diff options
author | Dongwon Kim <dongwon.kim@intel.com> | 2021-08-23 20:27:50 -0700 |
---|---|---|
committer | Anuj Mittal <anuj.mittal@intel.com> | 2021-08-24 16:05:35 +0800 |
commit | 6f8e742aab2526a35aafc671d2197bf4604bae26 (patch) | |
tree | 2e7f8f62338178072114f05546ec595eab0bf0f2 /dynamic-layers/clang-layer/recipes-opencl/igc | |
parent | 6a628fa0b507eefd3ed1608b39f4702ea56d1093 (diff) | |
download | meta-intel-6f8e742aab2526a35aafc671d2197bf4604bae26.tar.gz |
intel-graphics-compiler: update IGC to 1.0.8365
Updating IGC to add ADL-P support.
All patches under files folder are rebased as well.
Signed-off-by: Dongwon Kim <dongwon.kim@intel.com>
Signed-off-by: Anuj Mittal <anuj.mittal@intel.com>
Diffstat (limited to 'dynamic-layers/clang-layer/recipes-opencl/igc')
9 files changed, 617 insertions, 1091 deletions
diff --git a/dynamic-layers/clang-layer/recipes-opencl/igc/files/0001-skip-execution-of-ElfPackager.patch b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0001-skip-execution-of-ElfPackager.patch index 65342d9c..b77823b8 100644 --- a/dynamic-layers/clang-layer/recipes-opencl/igc/files/0001-skip-execution-of-ElfPackager.patch +++ b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0001-skip-execution-of-ElfPackager.patch | |||
@@ -1,7 +1,7 @@ | |||
1 | From 5501fab763aa2b11437fbdb19e07a11157a7d3e6 Mon Sep 17 00:00:00 2001 | 1 | From 562a19fe7ef3f4732365efdfc6d2a30763a5b8b0 Mon Sep 17 00:00:00 2001 |
2 | From: Naveen Saini <naveen.kumar.saini@intel.com> | 2 | From: Dongwon Kim <dongwon.kim@intel.com> |
3 | Date: Thu, 6 Feb 2020 14:56:56 +0800 | 3 | Date: Wed, 18 Aug 2021 18:49:49 -0700 |
4 | Subject: [PATCH] skip execution of ElfPackager | 4 | Subject: [PATCH 1/5] skip execution of ElfPackager |
5 | 5 | ||
6 | ElfPackager adds the ability to convert llvm bitcode into elf files for | 6 | ElfPackager adds the ability to convert llvm bitcode into elf files for |
7 | easier partitioning. Skip for now until we start building a native | 7 | easier partitioning. Skip for now until we start building a native |
@@ -11,33 +11,34 @@ Upstream-Status: Inappropriate [configuration specific] | |||
11 | 11 | ||
12 | Signed-off-by: Anuj Mittal <anuj.mittal@intel.com> | 12 | Signed-off-by: Anuj Mittal <anuj.mittal@intel.com> |
13 | Signed-off-by: Naveen Saini <naveen.kumar.saini@intel.com> | 13 | Signed-off-by: Naveen Saini <naveen.kumar.saini@intel.com> |
14 | Signed-off-by: Dongwon Kim <dongwon.kim@intel.com> | ||
14 | --- | 15 | --- |
15 | IGC/ElfPackager/CMakeLists.txt | 12 ++++++------ | 16 | IGC/ElfPackager/CMakeLists.txt | 12 ++++++------ |
16 | 1 file changed, 6 insertions(+), 6 deletions(-) | 17 | 1 file changed, 6 insertions(+), 6 deletions(-) |
17 | 18 | ||
18 | diff --git a/IGC/ElfPackager/CMakeLists.txt b/IGC/ElfPackager/CMakeLists.txt | 19 | diff --git a/IGC/ElfPackager/CMakeLists.txt b/IGC/ElfPackager/CMakeLists.txt |
19 | index 563ac9d6..c6c74ce9 100644 | 20 | index 8959a3ebb..7c3a0b42f 100644 |
20 | --- a/IGC/ElfPackager/CMakeLists.txt | 21 | --- a/IGC/ElfPackager/CMakeLists.txt |
21 | +++ b/IGC/ElfPackager/CMakeLists.txt | 22 | +++ b/IGC/ElfPackager/CMakeLists.txt |
22 | @@ -71,12 +71,12 @@ include_directories ( | 23 | @@ -51,12 +51,12 @@ if(NOT TARGET ${IGC_BUILD__PROJ__ElfPackager}) |
23 | ${IGC_SOURCE_DIR}/AdaptorOCL/CLElfLib/ | 24 | ${IGC_SOURCE_DIR}/AdaptorOCL/CLElfLib/ |
24 | ) | 25 | ) |
25 | 26 | ||
26 | -if(NOT ANDROID) | 27 | - if(NOT ANDROID) |
27 | - add_custom_command(TARGET ${IGC_BUILD__PROJ__ElfPackager} | 28 | - add_custom_command(TARGET ${IGC_BUILD__PROJ__ElfPackager} |
28 | - POST_BUILD | 29 | - POST_BUILD |
29 | - COMMAND $<TARGET_FILE:${IGC_BUILD__PROJ__ElfPackager}> -includeSizet -funcList ${CMAKE_CURRENT_SOURCE_DIR}/function_bin.txt ${IGC_BUILD__BIF_DIR}/OCLBiFImpl.bc ${IGC_BUILD__BIF_DIR}/igdclbif.bin | 30 | - COMMAND $<TARGET_FILE:${IGC_BUILD__PROJ__ElfPackager}> -includeSizet -funcList ${CMAKE_CURRENT_SOURCE_DIR}/function_bin.txt ${IGC_BUILD__BIF_DIR}/OCLBiFImpl.bc ${IGC_BUILD__BIF_DIR}/igdclbif.bin |
30 | - ) | 31 | - ) |
31 | -endif() | 32 | - endif() |
32 | +#if(NOT ANDROID) | 33 | +# if(NOT ANDROID) |
33 | +# add_custom_command(TARGET ${IGC_BUILD__PROJ__ElfPackager} | 34 | +# add_custom_command(TARGET ${IGC_BUILD__PROJ__ElfPackager} |
34 | +# POST_BUILD | 35 | +# POST_BUILD |
35 | +# COMMAND $<TARGET_FILE:${IGC_BUILD__PROJ__ElfPackager}> -includeSizet -funcList ${CMAKE_CURRENT_SOURCE_DIR}/function_bin.txt ${IGC_BUILD__BIF_DIR}/OCLBiFImpl.bc ${IGC_BUILD__BIF_DIR}/igdclbif.bin | 36 | +# COMMAND $<TARGET_FILE:${IGC_BUILD__PROJ__ElfPackager}> -includeSizet -funcList ${CMAKE_CURRENT_SOURCE_DIR}/function_bin.txt ${IGC_BUILD__BIF_DIR}/OCLBiFImpl.bc ${IGC_BUILD__BIF_DIR}/igdclbif.bin |
36 | +# ) | 37 | +# ) |
37 | +#endif() | 38 | +# endif() |
38 | 39 | ||
39 | 40 | ||
40 | add_dependencies("${IGC_BUILD__PROJ__ElfPackager}" "${IGC_BUILD__PROJ__BiFModule_OCL}") | 41 | add_dependencies("${IGC_BUILD__PROJ__ElfPackager}" "${IGC_BUILD__PROJ__BiFModule_OCL}") |
41 | -- | 42 | -- |
42 | 2.17.1 | 43 | 2.20.1 |
43 | 44 | ||
diff --git a/dynamic-layers/clang-layer/recipes-opencl/igc/files/0002-IGC-VectorCompiler-CMakeLists.txt-link-to-external-L.patch b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0002-IGC-VectorCompiler-CMakeLists.txt-link-to-external-L.patch new file mode 100644 index 00000000..db834c99 --- /dev/null +++ b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0002-IGC-VectorCompiler-CMakeLists.txt-link-to-external-L.patch | |||
@@ -0,0 +1,47 @@ | |||
1 | From 92a663dd55917fdfad5ae4daa173a9d18df0ee9e Mon Sep 17 00:00:00 2001 | ||
2 | From: Dongwon Kim <dongwon.kim@intel.com> | ||
3 | Date: Wed, 18 Aug 2021 19:14:28 -0700 | ||
4 | Subject: [PATCH 2/5] IGC/VectorCompiler/CMakeLists.txt: link to external | ||
5 | LLVMGenXIntrinsics | ||
6 | |||
7 | By default LLVMGenXIntrinsics is to be build In-tree, but we want to | ||
8 | link externally. | ||
9 | |||
10 | Upstream-Status: Inappropriate [configuration specific] | ||
11 | |||
12 | Signed-off-by: Naveen Saini <naveen.kumar.saini@intel.com> | ||
13 | Signed-off-by: Dongwon Kim <dongwon.kim@intel.com> | ||
14 | --- | ||
15 | IGC/VectorCompiler/CMakeLists.txt | 16 ---------------- | ||
16 | 1 file changed, 16 deletions(-) | ||
17 | |||
18 | diff --git a/IGC/VectorCompiler/CMakeLists.txt b/IGC/VectorCompiler/CMakeLists.txt | ||
19 | index 0b26072f0..bf1cd939d 100644 | ||
20 | --- a/IGC/VectorCompiler/CMakeLists.txt | ||
21 | +++ b/IGC/VectorCompiler/CMakeLists.txt | ||
22 | @@ -43,22 +43,6 @@ if(LLVM_ON_WIN32) | ||
23 | endif() | ||
24 | |||
25 | # --- VC Intrinsics --- | ||
26 | - | ||
27 | -if(DEFINED VC_INTRINSICS_SRC) | ||
28 | - set(INTRSRC "${VC_INTRINSICS_SRC}/GenXIntrinsics") | ||
29 | -endif() | ||
30 | - | ||
31 | -if(NOT DEFINED INTRSRC) | ||
32 | - set(INTRSRC "${CMAKE_CURRENT_SOURCE_DIR}/../../../vc-intrinsics/GenXIntrinsics") | ||
33 | -endif() | ||
34 | - | ||
35 | -message(STATUS "[VC] Using vc-intrinsics source from: ${INTRSRC}") | ||
36 | -# Trick intrinsics. | ||
37 | -set(BUILD_EXTERNAL YES) | ||
38 | -# We are using prebuilt SPIRV and building intrinsics. | ||
39 | -set(INTRBUILD "${CMAKE_CURRENT_BINARY_DIR}/intrbuild") | ||
40 | -add_subdirectory(${INTRSRC} ${INTRBUILD} EXCLUDE_FROM_ALL) | ||
41 | - | ||
42 | include(cmake/spirv.cmake) | ||
43 | |||
44 | set(IGC_OPTION__VC_DISABLE_BIF_DEFAULT OFF) | ||
45 | -- | ||
46 | 2.20.1 | ||
47 | |||
diff --git a/dynamic-layers/clang-layer/recipes-opencl/igc/files/0002-Review-fixes-for-LLVM-12-phase-1.patch b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0002-Review-fixes-for-LLVM-12-phase-1.patch deleted file mode 100644 index 6580df75..00000000 --- a/dynamic-layers/clang-layer/recipes-opencl/igc/files/0002-Review-fixes-for-LLVM-12-phase-1.patch +++ /dev/null | |||
@@ -1,318 +0,0 @@ | |||
1 | From 60136b453bb3a109bfc88c4040b871af9d522ed5 Mon Sep 17 00:00:00 2001 | ||
2 | From: =?UTF-8?q?Zolt=C3=A1n=20B=C3=B6sz=C3=B6rm=C3=A9nyi?= <zboszor@pr.hu> | ||
3 | Date: Thu, 25 Feb 2021 19:40:21 +0100 | ||
4 | Subject: [PATCH 2/3] Review fixes for LLVM 12 phase 1 | ||
5 | MIME-Version: 1.0 | ||
6 | Content-Type: text/plain; charset=UTF-8 | ||
7 | Content-Transfer-Encoding: 8bit | ||
8 | |||
9 | Upstream-Status: Pending | ||
10 | |||
11 | Signed-off-by: Zoltán Böszörményi <zboszor@gmail.com> | ||
12 | Signed-off-by: Naveen Saini <naveen.kumar.saini@intel.com> | ||
13 | --- | ||
14 | IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | 6 +----- | ||
15 | IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp | 6 +----- | ||
16 | IGC/Compiler/CISACodeGen/AdvMemOpt.cpp | 7 ++----- | ||
17 | IGC/Compiler/CISACodeGen/Simd32Profitability.cpp | 7 ++----- | ||
18 | IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp | 11 +---------- | ||
19 | IGC/Compiler/GenTTI.cpp | 9 ++------- | ||
20 | IGC/Compiler/Legalizer/InstPromoter.cpp | 8 +------- | ||
21 | .../DeviceEnqueueFuncs/TransformBlocks.cpp | 6 +----- | ||
22 | .../OpenCLPasses/WIFuncs/WIFuncResolution.cpp | 4 ---- | ||
23 | IGC/Compiler/Optimizer/Scalarizer.cpp | 2 +- | ||
24 | IGC/WrapperLLVM/include/llvmWrapper/IR/DerivedTypes.h | 10 ++++++++++ | ||
25 | .../include/llvmWrapper/Support/TypeSize.h | 4 +--- | ||
26 | .../include/llvmWrapper/Transforms/Utils/LoopUtils.h | 8 ++++++++ | ||
27 | IGC/common/igc_resourceDimTypes.h | 5 ++--- | ||
28 | 14 files changed, 33 insertions(+), 60 deletions(-) | ||
29 | |||
30 | diff --git a/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp b/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | ||
31 | index 725a1512..12f42be8 100644 | ||
32 | --- a/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | ||
33 | +++ b/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | ||
34 | @@ -1929,11 +1929,7 @@ SPIRVToLLVM::transType(SPIRVType *T) { | ||
35 | auto name = isSubgroupAvcINTELTypeOpCode(OC) ? | ||
36 | OCLSubgroupINTELTypeOpCodeMap::rmap(OC) : | ||
37 | BuiltinOpaqueGenericTypeOpCodeMap::rmap(OC); | ||
38 | -#if LLVM_VERSION_MAJOR >= 12 | ||
39 | - auto *pST = llvm::StructType::getTypeByName(M->getContext(), name); | ||
40 | -#else | ||
41 | - auto *pST = M->getTypeByName(name); | ||
42 | -#endif | ||
43 | + auto *pST = IGCLLVM::getTypeByName(M, name); | ||
44 | pST = pST ? pST : StructType::create(*Context, name); | ||
45 | |||
46 | return mapType(T, PointerType::get(pST, getOCLOpaqueTypeAddrSpace(OC))); | ||
47 | diff --git a/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp b/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp | ||
48 | index 91b4623c..f4f0dee5 100644 | ||
49 | --- a/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp | ||
50 | +++ b/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp | ||
51 | @@ -93,11 +93,7 @@ saveLLVMModule(Module *M, const std::string &OutputFile) { | ||
52 | PointerType* | ||
53 | getOrCreateOpaquePtrType(Module *M, const std::string &Name, | ||
54 | unsigned AddrSpace) { | ||
55 | -#if LLVM_VERSION_MAJOR >= 12 | ||
56 | - auto OpaqueType = llvm::StructType::getTypeByName(M->getContext(), Name); | ||
57 | -#else | ||
58 | - auto OpaqueType = M->getTypeByName(Name); | ||
59 | -#endif | ||
60 | + auto OpaqueType = IGCLLVM::getTypeByName(M, Name); | ||
61 | if (!OpaqueType) | ||
62 | OpaqueType = StructType::create(M->getContext(), Name); | ||
63 | return PointerType::get(OpaqueType, AddrSpace); | ||
64 | diff --git a/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp b/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp | ||
65 | index fc45a510..a612a473 100644 | ||
66 | --- a/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp | ||
67 | +++ b/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp | ||
68 | @@ -33,6 +33,7 @@ IN THE SOFTWARE. | ||
69 | #include <llvm/Support/Debug.h> | ||
70 | #include <llvm/Support/raw_ostream.h> | ||
71 | #include <llvm/Transforms/Utils/Local.h> | ||
72 | +#include "llvmWrapper/Transforms/Utils/LoopUtils.h" | ||
73 | #include "common/LLVMWarningsPop.hpp" | ||
74 | #include "GenISAIntrinsics/GenIntrinsics.h" | ||
75 | #include "Compiler/CISACodeGen/ShaderCodeGen.hpp" | ||
76 | @@ -134,11 +135,7 @@ bool AdvMemOpt::runOnFunction(Function& F) { | ||
77 | for (auto I = LI->begin(), E = LI->end(); I != E; ++I) | ||
78 | for (auto DFI = df_begin(*I), DFE = df_end(*I); DFI != DFE; ++DFI) { | ||
79 | Loop* L = *DFI; | ||
80 | -#if LLVM_VERSION_MAJOR >= 12 | ||
81 | - if (L->isInnermost()) | ||
82 | -#else | ||
83 | - if (L->empty()) | ||
84 | -#endif | ||
85 | + if (IGCLLVM::isInnermost(L)) | ||
86 | InnermostLoops.push_back(L); | ||
87 | } | ||
88 | |||
89 | diff --git a/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp b/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp | ||
90 | index c1f4a419..5393d4e8 100644 | ||
91 | --- a/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp | ||
92 | +++ b/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp | ||
93 | @@ -28,6 +28,7 @@ IN THE SOFTWARE. | ||
94 | #include "Compiler/CISACodeGen/Platform.hpp" | ||
95 | #include "common/LLVMWarningsPush.hpp" | ||
96 | #include <llvmWrapper/IR/DerivedTypes.h> | ||
97 | +#include <llvmWrapper/Transforms/Utils/LoopUtils.h> | ||
98 | #include <llvm/IR/InstIterator.h> | ||
99 | #include <llvm/IR/Operator.h> | ||
100 | #include <llvmWrapper/IR/DerivedTypes.h> | ||
101 | @@ -995,11 +996,7 @@ static bool hasLongStridedLdStInLoop(Function* F, LoopInfo* LI, WIAnalysis* WI) | ||
102 | // Collect innermost simple loop. | ||
103 | for (auto I = LI->begin(), E = LI->end(); I != E; ++I) { | ||
104 | auto L = *I; | ||
105 | -#if LLVM_VERSION_MAJOR >= 12 | ||
106 | - if (!L->isInnermost()) | ||
107 | -#else | ||
108 | - if (!L->empty()) | ||
109 | -#endif | ||
110 | + if (!IGCLLVM::isInnermost(L)) | ||
111 | continue; | ||
112 | if (L->getNumBlocks() != 2) | ||
113 | continue; | ||
114 | diff --git a/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp b/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp | ||
115 | index adf992cb..33473c23 100644 | ||
116 | --- a/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp | ||
117 | +++ b/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp | ||
118 | @@ -153,21 +153,12 @@ void ConvertMSAAPayloadTo16Bit::visitCallInst(CallInst& I) | ||
119 | // In OGL there are uses of ldmcs other then ldms, using vec4float type. | ||
120 | // Fix them to use newly created 16bit ldmcs. | ||
121 | if (ldmcs->getType()->isVectorTy() && | ||
122 | -#if LLVM_VERSION_MAJOR >= 12 | ||
123 | - ldmcs->getType()->getScalarType()->isFloatTy()) | ||
124 | -#else | ||
125 | - ldmcs->getType()->getVectorElementType()->isFloatTy()) | ||
126 | -#endif | ||
127 | + cast<IGCLLVM::FixedVectorType>(ldmcs->getType())->getElementType()->isFloatTy()) | ||
128 | { | ||
129 | m_builder->SetInsertPoint(ldmcs); | ||
130 | |||
131 | -#if LLVM_VERSION_MAJOR >= 12 | ||
132 | uint ldmcsNumOfElements = cast<IGCLLVM::FixedVectorType>(ldmcs->getType())->getNumElements(); | ||
133 | uint new_mcs_callNumOfElements = cast<IGCLLVM::FixedVectorType>(new_mcs_call->getType())->getNumElements(); | ||
134 | -#else | ||
135 | - uint ldmcsNumOfElements = ldmcs->getType()->getVectorNumElements(); | ||
136 | - uint new_mcs_callNumOfElements = new_mcs_call->getType()->getVectorNumElements(); | ||
137 | -#endif | ||
138 | |||
139 | // vec of 16bit ints to vec of 32bit ints | ||
140 | Type* new_mcs_callVecType = IGCLLVM::FixedVectorType::get(m_builder->getInt32Ty(), new_mcs_callNumOfElements); | ||
141 | diff --git a/IGC/Compiler/GenTTI.cpp b/IGC/Compiler/GenTTI.cpp | ||
142 | index 9e4d2f26..53e3ec9e 100644 | ||
143 | --- a/IGC/Compiler/GenTTI.cpp | ||
144 | +++ b/IGC/Compiler/GenTTI.cpp | ||
145 | @@ -37,6 +37,7 @@ IN THE SOFTWARE. | ||
146 | #include "llvm/Analysis/LoopInfo.h" | ||
147 | #include "llvm/Analysis/ScalarEvolution.h" | ||
148 | #include "llvm/Support/raw_ostream.h" | ||
149 | +#include "llvmWrapper/Transforms/Utils/LoopUtils.h" | ||
150 | #include "common/LLVMWarningsPop.hpp" | ||
151 | |||
152 | using namespace llvm; | ||
153 | @@ -216,13 +217,7 @@ namespace llvm { | ||
154 | |||
155 | // Skip non-simple loop. | ||
156 | if (L->getNumBlocks() != 1) { | ||
157 | - if (IGC_IS_FLAG_ENABLED(EnableAdvRuntimeUnroll) && | ||
158 | -#if LLVM_VERSION_MAJOR >= 12 | ||
159 | - L->isInnermost() | ||
160 | -#else | ||
161 | - L->empty() | ||
162 | -#endif | ||
163 | - ) { | ||
164 | + if (IGC_IS_FLAG_ENABLED(EnableAdvRuntimeUnroll) && IGCLLVM::isInnermost(L)) { | ||
165 | auto countNonPHI = [](BasicBlock* BB) { | ||
166 | unsigned Total = BB->size(); | ||
167 | unsigned PHIs = 0; | ||
168 | diff --git a/IGC/Compiler/Legalizer/InstPromoter.cpp b/IGC/Compiler/Legalizer/InstPromoter.cpp | ||
169 | index 8fadf89f..63cbccb5 100644 | ||
170 | --- a/IGC/Compiler/Legalizer/InstPromoter.cpp | ||
171 | +++ b/IGC/Compiler/Legalizer/InstPromoter.cpp | ||
172 | @@ -398,13 +398,7 @@ bool InstPromoter::visitBitCastInst(BitCastInst& I) { | ||
173 | IRB->CreateBitCast(Val, IGCLLVM::FixedVectorType::get(DestTy->getScalarType(), N)); | ||
174 | |||
175 | std::vector<Constant*> Vals; | ||
176 | - for (unsigned i = 0; | ||
177 | -#if LLVM_VERSION_MAJOR >= 12 | ||
178 | - i < cast<IGCLLVM::FixedVectorType>(DestTy)->getNumElements(); | ||
179 | -#else | ||
180 | - i < DestTy->getVectorNumElements(); | ||
181 | -#endif | ||
182 | - i++) | ||
183 | + for (unsigned i = 0; i < cast<IGCLLVM::FixedVectorType>(DestTy)->getNumElements(); i++) | ||
184 | Vals.push_back(IRB->getInt32(i)); | ||
185 | |||
186 | Value* Mask = ConstantVector::get(Vals); | ||
187 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp | ||
188 | index 119520ed..a3681b79 100644 | ||
189 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp | ||
190 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp | ||
191 | @@ -952,11 +952,7 @@ namespace //Anonymous | ||
192 | { | ||
193 | auto ndrangeStructName = "struct.ndrange_t"; | ||
194 | auto module = _deviceExecCall->getModule(); | ||
195 | -#if LLVM_VERSION_MAJOR >= 12 | ||
196 | - auto ndrangeTy = llvm::StructType::getTypeByName(module->getContext(), ndrangeStructName); | ||
197 | -#else | ||
198 | - auto ndrangeTy = module->getTypeByName(ndrangeStructName); | ||
199 | -#endif | ||
200 | + auto ndrangeTy = IGCLLVM::getTypeByName(module, ndrangeStructName); | ||
201 | if (ndrangeTy == nullptr) | ||
202 | { | ||
203 | //create struct type | ||
204 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp | ||
205 | index 535d6268..c23c661d 100644 | ||
206 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp | ||
207 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp | ||
208 | @@ -303,11 +303,7 @@ static Value* BuildLoadInst(CallInst& CI, unsigned int Offset, Type* DataType) | ||
209 | auto Size = ElemByteSize; | ||
210 | if (DataType->isVectorTy()) | ||
211 | { | ||
212 | -#if LLVM_VERSION_MAJOR >= 12 | ||
213 | Size *= cast<IGCLLVM::FixedVectorType>(DataType)->getNumElements(); | ||
214 | -#else | ||
215 | - Size *= DataType->getVectorNumElements(); | ||
216 | -#endif | ||
217 | } | ||
218 | unsigned int AlignedOffset = (Offset / ElemByteSize) * ElemByteSize; | ||
219 | unsigned int LoadByteSize = (Offset == AlignedOffset) ? Size : Size * 2; | ||
220 | diff --git a/IGC/Compiler/Optimizer/Scalarizer.cpp b/IGC/Compiler/Optimizer/Scalarizer.cpp | ||
221 | index a4e73a6d..38627553 100644 | ||
222 | --- a/IGC/Compiler/Optimizer/Scalarizer.cpp | ||
223 | +++ b/IGC/Compiler/Optimizer/Scalarizer.cpp | ||
224 | @@ -778,7 +778,7 @@ void ScalarizeFunction::scalarizeInstruction(ShuffleVectorInst* SI) | ||
225 | |||
226 | // Generate array for shuffled scalar values | ||
227 | SmallVector<Value*, MAX_INPUT_VECTOR_WIDTH>newVector; | ||
228 | - unsigned width = int_cast<unsigned>(dyn_cast<IGCLLVM::FixedVectorType>(SI->getType())->getNumElements()); | ||
229 | + unsigned width = int_cast<unsigned>(cast<IGCLLVM::FixedVectorType>(SI->getType())->getNumElements()); | ||
230 | |||
231 | // Generate undef value, which may be needed as some scalar elements | ||
232 | UndefValue* undef = UndefValue::get(inputType->getElementType()); | ||
233 | diff --git a/IGC/WrapperLLVM/include/llvmWrapper/IR/DerivedTypes.h b/IGC/WrapperLLVM/include/llvmWrapper/IR/DerivedTypes.h | ||
234 | index a3f5a0b8..6a5407bb 100644 | ||
235 | --- a/IGC/WrapperLLVM/include/llvmWrapper/IR/DerivedTypes.h | ||
236 | +++ b/IGC/WrapperLLVM/include/llvmWrapper/IR/DerivedTypes.h | ||
237 | @@ -29,6 +29,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. | ||
238 | |||
239 | #include "llvm/Config/llvm-config.h" | ||
240 | #include "llvm/IR/DerivedTypes.h" | ||
241 | +#include "llvm/IR/Module.h" | ||
242 | |||
243 | namespace IGCLLVM | ||
244 | { | ||
245 | @@ -62,6 +63,15 @@ namespace IGCLLVM | ||
246 | return false; | ||
247 | #endif | ||
248 | } | ||
249 | + | ||
250 | + inline llvm::StructType *getTypeByName(llvm::Module *M, llvm::StringRef Name) { | ||
251 | +#if LLVM_VERSION_MAJOR >= 12 | ||
252 | + return llvm::StructType::getTypeByName(M->getContext(), Name); | ||
253 | +#else | ||
254 | + return M->getTypeByName(Name); | ||
255 | +#endif | ||
256 | + } | ||
257 | + | ||
258 | } | ||
259 | |||
260 | #endif | ||
261 | diff --git a/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h b/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h | ||
262 | index 30e29720..7021820c 100644 | ||
263 | --- a/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h | ||
264 | +++ b/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h | ||
265 | @@ -39,12 +39,10 @@ inline unsigned getElementCount(unsigned EC) { return EC; } | ||
266 | inline ElementCount getElementCount(unsigned EC) { | ||
267 | return ElementCount(EC, false); | ||
268 | } | ||
269 | -#elif LLVM_VERSION_MAJOR == 12 | ||
270 | +#else | ||
271 | inline ElementCount getElementCount(unsigned EC) { | ||
272 | return ElementCount::get(EC, false); | ||
273 | } | ||
274 | -#else | ||
275 | -#error "unsupported LLVM version" | ||
276 | #endif | ||
277 | } // namespace IGCLLVM | ||
278 | |||
279 | diff --git a/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Utils/LoopUtils.h b/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Utils/LoopUtils.h | ||
280 | index db47b00b..bce9cfc1 100644 | ||
281 | --- a/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Utils/LoopUtils.h | ||
282 | +++ b/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Utils/LoopUtils.h | ||
283 | @@ -41,6 +41,14 @@ namespace IGCLLVM | ||
284 | return llvm::InsertPreheaderForLoop(L, DT, LI, nullptr, PreserveLCSSA); | ||
285 | } | ||
286 | #endif | ||
287 | + | ||
288 | + inline bool isInnermost(llvm::Loop *L) { | ||
289 | +#if LLVM_VERSION_MAJOR >= 12 | ||
290 | + return L->isInnermost(); | ||
291 | +#else | ||
292 | + return L->empty(); | ||
293 | +#endif | ||
294 | + } | ||
295 | } | ||
296 | |||
297 | #endif | ||
298 | diff --git a/IGC/common/igc_resourceDimTypes.h b/IGC/common/igc_resourceDimTypes.h | ||
299 | index d790330f..2d675969 100644 | ||
300 | --- a/IGC/common/igc_resourceDimTypes.h | ||
301 | +++ b/IGC/common/igc_resourceDimTypes.h | ||
302 | @@ -67,10 +67,9 @@ namespace IGC | ||
303 | resourceDimTypeId == DIM_3D_TYPE || resourceDimTypeId == DIM_CUBE_TYPE || resourceDimTypeId == DIM_CUBE_ARRAY_TYPE)); | ||
304 | |||
305 | #if LLVM_VERSION_MAJOR >= 12 | ||
306 | - llvm::LLVMContext& llvmCtx = module.getContext(); | ||
307 | - return llvm::StructType::getTypeByName(llvmCtx, ResourceDimensionTypeName[resourceDimTypeId]); | ||
308 | + return llvm::StructType::getTypeByName(module.getContext(), ResourceDimensionTypeName[resourceDimTypeId]); | ||
309 | #else | ||
310 | return module.getTypeByName(ResourceDimensionTypeName[resourceDimTypeId]); | ||
311 | #endif | ||
312 | } | ||
313 | -} | ||
314 | \ No newline at end of file | ||
315 | +} | ||
316 | -- | ||
317 | 2.17.1 | ||
318 | |||
diff --git a/dynamic-layers/clang-layer/recipes-opencl/igc/files/improve_src_package_reproducibility.patch b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0003-Improve-Reproducibility-for-src-package.patch index c694836f..7dd36aec 100644 --- a/dynamic-layers/clang-layer/recipes-opencl/igc/files/improve_src_package_reproducibility.patch +++ b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0003-Improve-Reproducibility-for-src-package.patch | |||
@@ -1,7 +1,7 @@ | |||
1 | From 2fa3f2da1179bd5b2eeac82839d386f0111e611c Mon Sep 17 00:00:00 2001 | 1 | From c2b7f30dd56568482b1b7c2f22bafdf68736fc88 Mon Sep 17 00:00:00 2001 |
2 | From: Lee Chee Yang <chee.yang.lee@intel.com> | 2 | From: Lee Chee Yang <chee.yang.lee@intel.com> |
3 | Date: Wed, 2 Sep 2020 08:28:35 +0800 | 3 | Date: Wed, 2 Sep 2020 08:28:35 +0800 |
4 | Subject: [PATCH] Improve Reproducibility for src package | 4 | Subject: [PATCH 3/5] Improve Reproducibility for src package |
5 | 5 | ||
6 | Improve reproducibility for intel-graphics-compiler-src package. | 6 | Improve reproducibility for intel-graphics-compiler-src package. |
7 | needs to pass build path as environment variable to the build. | 7 | needs to pass build path as environment variable to the build. |
@@ -14,10 +14,10 @@ Signed-off-by: Lee Chee Yang <chee.yang.lee@intel.com> | |||
14 | 1 file changed, 5 insertions(+), 2 deletions(-) | 14 | 1 file changed, 5 insertions(+), 2 deletions(-) |
15 | 15 | ||
16 | diff --git a/visa/CMakeLists.txt b/visa/CMakeLists.txt | 16 | diff --git a/visa/CMakeLists.txt b/visa/CMakeLists.txt |
17 | index 981c35d2..d5944370 100644 | 17 | index 65dbb4934..8cd607a69 100644 |
18 | --- a/visa/CMakeLists.txt | 18 | --- a/visa/CMakeLists.txt |
19 | +++ b/visa/CMakeLists.txt | 19 | +++ b/visa/CMakeLists.txt |
20 | @@ -109,8 +109,11 @@ endif() | 20 | @@ -123,8 +123,11 @@ endif() |
21 | set(bison_output_file ${CMAKE_CURRENT_BINARY_DIR}/CISA.tab.cpp) | 21 | set(bison_output_file ${CMAKE_CURRENT_BINARY_DIR}/CISA.tab.cpp) |
22 | set(flex_output_file ${CMAKE_CURRENT_BINARY_DIR}/lex.CISA.cpp) | 22 | set(flex_output_file ${CMAKE_CURRENT_BINARY_DIR}/lex.CISA.cpp) |
23 | 23 | ||
@@ -32,5 +32,5 @@ index 981c35d2..d5944370 100644 | |||
32 | set(CISAScanner_dependencies) | 32 | set(CISAScanner_dependencies) |
33 | 33 | ||
34 | -- | 34 | -- |
35 | 2.28.0 | 35 | 2.20.1 |
36 | 36 | ||
diff --git a/dynamic-layers/clang-layer/recipes-opencl/igc/files/0003-Review-fixes-for-LLVM-12-phase-2.patch b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0003-Review-fixes-for-LLVM-12-phase-2.patch deleted file mode 100644 index 189c2b3f..00000000 --- a/dynamic-layers/clang-layer/recipes-opencl/igc/files/0003-Review-fixes-for-LLVM-12-phase-2.patch +++ /dev/null | |||
@@ -1,123 +0,0 @@ | |||
1 | From c6d333637537263930acb1b6c5dadb0467d745f6 Mon Sep 17 00:00:00 2001 | ||
2 | From: =?UTF-8?q?Zolt=C3=A1n=20B=C3=B6sz=C3=B6rm=C3=A9nyi?= <zboszor@pr.hu> | ||
3 | Date: Fri, 26 Feb 2021 06:39:35 +0100 | ||
4 | Subject: [PATCH 3/3] Review fixes for LLVM 12 phase 2 | ||
5 | MIME-Version: 1.0 | ||
6 | Content-Type: text/plain; charset=UTF-8 | ||
7 | Content-Transfer-Encoding: 8bit | ||
8 | |||
9 | Upstream-Status: Pending | ||
10 | |||
11 | Signed-off-by: Zoltán Böszörményi <zboszor@gmail.com> | ||
12 | Signed-off-by: Naveen Saini <naveen.kumar.saini@intel.com> | ||
13 | --- | ||
14 | IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | 4 ---- | ||
15 | .../AddressSpaceAliasAnalysis.cpp | 10 +++++----- | ||
16 | .../PrivateMemory/PrivateMemoryResolution.cpp | 4 ---- | ||
17 | IGC/DebugInfo/DebugInfoUtils.hpp | 4 ---- | ||
18 | IGC/DebugInfo/DwarfDebug.cpp | 8 -------- | ||
19 | 5 files changed, 5 insertions(+), 25 deletions(-) | ||
20 | |||
21 | diff --git a/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp b/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | ||
22 | index 12f42be8..c4f9d1ea 100644 | ||
23 | --- a/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | ||
24 | +++ b/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | ||
25 | @@ -1576,11 +1576,7 @@ void SPIRVToLLVMDbgTran::transDbgInfo(SPIRVValue *SV, Value *V) { | ||
26 | Line->getColumn(), scope, iat); | ||
27 | |||
28 | if(scope && !isa<DIFile>(scope)) | ||
29 | -#if LLVM_VERSION_MAJOR >= 12 | ||
30 | I->setDebugLoc(DILocation::get(scope->getContext(), Line->getLine(), Line->getColumn(), | ||
31 | -#else | ||
32 | - I->setDebugLoc(DebugLoc::get(Line->getLine(), Line->getColumn(), | ||
33 | -#endif | ||
34 | scope, iat)); | ||
35 | } | ||
36 | } | ||
37 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp | ||
38 | index e9c07b34..b6b779da 100644 | ||
39 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp | ||
40 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp | ||
41 | @@ -23,8 +23,7 @@ IN THE SOFTWARE. | ||
42 | ============================= end_copyright_notice ===========================*/ | ||
43 | |||
44 | #include "llvm/Config/llvm-config.h" | ||
45 | -#include "llvmWrapper/IR/DerivedTypes.h" | ||
46 | -#include "llvmWrapper/Analysis/TargetLibraryInfo.h" | ||
47 | +#include <llvm/Analysis/TargetLibraryInfo.h> | ||
48 | #include "Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.h" | ||
49 | #include "Compiler/CodeGenPublic.h" | ||
50 | #include "Compiler/IGCPassSupport.h" | ||
51 | @@ -180,11 +179,12 @@ namespace { | ||
52 | bool doInitialization(Module& M) override { | ||
53 | if(M.size() > 0) | ||
54 | { | ||
55 | + Result.reset(new AddressSpaceAAResult( | ||
56 | + getAnalysis<TargetLibraryInfoWrapperPass>().getTLI( | ||
57 | #if LLVM_VERSION_MAJOR >= 10 | ||
58 | - Function &F = *M.begin(); | ||
59 | + *M.begin() | ||
60 | #endif | ||
61 | - Result.reset(new AddressSpaceAAResult( | ||
62 | - getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(), | ||
63 | + ), | ||
64 | *getAnalysis<CodeGenContextWrapper>().getCodeGenContext())); | ||
65 | } | ||
66 | return false; | ||
67 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp | ||
68 | index 07f85f4c..98ea616f 100644 | ||
69 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp | ||
70 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp | ||
71 | @@ -816,11 +816,7 @@ bool PrivateMemoryResolution::resolveAllocaInstructions(bool privateOnStack) | ||
72 | // Construct an empty DebugLoc. | ||
73 | IF_DEBUG_INFO(DebugLoc entryDebugLoc); | ||
74 | // Assign with the function location if available. | ||
75 | -#if LLVM_VERSION_MAJOR >= 12 | ||
76 | IF_DEBUG_INFO_IF(DISubprogram *subprogram = m_currFunction->getSubprogram(), entryDebugLoc = DILocation::get(subprogram->getContext(), subprogram->getLine(), 0, subprogram);); | ||
77 | -#else | ||
78 | - IF_DEBUG_INFO_IF(DISubprogram *subprogram = m_currFunction->getSubprogram(), entryDebugLoc = DebugLoc::get(subprogram->getLine(), 0, subprogram);); | ||
79 | -#endif | ||
80 | IF_DEBUG_INFO(entryBuilder.SetCurrentDebugLocation(entryDebugLoc)); | ||
81 | |||
82 | if (privateOnStack) | ||
83 | diff --git a/IGC/DebugInfo/DebugInfoUtils.hpp b/IGC/DebugInfo/DebugInfoUtils.hpp | ||
84 | index b77a550d..88b30a75 100644 | ||
85 | --- a/IGC/DebugInfo/DebugInfoUtils.hpp | ||
86 | +++ b/IGC/DebugInfo/DebugInfoUtils.hpp | ||
87 | @@ -108,11 +108,7 @@ namespace IGC | ||
88 | IGCLLVM::DIBuilder Builder(M); | ||
89 | llvm::DIGlobalVariable* GV = GVs[j]->getVariable(); | ||
90 | llvm::DIScope* scopeToUse = GV->getScope(); | ||
91 | -#if LLVM_VERSION_MAJOR >= 12 | ||
92 | llvm::DILocation* locToUse = llvm::DILocation::get(scopeToUse->getContext(), GV->getLine(), 0, scopeToUse, loc); | ||
93 | -#else | ||
94 | - llvm::DILocation* locToUse = llvm::DebugLoc::get(GV->getLine(), 0, scopeToUse, loc); | ||
95 | -#endif | ||
96 | if (llvm::isa<llvm::DICompileUnit>(GV->getScope())) | ||
97 | { | ||
98 | // Function has no DebugLoc so it is either internal | ||
99 | diff --git a/IGC/DebugInfo/DwarfDebug.cpp b/IGC/DebugInfo/DwarfDebug.cpp | ||
100 | index bd9f17b7..3d9f0835 100644 | ||
101 | --- a/IGC/DebugInfo/DwarfDebug.cpp | ||
102 | +++ b/IGC/DebugInfo/DwarfDebug.cpp | ||
103 | @@ -2102,17 +2102,9 @@ static DebugLoc getFnDebugLoc(DebugLoc DL, const LLVMContext& Ctx) | ||
104 | // Check for number of operands since the compatibility is cheap here. | ||
105 | if (SP->getNumOperands() > 19) | ||
106 | { | ||
107 | -#if LLVM_VERSION_MAJOR >= 12 | ||
108 | return DILocation::get(SP->getContext(), SP->getScopeLine(), 0, SP); | ||
109 | -#else | ||
110 | - return DebugLoc::get(SP->getScopeLine(), 0, SP); | ||
111 | -#endif | ||
112 | } | ||
113 | -#if LLVM_VERSION_MAJOR >= 12 | ||
114 | return DILocation::get(SP->getContext(), SP->getLine(), 0, SP); | ||
115 | -#else | ||
116 | - return DebugLoc::get(SP->getLine(), 0, SP); | ||
117 | -#endif | ||
118 | } | ||
119 | |||
120 | return DebugLoc(); | ||
121 | -- | ||
122 | 2.17.1 | ||
123 | |||
diff --git a/dynamic-layers/clang-layer/recipes-opencl/igc/files/0004-find-external-llvm-tblgen.patch b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0004-find-external-llvm-tblgen.patch new file mode 100644 index 00000000..79f79976 --- /dev/null +++ b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0004-find-external-llvm-tblgen.patch | |||
@@ -0,0 +1,30 @@ | |||
1 | From c9fe51ec555fadd098cfc98804ce91b1cf3029d4 Mon Sep 17 00:00:00 2001 | ||
2 | From: Dongwon Kim <dongwon.kim@intel.com> | ||
3 | Date: Thu, 19 Aug 2021 08:28:03 -0700 | ||
4 | Subject: [PATCH 4/5] find external llvm-tblgen | ||
5 | |||
6 | Upstream-Status: Pending | ||
7 | Signed-off-by: Dongwon Kim <dongwon.kim@intel.com> | ||
8 | --- | ||
9 | IGC/cmake/igc_llvm.cmake | 5 ++++- | ||
10 | 1 file changed, 4 insertions(+), 1 deletion(-) | ||
11 | |||
12 | diff --git a/IGC/cmake/igc_llvm.cmake b/IGC/cmake/igc_llvm.cmake | ||
13 | index 541793f21..bc82922b1 100644 | ||
14 | --- a/IGC/cmake/igc_llvm.cmake | ||
15 | +++ b/IGC/cmake/igc_llvm.cmake | ||
16 | @@ -24,7 +24,10 @@ set(CMAKE_MODULE_PATH | ||
17 | ${CMAKE_MODULE_PATH} | ||
18 | ) | ||
19 | |||
20 | -set(LLVM_TABLEGEN_EXE "llvm-tblgen") | ||
21 | +find_program(LLVM_TABLEGEN_EXE "llvm-tblgen") | ||
22 | +if(LLVM_TABLEGEN_EXE-NOTFOUND) | ||
23 | + message(FATAL_ERROR "[VC] llvm-tblgen is not found") | ||
24 | +endif() | ||
25 | |||
26 | include(AddLLVM) | ||
27 | include(TableGen) | ||
28 | -- | ||
29 | 2.20.1 | ||
30 | |||
diff --git a/dynamic-layers/clang-layer/recipes-opencl/igc/files/0001-Fix-build-with-LLVM-12.patch b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0005-Temporary-LLVM-12-compatiblity-fix.patch index 1d2c5767..000595a9 100644 --- a/dynamic-layers/clang-layer/recipes-opencl/igc/files/0001-Fix-build-with-LLVM-12.patch +++ b/dynamic-layers/clang-layer/recipes-opencl/igc/files/0005-Temporary-LLVM-12-compatiblity-fix.patch | |||
@@ -1,126 +1,123 @@ | |||
1 | From 869cdb784aa062bd08bd26465bf0b5651e8f112e Mon Sep 17 00:00:00 2001 | 1 | From 2207600aa9ddf8be1796bcc4667bc0a3b60aef03 Mon Sep 17 00:00:00 2001 |
2 | From: =?UTF-8?q?Zolt=C3=A1n=20B=C3=B6sz=C3=B6rm=C3=A9nyi?= <zboszor@pr.hu> | 2 | From: Dongwon Kim <dongwon.kim@intel.com> |
3 | Date: Thu, 25 Feb 2021 10:39:27 +0100 | 3 | Date: Thu, 19 Aug 2021 20:09:23 -0700 |
4 | Subject: [PATCH 1/3] Fix build with LLVM 12 | 4 | Subject: [PATCH 5/5] Temporary LLVM 12 compatiblity fix |
5 | MIME-Version: 1.0 | 5 | MIME-Version: 1.0 |
6 | Content-Type: text/plain; charset=UTF-8 | 6 | Content-Type: text/plain; charset=UTF-8 |
7 | Content-Transfer-Encoding: 8bit | 7 | Content-Transfer-Encoding: 8bit |
8 | 8 | ||
9 | Upstream-Status: Pending | 9 | Upstream-Status: Submitted |
10 | 10 | ||
11 | Signed-off-by: Zoltán Böszörményi <zboszor@gmail.com> | 11 | Signed-off-by: Zoltán Böszörményi <zboszor@gmail.com> |
12 | Signed-off-by: Naveen Saini <naveen.kumar.saini@intel.com> | 12 | Signed-off-by: Naveen Saini <naveen.kumar.saini@intel.com> |
13 | Signed-off-by: Dongwon Kim <dongwon.kim@intel.com> | ||
13 | --- | 14 | --- |
14 | .../LegalizeFunctionSignatures.cpp | 2 +- | 15 | .../LegalizeFunctionSignatures.cpp | 2 +- |
15 | IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | 26 +++++++++----- | 16 | IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | 22 +++++----- |
16 | IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp | 6 +++- | 17 | IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp | 4 +- |
17 | IGC/Compiler/CISACodeGen/AdvMemOpt.cpp | 4 +++ | 18 | IGC/Compiler/CISACodeGen/AdvMemOpt.cpp | 3 +- |
18 | IGC/Compiler/CISACodeGen/CShader.cpp | 14 ++++---- | 19 | IGC/Compiler/CISACodeGen/CShader.cpp | 14 +++---- |
19 | .../CISACodeGen/ConstantCoalescing.cpp | 14 ++++---- | 20 | .../CISACodeGen/ConstantCoalescing.cpp | 14 +++---- |
20 | IGC/Compiler/CISACodeGen/DeSSA.cpp | 3 +- | 21 | IGC/Compiler/CISACodeGen/DeSSA.cpp | 5 ++- |
21 | IGC/Compiler/CISACodeGen/EmitVISAPass.cpp | 35 +++++++++--------- | 22 | IGC/Compiler/CISACodeGen/EmitVISAPass.cpp | 42 +++++++++---------- |
22 | IGC/Compiler/CISACodeGen/Emu64OpsPass.cpp | 4 +-- | 23 | IGC/Compiler/CISACodeGen/Emu64OpsPass.cpp | 4 +- |
23 | IGC/Compiler/CISACodeGen/GenIRLowering.cpp | 4 +-- | 24 | IGC/Compiler/CISACodeGen/GenIRLowering.cpp | 4 +- |
24 | .../CISACodeGen/GenSimplification.cpp | 5 +-- | 25 | .../CISACodeGen/GenSimplification.cpp | 5 ++- |
25 | .../CISACodeGen/GeometryShaderLowering.cpp | 3 +- | 26 | .../CISACodeGen/GeometryShaderLowering.cpp | 3 +- |
26 | IGC/Compiler/CISACodeGen/LdShrink.cpp | 10 ++++-- | 27 | IGC/Compiler/CISACodeGen/LdShrink.cpp | 10 +++-- |
27 | .../CISACodeGen/LowerGEPForPrivMem.cpp | 8 ++--- | 28 | .../CISACodeGen/LowerGEPForPrivMem.cpp | 8 ++-- |
28 | IGC/Compiler/CISACodeGen/MemOpt.cpp | 8 ++--- | 29 | IGC/Compiler/CISACodeGen/MemOpt.cpp | 8 ++-- |
29 | IGC/Compiler/CISACodeGen/MemOpt2.cpp | 5 +-- | 30 | IGC/Compiler/CISACodeGen/MemOpt2.cpp | 5 ++- |
30 | .../CISACodeGen/OpenCLKernelCodeGen.cpp | 2 +- | 31 | .../CISACodeGen/OpenCLKernelCodeGen.cpp | 2 +- |
31 | IGC/Compiler/CISACodeGen/PatternMatchPass.cpp | 4 +-- | 32 | IGC/Compiler/CISACodeGen/PatternMatchPass.cpp | 6 +-- |
32 | IGC/Compiler/CISACodeGen/PushAnalysis.cpp | 4 +-- | 33 | IGC/Compiler/CISACodeGen/PushAnalysis.cpp | 4 +- |
33 | .../CISACodeGen/RegisterEstimator.cpp | 3 +- | 34 | .../CISACodeGen/RegisterEstimator.cpp | 3 +- |
34 | .../CISACodeGen/RegisterPressureEstimate.hpp | 3 +- | 35 | .../CISACodeGen/RegisterPressureEstimate.hpp | 3 +- |
35 | IGC/Compiler/CISACodeGen/SLMConstProp.cpp | 3 +- | 36 | IGC/Compiler/CISACodeGen/SLMConstProp.cpp | 3 +- |
36 | .../CISACodeGen/ScalarizerCodeGen.cpp | 4 +-- | 37 | .../CISACodeGen/ScalarizerCodeGen.cpp | 4 +- |
37 | IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp | 4 +++ | 38 | IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp | 4 ++ |
38 | .../CISACodeGen/Simd32Profitability.cpp | 11 ++++-- | 39 | .../CISACodeGen/Simd32Profitability.cpp | 10 +++-- |
39 | IGC/Compiler/CISACodeGen/SimplifyConstant.cpp | 12 +++---- | 40 | IGC/Compiler/CISACodeGen/SimplifyConstant.cpp | 12 +++--- |
40 | IGC/Compiler/CISACodeGen/TypeDemote.cpp | 3 +- | 41 | IGC/Compiler/CISACodeGen/TypeDemote.cpp | 3 +- |
41 | .../CISACodeGen/VariableReuseAnalysis.cpp | 5 +-- | 42 | .../CISACodeGen/VariableReuseAnalysis.cpp | 5 ++- |
42 | .../CISACodeGen/VariableReuseAnalysis.hpp | 3 +- | 43 | .../CISACodeGen/VariableReuseAnalysis.hpp | 3 +- |
43 | IGC/Compiler/CISACodeGen/VectorPreProcess.cpp | 24 ++++++------- | 44 | IGC/Compiler/CISACodeGen/VectorPreProcess.cpp | 24 +++++------ |
44 | IGC/Compiler/CISACodeGen/VectorProcess.cpp | 4 +-- | 45 | IGC/Compiler/CISACodeGen/VectorProcess.cpp | 4 +- |
45 | IGC/Compiler/CISACodeGen/helper.cpp | 10 +++--- | 46 | IGC/Compiler/CISACodeGen/helper.cpp | 10 ++--- |
46 | IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp | 15 ++++++-- | 47 | IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp | 7 ++-- |
47 | IGC/Compiler/CustomSafeOptPass.cpp | 26 +++++++------- | 48 | IGC/Compiler/CustomSafeOptPass.cpp | 26 ++++++------ |
48 | IGC/Compiler/DebugInfo/ScalarVISAModule.cpp | 3 +- | 49 | IGC/Compiler/DebugInfo/ScalarVISAModule.cpp | 3 +- |
49 | IGC/Compiler/GenTTI.cpp | 8 ++++- | 50 | IGC/Compiler/GenTTI.cpp | 3 +- |
50 | IGC/Compiler/GenUpdateCB.cpp | 3 +- | 51 | IGC/Compiler/GenUpdateCB.cpp | 3 +- |
51 | IGC/Compiler/HandleFRemInstructions.cpp | 3 +- | 52 | IGC/Compiler/HandleFRemInstructions.cpp | 3 +- |
52 | IGC/Compiler/HandleLoadStoreInstructions.cpp | 4 +-- | 53 | IGC/Compiler/HandleLoadStoreInstructions.cpp | 4 +- |
53 | IGC/Compiler/LegalizationPass.cpp | 28 +++++++-------- | 54 | IGC/Compiler/LegalizationPass.cpp | 28 ++++++------- |
54 | IGC/Compiler/Legalizer/InstPromoter.cpp | 11 ++++-- | 55 | IGC/Compiler/Legalizer/InstPromoter.cpp | 2 +- |
55 | IGC/Compiler/Legalizer/InstScalarizer.cpp | 6 ++-- | 56 | IGC/Compiler/Legalizer/InstScalarizer.cpp | 6 +-- |
56 | .../Legalizer/PeepholeTypeLegalizer.cpp | 6 ++-- | 57 | .../Legalizer/PeepholeTypeLegalizer.cpp | 6 +-- |
57 | IGC/Compiler/Legalizer/TypeLegalizer.cpp | 3 +- | 58 | IGC/Compiler/Legalizer/TypeLegalizer.cpp | 3 +- |
58 | IGC/Compiler/Legalizer/TypeLegalizer.h | 3 +- | 59 | IGC/Compiler/Legalizer/TypeLegalizer.h | 3 +- |
59 | IGC/Compiler/LowPrecisionOptPass.cpp | 2 +- | 60 | IGC/Compiler/LowPrecisionOptPass.cpp | 2 +- |
60 | .../AddressSpaceAliasAnalysis.cpp | 9 ++--- | 61 | .../AddressSpaceAliasAnalysis.cpp | 1 + |
61 | .../AggregateArguments/AggregateArguments.cpp | 3 +- | 62 | .../AggregateArguments/AggregateArguments.cpp | 3 +- |
62 | .../CorrectlyRoundedDivSqrt.cpp | 3 +- | 63 | .../CorrectlyRoundedDivSqrt.cpp | 3 +- |
63 | .../DeviceEnqueueFuncs/TransformBlocks.cpp | 6 +++- | 64 | .../DeviceEnqueueFuncs/TransformBlocks.cpp | 4 +- |
64 | .../Optimizer/OpenCLPasses/KernelArgs.cpp | 3 +- | 65 | .../Optimizer/OpenCLPasses/KernelArgs.cpp | 3 +- |
65 | .../OpenCLPrintf/OpenCLPrintfResolution.cpp | 4 +-- | 66 | .../OpenCLPrintf/OpenCLPrintfResolution.cpp | 4 +- |
66 | .../PrivateMemory/PrivateMemoryResolution.cpp | 11 ++++-- | 67 | .../PrivateMemory/PrivateMemoryResolution.cpp | 6 +-- |
67 | .../ReplaceUnsupportedIntrinsics.cpp | 4 +-- | 68 | .../ReplaceUnsupportedIntrinsics.cpp | 4 +- |
68 | .../OpenCLPasses/WIFuncs/WIFuncResolution.cpp | 4 +++ | 69 | .../OpenCLPasses/WIFuncs/WIFuncResolution.cpp | 2 +- |
69 | .../Optimizer/PreCompiledFuncImport.cpp | 4 +-- | 70 | .../Optimizer/PreCompiledFuncImport.cpp | 4 +- |
70 | IGC/Compiler/Optimizer/Scalarizer.cpp | 36 +++++++++---------- | 71 | IGC/Compiler/Optimizer/Scalarizer.cpp | 32 +++++++------- |
71 | IGC/Compiler/Optimizer/ValueTracker.cpp | 2 +- | 72 | IGC/Compiler/Optimizer/ValueTracker.cpp | 2 +- |
72 | IGC/Compiler/VerificationPass.cpp | 2 +- | 73 | IGC/Compiler/VerificationPass.cpp | 2 +- |
73 | IGC/DebugInfo/DebugInfoUtils.hpp | 4 +++ | 74 | IGC/DebugInfo/DebugInfoUtils.hpp | 2 +- |
74 | IGC/DebugInfo/DwarfDebug.cpp | 8 +++++ | 75 | IGC/DebugInfo/DwarfDebug.cpp | 4 +- |
75 | IGC/GenISAIntrinsics/GenIntrinsics.cpp | 6 ++-- | 76 | IGC/GenISAIntrinsics/GenIntrinsics.cpp | 6 +-- |
76 | .../BuiltinsFrontendDefinitions.hpp | 2 +- | 77 | .../BuiltinsFrontendDefinitions.hpp | 2 +- |
77 | .../include/llvmWrapper/Support/TypeSize.h | 8 ++++- | 78 | .../include/llvmWrapper/IR/DerivedTypes.h | 10 +++++ |
78 | .../include/llvmWrapper/Transforms/Scalar.h | 4 +-- | 79 | .../include/llvmWrapper/Support/TypeSize.h | 6 ++- |
79 | IGC/common/igc_resourceDimTypes.h | 5 +++ | 80 | .../include/llvmWrapper/Transforms/Scalar.h | 4 +- |
80 | 66 files changed, 303 insertions(+), 196 deletions(-) | 81 | .../llvmWrapper/Transforms/Utils/LoopUtils.h | 8 ++++ |
82 | IGC/common/igc_resourceDimTypes.h | 6 ++- | ||
83 | 68 files changed, 263 insertions(+), 207 deletions(-) | ||
81 | 84 | ||
82 | diff --git a/IGC/AdaptorCommon/LegalizeFunctionSignatures.cpp b/IGC/AdaptorCommon/LegalizeFunctionSignatures.cpp | 85 | diff --git a/IGC/AdaptorCommon/LegalizeFunctionSignatures.cpp b/IGC/AdaptorCommon/LegalizeFunctionSignatures.cpp |
83 | index 4046b4a6..8ed36089 100644 | 86 | index d06c26de4..ee43be301 100644 |
84 | --- a/IGC/AdaptorCommon/LegalizeFunctionSignatures.cpp | 87 | --- a/IGC/AdaptorCommon/LegalizeFunctionSignatures.cpp |
85 | +++ b/IGC/AdaptorCommon/LegalizeFunctionSignatures.cpp | 88 | +++ b/IGC/AdaptorCommon/LegalizeFunctionSignatures.cpp |
86 | @@ -156,7 +156,7 @@ inline Type* LegalizedIntVectorType(Module& M, const Type* const oldTy) | 89 | @@ -133,7 +133,7 @@ inline Type* LegalizedIntVectorType(Module& M, Type* ty) |
87 | else if (size <= 64) newSize = 64; | ||
88 | else IGC_ASSERT_MESSAGE(0, "Currently don't support upscaling int sizes > 64 bits"); | ||
89 | 90 | ||
90 | - return IGCLLVM::FixedVectorType::get(IntegerType::get(M.getContext(), newSize), (unsigned)cast<VectorType>(oldTy)->getNumElements()); | 91 | return ty->isIntegerTy() ? |
91 | + return IGCLLVM::FixedVectorType::get(IntegerType::get(M.getContext(), newSize), (unsigned)cast<IGCLLVM::FixedVectorType>(oldTy)->getNumElements()); | 92 | cast<Type>(IntegerType::get(M.getContext(), newSize)) : |
93 | - IGCLLVM::FixedVectorType::get(IntegerType::get(M.getContext(), newSize), (unsigned)cast<VectorType>(ty)->getNumElements()); | ||
94 | + IGCLLVM::FixedVectorType::get(IntegerType::get(M.getContext(), newSize), (unsigned)cast<IGCLLVM::FixedVectorType>(ty)->getNumElements()); | ||
92 | } | 95 | } |
93 | 96 | ||
94 | void LegalizeFunctionSignatures::FixFunctionSignatures() | 97 | // Returns true for small structures that only contain primitive types |
95 | diff --git a/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp b/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | 98 | diff --git a/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp b/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp |
96 | index 41d1f2ea..725a1512 100644 | 99 | index 66f31a0ce..65e672afe 100644 |
97 | --- a/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | 100 | --- a/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp |
98 | +++ b/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp | 101 | +++ b/IGC/AdaptorOCL/SPIRV/SPIRVReader.cpp |
99 | @@ -1576,7 +1576,11 @@ void SPIRVToLLVMDbgTran::transDbgInfo(SPIRVValue *SV, Value *V) { | 102 | @@ -1610,7 +1610,7 @@ void SPIRVToLLVMDbgTran::transDbgInfo(SPIRVValue *SV, Value *V) { |
100 | Line->getColumn(), scope, iat); | 103 | Line->getColumn(), scope, iat); |
101 | 104 | ||
102 | if(scope && !isa<DIFile>(scope)) | 105 | if(scope && !isa<DIFile>(scope)) |
103 | +#if LLVM_VERSION_MAJOR >= 12 | 106 | - I->setDebugLoc(DebugLoc::get(Line->getLine(), Line->getColumn(), |
104 | + I->setDebugLoc(DILocation::get(scope->getContext(), Line->getLine(), Line->getColumn(), | 107 | + I->setDebugLoc(DILocation::get(scope->getContext(), Line->getLine(), Line->getColumn(), |
105 | +#else | ||
106 | I->setDebugLoc(DebugLoc::get(Line->getLine(), Line->getColumn(), | ||
107 | +#endif | ||
108 | scope, iat)); | 108 | scope, iat)); |
109 | } | 109 | } |
110 | } | 110 | } |
111 | @@ -1925,7 +1929,11 @@ SPIRVToLLVM::transType(SPIRVType *T) { | 111 | @@ -2063,7 +2063,7 @@ SPIRVToLLVM::transType(SPIRVType *T) { |
112 | auto name = isSubgroupAvcINTELTypeOpCode(OC) ? | 112 | auto name = isSubgroupAvcINTELTypeOpCode(OC) ? |
113 | OCLSubgroupINTELTypeOpCodeMap::rmap(OC) : | 113 | OCLSubgroupINTELTypeOpCodeMap::rmap(OC) : |
114 | BuiltinOpaqueGenericTypeOpCodeMap::rmap(OC); | 114 | BuiltinOpaqueGenericTypeOpCodeMap::rmap(OC); |
115 | +#if LLVM_VERSION_MAJOR >= 12 | 115 | - auto *pST = M->getTypeByName(name); |
116 | + auto *pST = llvm::StructType::getTypeByName(M->getContext(), name); | 116 | + auto *pST = IGCLLVM::getTypeByName(M, name); |
117 | +#else | ||
118 | auto *pST = M->getTypeByName(name); | ||
119 | +#endif | ||
120 | pST = pST ? pST : StructType::create(*Context, name); | 117 | pST = pST ? pST : StructType::create(*Context, name); |
121 | 118 | ||
122 | return mapType(T, PointerType::get(pST, getOCLOpaqueTypeAddrSpace(OC))); | 119 | return mapType(T, PointerType::get(pST, getOCLOpaqueTypeAddrSpace(OC))); |
123 | @@ -2403,7 +2411,7 @@ Value *SPIRVToLLVM::promoteBool(Value *pVal, BasicBlock *BB) | 120 | @@ -2541,7 +2541,7 @@ Value *SPIRVToLLVM::promoteBool(Value *pVal, BasicBlock *BB) |
124 | 121 | ||
125 | auto *PromoType = isa<VectorType>(pVal->getType()) ? | 122 | auto *PromoType = isa<VectorType>(pVal->getType()) ? |
126 | cast<Type>(IGCLLVM::FixedVectorType::get(Type::getInt8Ty(pVal->getContext()), | 123 | cast<Type>(IGCLLVM::FixedVectorType::get(Type::getInt8Ty(pVal->getContext()), |
@@ -129,7 +126,7 @@ index 41d1f2ea..725a1512 100644 | |||
129 | Type::getInt8Ty(pVal->getContext()); | 126 | Type::getInt8Ty(pVal->getContext()); |
130 | 127 | ||
131 | if (auto *C = dyn_cast<Constant>(pVal)) | 128 | if (auto *C = dyn_cast<Constant>(pVal)) |
132 | @@ -2445,7 +2453,7 @@ Value *SPIRVToLLVM::truncBool(Value *pVal, BasicBlock *BB) | 129 | @@ -2583,7 +2583,7 @@ Value *SPIRVToLLVM::truncBool(Value *pVal, BasicBlock *BB) |
133 | 130 | ||
134 | auto *TruncType = isa<VectorType>(pVal->getType()) ? | 131 | auto *TruncType = isa<VectorType>(pVal->getType()) ? |
135 | cast<Type>(IGCLLVM::FixedVectorType::get(Type::getInt1Ty(pVal->getContext()), | 132 | cast<Type>(IGCLLVM::FixedVectorType::get(Type::getInt1Ty(pVal->getContext()), |
@@ -138,7 +135,7 @@ index 41d1f2ea..725a1512 100644 | |||
138 | Type::getInt1Ty(pVal->getContext()); | 135 | Type::getInt1Ty(pVal->getContext()); |
139 | 136 | ||
140 | if (auto *C = dyn_cast<Constant>(pVal)) | 137 | if (auto *C = dyn_cast<Constant>(pVal)) |
141 | @@ -2491,7 +2499,7 @@ Type *SPIRVToLLVM::truncBoolType(SPIRVType *SPVType, Type *LLType) | 138 | @@ -2629,7 +2629,7 @@ Type *SPIRVToLLVM::truncBoolType(SPIRVType *SPVType, Type *LLType) |
142 | 139 | ||
143 | return isa<VectorType>(LLType) ? | 140 | return isa<VectorType>(LLType) ? |
144 | cast<Type>(IGCLLVM::FixedVectorType::get(Type::getInt1Ty(LLType->getContext()), | 141 | cast<Type>(IGCLLVM::FixedVectorType::get(Type::getInt1Ty(LLType->getContext()), |
@@ -147,7 +144,7 @@ index 41d1f2ea..725a1512 100644 | |||
147 | Type::getInt1Ty(LLType->getContext()); | 144 | Type::getInt1Ty(LLType->getContext()); |
148 | } | 145 | } |
149 | 146 | ||
150 | @@ -2648,7 +2656,7 @@ SPIRVToLLVM::transValueWithoutDecoration(SPIRVValue *BV, Function *F, | 147 | @@ -2837,7 +2837,7 @@ SPIRVToLLVM::transValueWithoutDecoration(SPIRVValue *BV, Function *F, |
151 | { | 148 | { |
152 | if(CV[i]->getType()->isVectorTy()) | 149 | if(CV[i]->getType()->isVectorTy()) |
153 | { | 150 | { |
@@ -156,7 +153,7 @@ index 41d1f2ea..725a1512 100644 | |||
156 | { | 153 | { |
157 | Value *v = ExtractElementInst::Create( CV[i],ConstantInt::get( *Context,APInt( 32,j ) ),BCC->getName(),BB ); | 154 | Value *v = ExtractElementInst::Create( CV[i],ConstantInt::get( *Context,APInt( 32,j ) ),BCC->getName(),BB ); |
158 | elm1 = CreateCompositeConstruct( elm1,v,pos++ ); | 155 | elm1 = CreateCompositeConstruct( elm1,v,pos++ ); |
159 | @@ -3336,7 +3344,7 @@ SPIRVToLLVM::transValueWithoutDecoration(SPIRVValue *BV, Function *F, | 156 | @@ -3508,7 +3508,7 @@ SPIRVToLLVM::transValueWithoutDecoration(SPIRVValue *BV, Function *F, |
160 | auto Vector = transValue(BI->getOperand(0), F, BB); | 157 | auto Vector = transValue(BI->getOperand(0), F, BB); |
161 | auto Scalar = transValue(BI->getOperand(1), F, BB); | 158 | auto Scalar = transValue(BI->getOperand(1), F, BB); |
162 | 159 | ||
@@ -165,7 +162,7 @@ index 41d1f2ea..725a1512 100644 | |||
165 | auto Undef = UndefValue::get(VecType); | 162 | auto Undef = UndefValue::get(VecType); |
166 | 163 | ||
167 | auto ScalarVec = InsertElementInst::Create(Undef, Scalar, | 164 | auto ScalarVec = InsertElementInst::Create(Undef, Scalar, |
168 | @@ -3361,7 +3369,7 @@ SPIRVToLLVM::transValueWithoutDecoration(SPIRVValue *BV, Function *F, | 165 | @@ -3533,7 +3533,7 @@ SPIRVToLLVM::transValueWithoutDecoration(SPIRVValue *BV, Function *F, |
169 | a->getType()->getScalarSizeInBits() - 1); | 166 | a->getType()->getScalarSizeInBits() - 1); |
170 | auto *ShiftOp = isa<VectorType>(a->getType()) ? | 167 | auto *ShiftOp = isa<VectorType>(a->getType()) ? |
171 | ConstantVector::getSplat( | 168 | ConstantVector::getSplat( |
@@ -174,7 +171,7 @@ index 41d1f2ea..725a1512 100644 | |||
174 | ShiftAmt; | 171 | ShiftAmt; |
175 | 172 | ||
176 | // OCL C: | 173 | // OCL C: |
177 | @@ -3705,15 +3713,15 @@ SPIRVToLLVM::transSPIRVBuiltinFromInst(SPIRVInstruction *BI, BasicBlock *BB) { | 174 | @@ -3887,15 +3887,15 @@ SPIRVToLLVM::transSPIRVBuiltinFromInst(SPIRVInstruction *BI, BasicBlock *BB) { |
178 | "", | 175 | "", |
179 | BB); | 176 | BB); |
180 | } | 177 | } |
@@ -194,22 +191,19 @@ index 41d1f2ea..725a1512 100644 | |||
194 | 191 | ||
195 | imageCoordinateWiden = new ShuffleVectorInst( | 192 | imageCoordinateWiden = new ShuffleVectorInst( |
196 | diff --git a/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp b/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp | 193 | diff --git a/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp b/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp |
197 | index 57821556..91b4623c 100644 | 194 | index 73ffca030..ecb045b88 100644 |
198 | --- a/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp | 195 | --- a/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp |
199 | +++ b/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp | 196 | +++ b/IGC/AdaptorOCL/SPIRV/SPIRVUtil.cpp |
200 | @@ -93,7 +93,11 @@ saveLLVMModule(Module *M, const std::string &OutputFile) { | 197 | @@ -76,7 +76,7 @@ saveLLVMModule(Module *M, const std::string &OutputFile) { |
201 | PointerType* | 198 | PointerType* |
202 | getOrCreateOpaquePtrType(Module *M, const std::string &Name, | 199 | getOrCreateOpaquePtrType(Module *M, const std::string &Name, |
203 | unsigned AddrSpace) { | 200 | unsigned AddrSpace) { |
204 | +#if LLVM_VERSION_MAJOR >= 12 | 201 | - auto OpaqueType = M->getTypeByName(Name); |
205 | + auto OpaqueType = llvm::StructType::getTypeByName(M->getContext(), Name); | 202 | + auto OpaqueType = IGCLLVM::getTypeByName(M, Name); |
206 | +#else | ||
207 | auto OpaqueType = M->getTypeByName(Name); | ||
208 | +#endif | ||
209 | if (!OpaqueType) | 203 | if (!OpaqueType) |
210 | OpaqueType = StructType::create(M->getContext(), Name); | 204 | OpaqueType = StructType::create(M->getContext(), Name); |
211 | return PointerType::get(OpaqueType, AddrSpace); | 205 | return PointerType::get(OpaqueType, AddrSpace); |
212 | @@ -159,7 +163,7 @@ std::string recursive_mangle(const Type* pType) | 206 | @@ -142,7 +142,7 @@ std::string recursive_mangle(const Type* pType) |
213 | return "i" + utostr(pType->getIntegerBitWidth()); | 207 | return "i" + utostr(pType->getIntegerBitWidth()); |
214 | case IGCLLVM::VectorTyID: | 208 | case IGCLLVM::VectorTyID: |
215 | { | 209 | { |
@@ -219,26 +213,31 @@ index 57821556..91b4623c 100644 | |||
219 | return "v" + utostr(vecLen) + recursive_mangle(pEltType); | 213 | return "v" + utostr(vecLen) + recursive_mangle(pEltType); |
220 | } | 214 | } |
221 | diff --git a/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp b/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp | 215 | diff --git a/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp b/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp |
222 | index 81acc6ce..fc45a510 100644 | 216 | index 876cb0a30..9621c4d53 100644 |
223 | --- a/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp | 217 | --- a/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp |
224 | +++ b/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp | 218 | +++ b/IGC/Compiler/CISACodeGen/AdvMemOpt.cpp |
225 | @@ -134,7 +134,11 @@ bool AdvMemOpt::runOnFunction(Function& F) { | 219 | @@ -17,6 +17,7 @@ SPDX-License-Identifier: MIT |
220 | #include <llvm/Support/Debug.h> | ||
221 | #include <llvm/Support/raw_ostream.h> | ||
222 | #include <llvm/Transforms/Utils/Local.h> | ||
223 | +#include "llvmWrapper/Transforms/Utils/LoopUtils.h" | ||
224 | #include "common/LLVMWarningsPop.hpp" | ||
225 | #include "GenISAIntrinsics/GenIntrinsics.h" | ||
226 | #include "Compiler/CISACodeGen/ShaderCodeGen.hpp" | ||
227 | @@ -118,7 +119,7 @@ bool AdvMemOpt::runOnFunction(Function& F) { | ||
226 | for (auto I = LI->begin(), E = LI->end(); I != E; ++I) | 228 | for (auto I = LI->begin(), E = LI->end(); I != E; ++I) |
227 | for (auto DFI = df_begin(*I), DFE = df_end(*I); DFI != DFE; ++DFI) { | 229 | for (auto DFI = df_begin(*I), DFE = df_end(*I); DFI != DFE; ++DFI) { |
228 | Loop* L = *DFI; | 230 | Loop* L = *DFI; |
229 | +#if LLVM_VERSION_MAJOR >= 12 | 231 | - if (L->empty()) |
230 | + if (L->isInnermost()) | 232 | + if (IGCLLVM::isInnermost(L)) |
231 | +#else | ||
232 | if (L->empty()) | ||
233 | +#endif | ||
234 | InnermostLoops.push_back(L); | 233 | InnermostLoops.push_back(L); |
235 | } | 234 | } |
236 | 235 | ||
237 | diff --git a/IGC/Compiler/CISACodeGen/CShader.cpp b/IGC/Compiler/CISACodeGen/CShader.cpp | 236 | diff --git a/IGC/Compiler/CISACodeGen/CShader.cpp b/IGC/Compiler/CISACodeGen/CShader.cpp |
238 | index e3cb8153..ebc99615 100644 | 237 | index 04b59cf00..16b77b050 100644 |
239 | --- a/IGC/Compiler/CISACodeGen/CShader.cpp | 238 | --- a/IGC/Compiler/CISACodeGen/CShader.cpp |
240 | +++ b/IGC/Compiler/CISACodeGen/CShader.cpp | 239 | +++ b/IGC/Compiler/CISACodeGen/CShader.cpp |
241 | @@ -430,7 +430,7 @@ void CShader::CreateAliasVars() | 240 | @@ -427,7 +427,7 @@ void CShader::CreateAliasVars() |
242 | continue; | 241 | continue; |
243 | 242 | ||
244 | Type* Ty = V->getType(); | 243 | Type* Ty = V->getType(); |
@@ -247,7 +246,7 @@ index e3cb8153..ebc99615 100644 | |||
247 | Type* BTy = VTy ? VTy->getElementType() : Ty; | 246 | Type* BTy = VTy ? VTy->getElementType() : Ty; |
248 | int nelts = (VTy ? (int)VTy->getNumElements() : 1); | 247 | int nelts = (VTy ? (int)VTy->getNumElements() : 1); |
249 | 248 | ||
250 | @@ -961,7 +961,7 @@ bool CShader::InsideDivergentCF(llvm::Instruction* inst) | 249 | @@ -1039,7 +1039,7 @@ bool CShader::InsideWorkgroupDivergentCF(const llvm::Instruction* inst) const |
251 | uint CShader::GetNbVectorElementAndMask(llvm::Value* val, uint32_t& mask) | 250 | uint CShader::GetNbVectorElementAndMask(llvm::Value* val, uint32_t& mask) |
252 | { | 251 | { |
253 | llvm::Type* type = val->getType(); | 252 | llvm::Type* type = val->getType(); |
@@ -256,25 +255,16 @@ index e3cb8153..ebc99615 100644 | |||
256 | mask = 0; | 255 | mask = 0; |
257 | // we don't process vector bigger than 31 elements as the mask has only 32bits | 256 | // we don't process vector bigger than 31 elements as the mask has only 32bits |
258 | // If we want to support longer vectors we need to extend the mask size | 257 | // If we want to support longer vectors we need to extend the mask size |
259 | @@ -1166,7 +1166,7 @@ uint32_t CShader::GetExtractMask(llvm::Value* vecVal) | 258 | @@ -1244,7 +1244,7 @@ CShader::ExtractMaskWrapper::ExtractMaskWrapper(CShader* pS, Value* VecVal) |
260 | { | 259 | m_EM = it->second; |
261 | return it->second; | 260 | return; |
262 | } | 261 | } |
263 | - const unsigned int numChannels = vecVal->getType()->isVectorTy() ? (unsigned)cast<VectorType>(vecVal->getType())->getNumElements() : 1; | 262 | - VectorType* VTy = dyn_cast<VectorType>(VecVal->getType()); |
264 | + const unsigned int numChannels = vecVal->getType()->isVectorTy() ? (unsigned)cast<IGCLLVM::FixedVectorType>(vecVal->getType())->getNumElements() : 1; | 263 | + IGCLLVM::FixedVectorType* VTy = dyn_cast<IGCLLVM::FixedVectorType>(VecVal->getType()); |
265 | IGC_ASSERT_MESSAGE(numChannels <= 32, "Mask has 32 bits maximally!"); | 264 | const unsigned int numChannels = VTy ? (unsigned)VTy->getNumElements() : 1; |
266 | return (1ULL << numChannels) - 1; | 265 | if (numChannels <= 32) |
267 | } | ||
268 | @@ -1174,7 +1174,7 @@ uint32_t CShader::GetExtractMask(llvm::Value* vecVal) | ||
269 | uint16_t CShader::AdjustExtractIndex(llvm::Value* vecVal, uint16_t index) | ||
270 | { | ||
271 | uint16_t result = index; | ||
272 | - if (cast<VectorType>(vecVal->getType())->getNumElements() < 32) | ||
273 | + if (cast<IGCLLVM::FixedVectorType>(vecVal->getType())->getNumElements() < 32) | ||
274 | { | 266 | { |
275 | uint32_t mask = GetExtractMask(vecVal); | 267 | @@ -1768,7 +1768,7 @@ CVariable* CShader::GetStructVariable(llvm::Value* v, bool forceVectorInit) |
276 | for (uint i = 0; i < index; ++i) | ||
277 | @@ -1591,7 +1591,7 @@ auto sizeToSIMDMode = [](uint32_t size) | ||
278 | 268 | ||
279 | CVariable* CShader::GetConstant(llvm::Constant* C, CVariable* dstVar) | 269 | CVariable* CShader::GetConstant(llvm::Constant* C, CVariable* dstVar) |
280 | { | 270 | { |
@@ -283,7 +273,7 @@ index e3cb8153..ebc99615 100644 | |||
283 | if (C && VTy) | 273 | if (C && VTy) |
284 | { // Vector constant | 274 | { // Vector constant |
285 | llvm::Type* eTy = VTy->getElementType(); | 275 | llvm::Type* eTy = VTy->getElementType(); |
286 | @@ -1816,7 +1816,7 @@ uint32_t CShader::GetNumElts(llvm::Type* type, bool isUniform) | 276 | @@ -1996,7 +1996,7 @@ uint32_t CShader::GetNumElts(llvm::Type* type, bool isUniform) |
287 | { | 277 | { |
288 | IGC_ASSERT(type->getContainedType(0)->isIntegerTy() || type->getContainedType(0)->isFloatingPointTy()); | 278 | IGC_ASSERT(type->getContainedType(0)->isIntegerTy() || type->getContainedType(0)->isFloatingPointTy()); |
289 | 279 | ||
@@ -291,21 +281,30 @@ index e3cb8153..ebc99615 100644 | |||
291 | + auto VT = cast<IGCLLVM::FixedVectorType>(type); | 281 | + auto VT = cast<IGCLLVM::FixedVectorType>(type); |
292 | numElts *= (uint16_t)VT->getNumElements(); | 282 | numElts *= (uint16_t)VT->getNumElements(); |
293 | } | 283 | } |
294 | return numElts; | 284 | else if (type->isStructTy()) |
295 | @@ -2516,7 +2516,7 @@ CVariable* CShader::GetSymbol(llvm::Value* value, bool fromConstantPool) | 285 | @@ -2709,7 +2709,7 @@ CVariable* CShader::GetSymbol(llvm::Value* value, bool fromConstantPool) |
296 | if (isVecType) | 286 | if (isVecType) |
297 | { | 287 | { |
298 | // Map the entire vector value to the CVar | 288 | // Map the entire vector value to the CVar |
299 | - unsigned numElements = (unsigned)cast<VectorType>(value->getType())->getNumElements(); | 289 | - unsigned numElements = (unsigned)cast<VectorType>(value->getType())->getNumElements(); |
300 | + unsigned numElements = (unsigned)cast<IGCLLVM::FixedVectorType>(value->getType())->getNumElements(); | 290 | + unsigned numElements = (unsigned)cast<IGCLLVM::FixedVectorType>(value->getType())->getNumElements(); |
301 | var = GetNewVariable(numElements, ISA_TYPE_UQ, (GetContext()->platform.getGRFSize() == 64) ? EALIGN_32WORD : EALIGN_HWORD, true, 1, valName); | 291 | var = GetNewVariable(numElements, ISA_TYPE_UQ, |
302 | symbolMapping.insert(std::pair<llvm::Value*, CVariable*>(value, var)); | 292 | (GetContext()->platform.getGRFSize() == 64) ? EALIGN_32WORD : EALIGN_HWORD, |
303 | 293 | WIBaseClass::UNIFORM_GLOBAL, 1, valName); | |
294 | @@ -3619,7 +3619,7 @@ unsigned int CShader::GetPrimitiveTypeSizeInRegisterInBits(const Type* Ty) const | ||
295 | { | ||
296 | sizeInBits = | ||
297 | GetContext()->getRegisterPointerSizeInBits(Ty->getPointerAddressSpace()); | ||
298 | - if (auto* VTy = dyn_cast<VectorType>(Ty)) | ||
299 | + if (auto* VTy = dyn_cast<IGCLLVM::FixedVectorType>(Ty)) | ||
300 | { | ||
301 | sizeInBits *= (unsigned)VTy->getNumElements(); | ||
302 | } | ||
304 | diff --git a/IGC/Compiler/CISACodeGen/ConstantCoalescing.cpp b/IGC/Compiler/CISACodeGen/ConstantCoalescing.cpp | 303 | diff --git a/IGC/Compiler/CISACodeGen/ConstantCoalescing.cpp b/IGC/Compiler/CISACodeGen/ConstantCoalescing.cpp |
305 | index 1efe116f..47b7c6c0 100644 | 304 | index 43fcfc7c0..8d58414b6 100644 |
306 | --- a/IGC/Compiler/CISACodeGen/ConstantCoalescing.cpp | 305 | --- a/IGC/Compiler/CISACodeGen/ConstantCoalescing.cpp |
307 | +++ b/IGC/Compiler/CISACodeGen/ConstantCoalescing.cpp | 306 | +++ b/IGC/Compiler/CISACodeGen/ConstantCoalescing.cpp |
308 | @@ -224,7 +224,7 @@ static void checkInsertExtractMatch(InsertElementInst* insertInst, Value* base, | 307 | @@ -182,7 +182,7 @@ static void checkInsertExtractMatch(InsertElementInst* insertInst, Value* base, |
309 | 308 | ||
310 | static bool canReplaceInsert(InsertElementInst* insertElt) | 309 | static bool canReplaceInsert(InsertElementInst* insertElt) |
311 | { | 310 | { |
@@ -314,7 +313,7 @@ index 1efe116f..47b7c6c0 100644 | |||
314 | ConstantInt* index = dyn_cast<ConstantInt>(insertElt->getOperand(2)); | 313 | ConstantInt* index = dyn_cast<ConstantInt>(insertElt->getOperand(2)); |
315 | if (!index || index->getZExtValue() != VTy->getNumElements() - 1) | 314 | if (!index || index->getZExtValue() != VTy->getNumElements() - 1) |
316 | { | 315 | { |
317 | @@ -312,7 +312,7 @@ void ConstantCoalescing::VectorizePrep(llvm::BasicBlock* bb) | 316 | @@ -270,7 +270,7 @@ void ConstantCoalescing::VectorizePrep(llvm::BasicBlock* bb) |
318 | { | 317 | { |
319 | if (load->getType()->isVectorTy() && wiAns->isUniform(load)) | 318 | if (load->getType()->isVectorTy() && wiAns->isUniform(load)) |
320 | { | 319 | { |
@@ -323,7 +322,7 @@ index 1efe116f..47b7c6c0 100644 | |||
323 | DenseMap<uint64_t, Instruction*> extractElementMap; | 322 | DenseMap<uint64_t, Instruction*> extractElementMap; |
324 | 323 | ||
325 | for (auto iter = load->user_begin(); iter != load->user_end(); iter++) | 324 | for (auto iter = load->user_begin(); iter != load->user_end(); iter++) |
326 | @@ -396,7 +396,7 @@ bool ConstantCoalescing::isProfitableLoad( | 325 | @@ -354,7 +354,7 @@ bool ConstantCoalescing::isProfitableLoad( |
327 | (isa<LoadInst>(I) && wiAns->isUniform(I)) ? | 326 | (isa<LoadInst>(I) && wiAns->isUniform(I)) ? |
328 | 16 : 4; | 327 | 16 : 4; |
329 | 328 | ||
@@ -332,7 +331,7 @@ index 1efe116f..47b7c6c0 100644 | |||
332 | return false; | 331 | return false; |
333 | 332 | ||
334 | MaxEltPlus = CheckVectorElementUses(I); | 333 | MaxEltPlus = CheckVectorElementUses(I); |
335 | @@ -1787,7 +1787,7 @@ void ConstantCoalescing::AdjustChunk(BufChunk* cov_chunk, uint start_adj, uint s | 334 | @@ -1933,7 +1933,7 @@ void ConstantCoalescing::AdjustChunk(BufChunk* cov_chunk, uint start_adj, uint s |
336 | WIAnalysis::WIDependancy loadDep = wiAns->whichDepend(cov_chunk->chunkIO); | 335 | WIAnalysis::WIDependancy loadDep = wiAns->whichDepend(cov_chunk->chunkIO); |
337 | irBuilder->SetInsertPoint(cov_chunk->chunkIO->getNextNode()); | 336 | irBuilder->SetInsertPoint(cov_chunk->chunkIO->getNextNode()); |
338 | Value* vec = UndefValue::get(originalType); | 337 | Value* vec = UndefValue::get(originalType); |
@@ -341,7 +340,7 @@ index 1efe116f..47b7c6c0 100644 | |||
341 | { | 340 | { |
342 | Value* channel = irBuilder->CreateExtractElement( | 341 | Value* channel = irBuilder->CreateExtractElement( |
343 | cov_chunk->chunkIO, irBuilder->getInt32(i + start_adj)); | 342 | cov_chunk->chunkIO, irBuilder->getInt32(i + start_adj)); |
344 | @@ -1851,7 +1851,7 @@ void ConstantCoalescing::MoveExtracts(BufChunk* cov_chunk, Instruction* load, ui | 343 | @@ -1997,7 +1997,7 @@ void ConstantCoalescing::MoveExtracts(BufChunk* cov_chunk, Instruction* load, ui |
345 | irBuilder->SetInsertPoint(load->getNextNode()); | 344 | irBuilder->SetInsertPoint(load->getNextNode()); |
346 | Type* vecType = load->getType(); | 345 | Type* vecType = load->getType(); |
347 | Value* vec = UndefValue::get(vecType); | 346 | Value* vec = UndefValue::get(vecType); |
@@ -350,7 +349,7 @@ index 1efe116f..47b7c6c0 100644 | |||
350 | { | 349 | { |
351 | Value* channel = irBuilder->CreateExtractElement( | 350 | Value* channel = irBuilder->CreateExtractElement( |
352 | cov_chunk->chunkIO, irBuilder->getInt32(i + start_adj)); | 351 | cov_chunk->chunkIO, irBuilder->getInt32(i + start_adj)); |
353 | @@ -1915,7 +1915,7 @@ void ConstantCoalescing::EnlargeChunk(BufChunk* cov_chunk, uint size_adj) | 352 | @@ -2061,7 +2061,7 @@ void ConstantCoalescing::EnlargeChunk(BufChunk* cov_chunk, uint size_adj) |
354 | WIAnalysis::WIDependancy loadDep = wiAns->whichDepend(cov_chunk->chunkIO); | 353 | WIAnalysis::WIDependancy loadDep = wiAns->whichDepend(cov_chunk->chunkIO); |
355 | irBuilder->SetInsertPoint(cov_chunk->chunkIO->getNextNode()); | 354 | irBuilder->SetInsertPoint(cov_chunk->chunkIO->getNextNode()); |
356 | Value* vec = UndefValue::get(originalType); | 355 | Value* vec = UndefValue::get(originalType); |
@@ -359,7 +358,7 @@ index 1efe116f..47b7c6c0 100644 | |||
359 | { | 358 | { |
360 | Value* channel = irBuilder->CreateExtractElement( | 359 | Value* channel = irBuilder->CreateExtractElement( |
361 | cov_chunk->chunkIO, irBuilder->getInt32(i)); | 360 | cov_chunk->chunkIO, irBuilder->getInt32(i)); |
362 | @@ -2343,7 +2343,7 @@ void ConstantCoalescing::ReplaceLoadWithSamplerLoad( | 361 | @@ -2498,7 +2498,7 @@ void ConstantCoalescing::ReplaceLoadWithSamplerLoad( |
363 | if (dstTy->isVectorTy()) | 362 | if (dstTy->isVectorTy()) |
364 | { | 363 | { |
365 | result = UndefValue::get(dstTy); | 364 | result = UndefValue::get(dstTy); |
@@ -369,18 +368,18 @@ index 1efe116f..47b7c6c0 100644 | |||
369 | Value* tmpData = ExtractFromSamplerData(cast<VectorType>(dstTy)->getElementType(), i); | 368 | Value* tmpData = ExtractFromSamplerData(cast<VectorType>(dstTy)->getElementType(), i); |
370 | result = irBuilder->CreateInsertElement(result, tmpData, irBuilder->getInt32(i)); | 369 | result = irBuilder->CreateInsertElement(result, tmpData, irBuilder->getInt32(i)); |
371 | diff --git a/IGC/Compiler/CISACodeGen/DeSSA.cpp b/IGC/Compiler/CISACodeGen/DeSSA.cpp | 370 | diff --git a/IGC/Compiler/CISACodeGen/DeSSA.cpp b/IGC/Compiler/CISACodeGen/DeSSA.cpp |
372 | index f540dc11..6fab0246 100644 | 371 | index bc06a00d2..8ac30bac2 100644 |
373 | --- a/IGC/Compiler/CISACodeGen/DeSSA.cpp | 372 | --- a/IGC/Compiler/CISACodeGen/DeSSA.cpp |
374 | +++ b/IGC/Compiler/CISACodeGen/DeSSA.cpp | 373 | +++ b/IGC/Compiler/CISACodeGen/DeSSA.cpp |
375 | @@ -83,6 +83,7 @@ IN THE SOFTWARE. | 374 | @@ -67,6 +67,7 @@ See LICENSE.TXT for details. |
376 | #include "Compiler/IGCPassSupport.h" | ||
377 | #include "common/LLVMWarningsPush.hpp" | 375 | #include "common/LLVMWarningsPush.hpp" |
376 | #include "llvmWrapper/IR/Instructions.h" | ||
378 | #include <llvm/IR/InstIterator.h> | 377 | #include <llvm/IR/InstIterator.h> |
379 | +#include <llvmWrapper/IR/DerivedTypes.h> | 378 | +#include <llvmWrapper/IR/DerivedTypes.h> |
379 | #include <llvm/IR/InlineAsm.h> | ||
380 | #include "common/LLVMWarningsPop.hpp" | 380 | #include "common/LLVMWarningsPop.hpp" |
381 | #include <algorithm> | 381 | #include <algorithm> |
382 | #include "Probe/Assertion.h" | 382 | @@ -1302,7 +1303,7 @@ int DeSSA::checkInsertElementAlias( |
383 | @@ -1284,7 +1285,7 @@ int DeSSA::checkInsertElementAlias( | ||
384 | // | 383 | // |
385 | // If found, return the actual vector size; | 384 | // If found, return the actual vector size; |
386 | // otherwise, return 0. | 385 | // otherwise, return 0. |
@@ -389,19 +388,18 @@ index f540dc11..6fab0246 100644 | |||
389 | IGC_ASSERT(nullptr != VTy); | 388 | IGC_ASSERT(nullptr != VTy); |
390 | int nelts = (int)VTy->getNumElements(); | 389 | int nelts = (int)VTy->getNumElements(); |
391 | AllIEIs.resize(nelts, nullptr); | 390 | AllIEIs.resize(nelts, nullptr); |
391 | @@ -1491,4 +1492,4 @@ bool DeSSA::alignInterfere(e_alignment a1, e_alignment a2) | ||
392 | return true; | ||
393 | } | ||
394 | return false; | ||
395 | -} | ||
396 | \ No newline at end of file | ||
397 | +} | ||
392 | diff --git a/IGC/Compiler/CISACodeGen/EmitVISAPass.cpp b/IGC/Compiler/CISACodeGen/EmitVISAPass.cpp | 398 | diff --git a/IGC/Compiler/CISACodeGen/EmitVISAPass.cpp b/IGC/Compiler/CISACodeGen/EmitVISAPass.cpp |
393 | index 3ab1cc5f..4502ef62 100644 | 399 | index dda1529eb..5541dee74 100644 |
394 | --- a/IGC/Compiler/CISACodeGen/EmitVISAPass.cpp | 400 | --- a/IGC/Compiler/CISACodeGen/EmitVISAPass.cpp |
395 | +++ b/IGC/Compiler/CISACodeGen/EmitVISAPass.cpp | 401 | +++ b/IGC/Compiler/CISACodeGen/EmitVISAPass.cpp |
396 | @@ -53,6 +53,7 @@ IN THE SOFTWARE. | 402 | @@ -1265,7 +1265,7 @@ bool EmitPass::canRelocatePhiMov( |
397 | #include "Compiler/IGCPassSupport.h" | ||
398 | #include "common/LLVMWarningsPush.hpp" | ||
399 | #include "llvmWrapper/IR/Instructions.h" | ||
400 | +#include "llvmWrapper/IR/DerivedTypes.h" | ||
401 | #include "llvm/Support/Path.h" | ||
402 | #include "llvmWrapper/IR/Intrinsics.h" | ||
403 | #include "common/LLVMWarningsPop.hpp" | ||
404 | @@ -1165,7 +1166,7 @@ bool EmitPass::canRelocatePhiMov( | ||
405 | if (dst != src) | 403 | if (dst != src) |
406 | { | 404 | { |
407 | int numElt = 1; | 405 | int numElt = 1; |
@@ -410,7 +408,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
410 | { | 408 | { |
411 | numElt = int_cast<int>(vTy->getNumElements()); | 409 | numElt = int_cast<int>(vTy->getNumElements()); |
412 | } | 410 | } |
413 | @@ -1302,7 +1303,7 @@ void EmitPass::MovPhiSources(llvm::BasicBlock* aBB) | 411 | @@ -1402,7 +1402,7 @@ void EmitPass::MovPhiSources(llvm::BasicBlock* aBB) |
414 | phiSrcDstList.push_back(phiInfo); | 412 | phiSrcDstList.push_back(phiInfo); |
415 | 413 | ||
416 | int numElt = 0; | 414 | int numElt = 0; |
@@ -419,7 +417,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
419 | { | 417 | { |
420 | numElt = int_cast<int>(vTy->getNumElements()); | 418 | numElt = int_cast<int>(vTy->getNumElements()); |
421 | } | 419 | } |
422 | @@ -5375,7 +5376,7 @@ void EmitPass::emitLegacySimdBlockWrite(llvm::Instruction* inst, llvm::Value* pt | 420 | @@ -5606,7 +5606,7 @@ void EmitPass::emitLegacySimdBlockWrite(llvm::Instruction* inst, llvm::Value* pt |
423 | bool useA64 = isA64Ptr(ptrType, m_currShader->GetContext()); | 421 | bool useA64 = isA64Ptr(ptrType, m_currShader->GetContext()); |
424 | 422 | ||
425 | Type* Ty = dataPtr->getType(); | 423 | Type* Ty = dataPtr->getType(); |
@@ -428,7 +426,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
428 | uint32_t nbElements = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; | 426 | uint32_t nbElements = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; |
429 | 427 | ||
430 | uint32_t typeSizeInBytes = Ty->getScalarSizeInBits() / 8; | 428 | uint32_t typeSizeInBytes = Ty->getScalarSizeInBits() / 8; |
431 | @@ -5621,7 +5622,7 @@ void EmitPass::emitLegacySimdBlockRead(llvm::Instruction* inst, llvm::Value* ptr | 429 | @@ -5852,7 +5852,7 @@ void EmitPass::emitLegacySimdBlockRead(llvm::Instruction* inst, llvm::Value* ptr |
432 | bool useA64 = isA64Ptr(ptrType, m_currShader->GetContext()); | 430 | bool useA64 = isA64Ptr(ptrType, m_currShader->GetContext()); |
433 | 431 | ||
434 | Type* Ty = inst->getType(); | 432 | Type* Ty = inst->getType(); |
@@ -437,7 +435,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
437 | uint32_t nbElements = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; | 435 | uint32_t nbElements = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; |
438 | 436 | ||
439 | uint32_t typeSizeInBytes = Ty->getScalarSizeInBits() / 8; | 437 | uint32_t typeSizeInBytes = Ty->getScalarSizeInBits() / 8; |
440 | @@ -5913,7 +5914,7 @@ void EmitPass::emitMediaBlockIO(const llvm::GenIntrinsicInst* inst, bool isRead) | 438 | @@ -6144,7 +6144,7 @@ void EmitPass::emitMediaBlockIO(const llvm::GenIntrinsicInst* inst, bool isRead) |
441 | }; | 439 | }; |
442 | 440 | ||
443 | uint nElts = isa<VectorType>(pDataType) ? | 441 | uint nElts = isa<VectorType>(pDataType) ? |
@@ -446,7 +444,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
446 | 1; | 444 | 1; |
447 | 445 | ||
448 | // Now, do the copies. | 446 | // Now, do the copies. |
449 | @@ -6003,7 +6004,7 @@ void EmitPass::emitSimdMediaBlockRead(llvm::Instruction* inst) | 447 | @@ -6234,7 +6234,7 @@ void EmitPass::emitSimdMediaBlockRead(llvm::Instruction* inst) |
450 | uint32_t nbElements = 1; | 448 | uint32_t nbElements = 1; |
451 | if (inst->getType()->isVectorTy()) | 449 | if (inst->getType()->isVectorTy()) |
452 | { | 450 | { |
@@ -455,7 +453,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
455 | } | 453 | } |
456 | IGC_ASSERT_MESSAGE(nbElements <= 8, "InValid Vector Size"); | 454 | IGC_ASSERT_MESSAGE(nbElements <= 8, "InValid Vector Size"); |
457 | 455 | ||
458 | @@ -6247,7 +6248,7 @@ void EmitPass::emitSimdMediaBlockWrite(llvm::Instruction* inst) | 456 | @@ -6478,7 +6478,7 @@ void EmitPass::emitSimdMediaBlockWrite(llvm::Instruction* inst) |
459 | uint32_t nbElements = 1; | 457 | uint32_t nbElements = 1; |
460 | if (dataPtr->getType()->isVectorTy()) | 458 | if (dataPtr->getType()->isVectorTy()) |
461 | { | 459 | { |
@@ -464,18 +462,29 @@ index 3ab1cc5f..4502ef62 100644 | |||
464 | } | 462 | } |
465 | IGC_ASSERT_MESSAGE(nbElements <= 8, "InValid Vector Size"); | 463 | IGC_ASSERT_MESSAGE(nbElements <= 8, "InValid Vector Size"); |
466 | 464 | ||
467 | @@ -8993,8 +8994,8 @@ void EmitPass::emitBitCast(llvm::BitCastInst* btCst) | 465 | @@ -9140,7 +9140,7 @@ void EmitPass::EmitInlineAsm(llvm::CallInst* inst) |
466 | { | ||
467 | if (inst->getType()->isVectorTy()) | ||
468 | { | ||
469 | - emitVectorCopy(dest, cv, int_cast<unsigned>(dyn_cast<VectorType>(inst->getType())->getNumElements())); | ||
470 | + emitVectorCopy(dest, cv, int_cast<unsigned>(dyn_cast<IGCLLVM::FixedVectorType>(inst->getType())->getNumElements())); | ||
471 | } | ||
472 | else | ||
473 | { | ||
474 | @@ -9415,9 +9415,9 @@ void EmitPass::emitBitCast(llvm::BitCastInst* btCst) | ||
468 | { | 475 | { |
469 | Type* srcType = btCst->getOperand(0)->getType(); | 476 | Type* srcType = btCst->getOperand(0)->getType(); |
470 | Type* dstType = btCst->getType(); | 477 | Type* dstType = btCst->getType(); |
471 | - unsigned int numSrcElement = srcType->isVectorTy() ? (unsigned)cast<VectorType>(srcType)->getNumElements() : 1; | 478 | - unsigned int numSrcElement = srcType->isVectorTy() ? (unsigned)cast<VectorType>(srcType)->getNumElements() : 1; |
472 | - unsigned int numDstElement = dstType->isVectorTy() ? (unsigned)cast<VectorType>(dstType)->getNumElements() : 1; | 479 | - unsigned int numDstElement = dstType->isVectorTy() ? (unsigned)cast<VectorType>(dstType)->getNumElements() : 1; |
480 | - | ||
473 | + unsigned int numSrcElement = srcType->isVectorTy() ? (unsigned)cast<IGCLLVM::FixedVectorType>(srcType)->getNumElements() : 1; | 481 | + unsigned int numSrcElement = srcType->isVectorTy() ? (unsigned)cast<IGCLLVM::FixedVectorType>(srcType)->getNumElements() : 1; |
474 | + unsigned int numDstElement = dstType->isVectorTy() ? (unsigned)cast<IGCLLVM::FixedVectorType>(dstType)->getNumElements() : 1; | 482 | + unsigned int numDstElement = dstType->isVectorTy() ? (unsigned)cast<IGCLLVM::FixedVectorType>(dstType)->getNumElements() : 1; |
475 | 483 | + | |
476 | if (srcType->isPointerTy()) | 484 | if (srcType->isPointerTy()) |
477 | { | 485 | { |
478 | @@ -9468,7 +9469,7 @@ void EmitPass::emitLoad3DInner(LdRawIntrinsic* inst, ResourceDescriptor& resourc | 486 | IGC_ASSERT_MESSAGE(dstType->isPointerTy(), "Expected both src and dst have pointer type."); |
487 | @@ -9863,7 +9863,7 @@ void EmitPass::emitLoad3DInner(LdRawIntrinsic* inst, ResourceDescriptor& resourc | ||
479 | { | 488 | { |
480 | IGC_ASSERT_MESSAGE(predDefSurface != ESURFACE_STATELESS, "scratch cannot be uniform"); | 489 | IGC_ASSERT_MESSAGE(predDefSurface != ESURFACE_STATELESS, "scratch cannot be uniform"); |
481 | Type* loadType = inst->getType(); | 490 | Type* loadType = inst->getType(); |
@@ -484,7 +493,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
484 | if (predDefSurface == ESURFACE_SLM) | 493 | if (predDefSurface == ESURFACE_SLM) |
485 | { | 494 | { |
486 | IGC_ASSERT(numElement <= 4); | 495 | IGC_ASSERT(numElement <= 4); |
487 | @@ -10795,7 +10796,7 @@ void EmitPass::emitInsert(llvm::Instruction* inst) | 496 | @@ -11278,7 +11278,7 @@ void EmitPass::emitInsert(llvm::Instruction* inst) |
488 | pVecVar = GetSymbol(pVec); | 497 | pVecVar = GetSymbol(pVec); |
489 | if (pVecVar != pInstVar) | 498 | if (pVecVar != pInstVar) |
490 | { | 499 | { |
@@ -493,16 +502,16 @@ index 3ab1cc5f..4502ef62 100644 | |||
493 | } | 502 | } |
494 | } | 503 | } |
495 | } | 504 | } |
496 | @@ -14596,7 +14597,7 @@ bool EmitPass::isUniformStoreOCL(llvm::StoreInst* SI) | 505 | @@ -15233,7 +15233,7 @@ bool EmitPass::isUniformStoreOCL(Value* ptr, Value* storeVal) |
506 | } | ||
497 | 507 | ||
498 | Value* storeVal = SI->getValueOperand(); | ||
499 | Type* Ty = storeVal->getType(); | 508 | Type* Ty = storeVal->getType(); |
500 | - VectorType* VTy = dyn_cast<VectorType>(Ty); | 509 | - VectorType* VTy = dyn_cast<VectorType>(Ty); |
501 | + IGCLLVM::FixedVectorType* VTy = dyn_cast<IGCLLVM::FixedVectorType>(Ty); | 510 | + IGCLLVM::FixedVectorType* VTy = dyn_cast<IGCLLVM::FixedVectorType>(Ty); |
502 | uint32_t elts = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; | 511 | uint32_t elts = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; |
503 | Type* eltTy = VTy ? VTy->getElementType() : Ty; | 512 | Type* eltTy = VTy ? VTy->getElementType() : Ty; |
504 | 513 | ||
505 | @@ -14626,7 +14627,7 @@ void EmitPass::emitVectorBitCast(llvm::BitCastInst* BCI) | 514 | @@ -15269,7 +15269,7 @@ void EmitPass::emitVectorBitCast(llvm::BitCastInst* BCI) |
506 | if (srcTy->isVectorTy()) | 515 | if (srcTy->isVectorTy()) |
507 | { | 516 | { |
508 | srcEltTy = cast<VectorType>(srcTy)->getElementType(); | 517 | srcEltTy = cast<VectorType>(srcTy)->getElementType(); |
@@ -511,7 +520,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
511 | } | 520 | } |
512 | else | 521 | else |
513 | { | 522 | { |
514 | @@ -14636,7 +14637,7 @@ void EmitPass::emitVectorBitCast(llvm::BitCastInst* BCI) | 523 | @@ -15279,7 +15279,7 @@ void EmitPass::emitVectorBitCast(llvm::BitCastInst* BCI) |
515 | if (dstTy->isVectorTy()) | 524 | if (dstTy->isVectorTy()) |
516 | { | 525 | { |
517 | dstEltTy = cast<VectorType>(dstTy)->getElementType(); | 526 | dstEltTy = cast<VectorType>(dstTy)->getElementType(); |
@@ -520,7 +529,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
520 | } | 529 | } |
521 | else | 530 | else |
522 | { | 531 | { |
523 | @@ -15215,7 +15216,7 @@ void EmitPass::emitVectorLoad(LoadInst* inst, Value* offset, ConstantInt* immOff | 532 | @@ -15867,7 +15867,7 @@ void EmitPass::emitVectorLoad(LoadInst* inst, Value* offset, ConstantInt* immOff |
524 | } | 533 | } |
525 | 534 | ||
526 | Type* Ty = inst->getType(); | 535 | Type* Ty = inst->getType(); |
@@ -529,7 +538,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
529 | Type* eltTy = VTy ? VTy->getElementType() : Ty; | 538 | Type* eltTy = VTy ? VTy->getElementType() : Ty; |
530 | uint32_t eltBytes = GetScalarTypeSizeInRegister(eltTy); | 539 | uint32_t eltBytes = GetScalarTypeSizeInRegister(eltTy); |
531 | IGC_ASSERT_MESSAGE((eltBytes == 1) || (eltBytes == 2) || (eltBytes == 4) || (eltBytes == 8), | 540 | IGC_ASSERT_MESSAGE((eltBytes == 1) || (eltBytes == 2) || (eltBytes == 4) || (eltBytes == 8), |
532 | @@ -15667,7 +15668,7 @@ void EmitPass::emitVectorStore(StoreInst* inst, Value* offset, ConstantInt* immO | 541 | @@ -16381,7 +16381,7 @@ void EmitPass::emitVectorStore(StoreInst* inst, Value* offset, ConstantInt* immO |
533 | 542 | ||
534 | Value* storedVal = inst->getValueOperand(); | 543 | Value* storedVal = inst->getValueOperand(); |
535 | Type* Ty = storedVal->getType(); | 544 | Type* Ty = storedVal->getType(); |
@@ -538,7 +547,7 @@ index 3ab1cc5f..4502ef62 100644 | |||
538 | Type* eltTy = VTy ? VTy->getElementType() : Ty; | 547 | Type* eltTy = VTy ? VTy->getElementType() : Ty; |
539 | uint32_t eltBytes = GetScalarTypeSizeInRegister(eltTy); | 548 | uint32_t eltBytes = GetScalarTypeSizeInRegister(eltTy); |
540 | 549 | ||
541 | @@ -16471,7 +16472,7 @@ void EmitPass::emitCopyAll(CVariable* Dst, CVariable* Src, llvm::Type* Ty) | 550 | @@ -17189,7 +17189,7 @@ void EmitPass::emitCopyAll(CVariable* Dst, CVariable* Src, llvm::Type* Ty) |
542 | } | 551 | } |
543 | else if (Ty->isVectorTy()) | 552 | else if (Ty->isVectorTy()) |
544 | { | 553 | { |
@@ -546,12 +555,24 @@ index 3ab1cc5f..4502ef62 100644 | |||
546 | + unsigned NElts = (unsigned)cast<IGCLLVM::FixedVectorType>(Ty)->getNumElements(); | 555 | + unsigned NElts = (unsigned)cast<IGCLLVM::FixedVectorType>(Ty)->getNumElements(); |
547 | emitVectorCopy(Dst, Src, NElts); | 556 | emitVectorCopy(Dst, Src, NElts); |
548 | } | 557 | } |
549 | else | 558 | else if (Ty->isStructTy()) |
559 | @@ -18347,9 +18347,9 @@ void EmitPass::emitfcvt(llvm::GenIntrinsicInst* GII) | ||
560 | CVariable* dst = m_destination; | ||
561 | |||
562 | Type* dTy = GII->getType(); | ||
563 | - VectorType* dVTy = dyn_cast<VectorType>(dTy); | ||
564 | + IGCLLVM::FixedVectorType* dVTy = dyn_cast<IGCLLVM::FixedVectorType>(dTy); | ||
565 | Type* sTy = sVal->getType(); | ||
566 | - VectorType* sVTy = dyn_cast<VectorType>(sTy); | ||
567 | + IGCLLVM::FixedVectorType* sVTy = dyn_cast<IGCLLVM::FixedVectorType>(sTy); | ||
568 | int nelts = dVTy ? (int)dVTy->getNumElements() : 1; | ||
569 | int src_nelts = sVTy ? (int)sVTy->getNumElements() : 1; | ||
570 | if (nelts != src_nelts) | ||
550 | diff --git a/IGC/Compiler/CISACodeGen/Emu64OpsPass.cpp b/IGC/Compiler/CISACodeGen/Emu64OpsPass.cpp | 571 | diff --git a/IGC/Compiler/CISACodeGen/Emu64OpsPass.cpp b/IGC/Compiler/CISACodeGen/Emu64OpsPass.cpp |
551 | index 28248f30..767b477a 100644 | 572 | index 1c30ae361..db77b514d 100644 |
552 | --- a/IGC/Compiler/CISACodeGen/Emu64OpsPass.cpp | 573 | --- a/IGC/Compiler/CISACodeGen/Emu64OpsPass.cpp |
553 | +++ b/IGC/Compiler/CISACodeGen/Emu64OpsPass.cpp | 574 | +++ b/IGC/Compiler/CISACodeGen/Emu64OpsPass.cpp |
554 | @@ -1961,7 +1961,7 @@ bool InstExpander::visitExtractElement(ExtractElementInst& EEI) { | 575 | @@ -1945,7 +1945,7 @@ bool InstExpander::visitExtractElement(ExtractElementInst& EEI) { |
555 | // later. | 576 | // later. |
556 | 577 | ||
557 | Value* V = EEI.getVectorOperand(); | 578 | Value* V = EEI.getVectorOperand(); |
@@ -560,7 +581,7 @@ index 28248f30..767b477a 100644 | |||
560 | V = IRB->CreateBitCast(V, Emu->getV2Int32Ty(NumElts)); | 581 | V = IRB->CreateBitCast(V, Emu->getV2Int32Ty(NumElts)); |
561 | // Re-calculate indices to Lo and Hi parts. | 582 | // Re-calculate indices to Lo and Hi parts. |
562 | Value* Idx = EEI.getIndexOperand(); | 583 | Value* Idx = EEI.getIndexOperand(); |
563 | @@ -1998,7 +1998,7 @@ bool InstExpander::visitInsertElement(InsertElementInst& IEI) { | 584 | @@ -1982,7 +1982,7 @@ bool InstExpander::visitInsertElement(InsertElementInst& IEI) { |
564 | 585 | ||
565 | // Create the emulated vector. | 586 | // Create the emulated vector. |
566 | Value* NewVal = IEI.getOperand(0); | 587 | Value* NewVal = IEI.getOperand(0); |
@@ -570,10 +591,10 @@ index 28248f30..767b477a 100644 | |||
570 | // Re-calculate indices to Lo and Hi parts. | 591 | // Re-calculate indices to Lo and Hi parts. |
571 | Value* Idx = IEI.getOperand(2); | 592 | Value* Idx = IEI.getOperand(2); |
572 | diff --git a/IGC/Compiler/CISACodeGen/GenIRLowering.cpp b/IGC/Compiler/CISACodeGen/GenIRLowering.cpp | 593 | diff --git a/IGC/Compiler/CISACodeGen/GenIRLowering.cpp b/IGC/Compiler/CISACodeGen/GenIRLowering.cpp |
573 | index c74a2ac6..78e8a4aa 100644 | 594 | index 71e3bdf06..311964b9c 100644 |
574 | --- a/IGC/Compiler/CISACodeGen/GenIRLowering.cpp | 595 | --- a/IGC/Compiler/CISACodeGen/GenIRLowering.cpp |
575 | +++ b/IGC/Compiler/CISACodeGen/GenIRLowering.cpp | 596 | +++ b/IGC/Compiler/CISACodeGen/GenIRLowering.cpp |
576 | @@ -453,7 +453,7 @@ Value* GEPLowering::getSExtOrTrunc(Value* Val, Type* NewTy) const { | 597 | @@ -437,7 +437,7 @@ Value* GEPLowering::getSExtOrTrunc(Value* Val, Type* NewTy) const { |
577 | 598 | ||
578 | IGC_ASSERT_MESSAGE(OldTy->isIntOrIntVectorTy(), "Index should be Integer or vector of Integer!"); | 599 | IGC_ASSERT_MESSAGE(OldTy->isIntOrIntVectorTy(), "Index should be Integer or vector of Integer!"); |
579 | 600 | ||
@@ -582,7 +603,7 @@ index c74a2ac6..78e8a4aa 100644 | |||
582 | OldWidth = (unsigned)OldVecTy->getNumElements() * OldVecTy->getElementType()->getIntegerBitWidth(); | 603 | OldWidth = (unsigned)OldVecTy->getNumElements() * OldVecTy->getElementType()->getIntegerBitWidth(); |
583 | NewWidth = (unsigned)OldVecTy->getNumElements() * NewTy->getIntegerBitWidth(); | 604 | NewWidth = (unsigned)OldVecTy->getNumElements() * NewTy->getIntegerBitWidth(); |
584 | } | 605 | } |
585 | @@ -805,7 +805,7 @@ bool GEPLowering::lowerGetElementPtrInst(GetElementPtrInst* GEP) const | 606 | @@ -789,7 +789,7 @@ bool GEPLowering::lowerGetElementPtrInst(GetElementPtrInst* GEP) const |
586 | } | 607 | } |
587 | else | 608 | else |
588 | { | 609 | { |
@@ -592,10 +613,10 @@ index c74a2ac6..78e8a4aa 100644 | |||
592 | for (uint32_t j = 0; j < (uint32_t)NewIdxVT->getNumElements(); j++) { | 613 | for (uint32_t j = 0; j < (uint32_t)NewIdxVT->getNumElements(); j++) { |
593 | result = Builder->CreateInsertElement(result, PointerValue, Builder->getInt32(j)); | 614 | result = Builder->CreateInsertElement(result, PointerValue, Builder->getInt32(j)); |
594 | diff --git a/IGC/Compiler/CISACodeGen/GenSimplification.cpp b/IGC/Compiler/CISACodeGen/GenSimplification.cpp | 615 | diff --git a/IGC/Compiler/CISACodeGen/GenSimplification.cpp b/IGC/Compiler/CISACodeGen/GenSimplification.cpp |
595 | index 8c38f229..f432061e 100644 | 616 | index 23ab58538..f97482b6c 100644 |
596 | --- a/IGC/Compiler/CISACodeGen/GenSimplification.cpp | 617 | --- a/IGC/Compiler/CISACodeGen/GenSimplification.cpp |
597 | +++ b/IGC/Compiler/CISACodeGen/GenSimplification.cpp | 618 | +++ b/IGC/Compiler/CISACodeGen/GenSimplification.cpp |
598 | @@ -35,6 +35,7 @@ IN THE SOFTWARE. | 619 | @@ -19,6 +19,7 @@ SPDX-License-Identifier: MIT |
599 | #include <llvm/IR/InstVisitor.h> | 620 | #include <llvm/IR/InstVisitor.h> |
600 | #include <llvm/IR/Instruction.h> | 621 | #include <llvm/IR/Instruction.h> |
601 | #include <llvm/Support/raw_ostream.h> | 622 | #include <llvm/Support/raw_ostream.h> |
@@ -603,7 +624,7 @@ index 8c38f229..f432061e 100644 | |||
603 | #include "common/LLVMWarningsPop.hpp" | 624 | #include "common/LLVMWarningsPop.hpp" |
604 | #include "Probe/Assertion.h" | 625 | #include "Probe/Assertion.h" |
605 | 626 | ||
606 | @@ -164,7 +165,7 @@ bool GenSimplification::simplifyVectorPHINodeCase2(PHINode& PN) const { | 627 | @@ -148,7 +149,7 @@ bool GenSimplification::simplifyVectorPHINodeCase2(PHINode& PN) const { |
607 | 628 | ||
608 | Type* Ty = PN.getType(); | 629 | Type* Ty = PN.getType(); |
609 | Type* EltTy = Ty->getScalarType(); | 630 | Type* EltTy = Ty->getScalarType(); |
@@ -612,7 +633,7 @@ index 8c38f229..f432061e 100644 | |||
612 | 633 | ||
613 | SmallVector<Value*, 8> Lanes; | 634 | SmallVector<Value*, 8> Lanes; |
614 | SmallVector<SmallVector<Value*, 8>, 4> Values; | 635 | SmallVector<SmallVector<Value*, 8>, 4> Values; |
615 | @@ -238,7 +239,7 @@ void GenSimplification::visitPHINode(PHINode& PN) { | 636 | @@ -222,7 +223,7 @@ void GenSimplification::visitPHINode(PHINode& PN) { |
616 | void GenSimplification::visitExtractElement(ExtractElementInst& EEI) { | 637 | void GenSimplification::visitExtractElement(ExtractElementInst& EEI) { |
617 | // Skip non-2-element vector. | 638 | // Skip non-2-element vector. |
618 | Value* Vec = EEI.getVectorOperand(); | 639 | Value* Vec = EEI.getVectorOperand(); |
@@ -622,10 +643,10 @@ index 8c38f229..f432061e 100644 | |||
622 | return; | 643 | return; |
623 | 644 | ||
624 | diff --git a/IGC/Compiler/CISACodeGen/GeometryShaderLowering.cpp b/IGC/Compiler/CISACodeGen/GeometryShaderLowering.cpp | 645 | diff --git a/IGC/Compiler/CISACodeGen/GeometryShaderLowering.cpp b/IGC/Compiler/CISACodeGen/GeometryShaderLowering.cpp |
625 | index e00e8469..a60b9892 100644 | 646 | index 20638a0fe..6982ed252 100644 |
626 | --- a/IGC/Compiler/CISACodeGen/GeometryShaderLowering.cpp | 647 | --- a/IGC/Compiler/CISACodeGen/GeometryShaderLowering.cpp |
627 | +++ b/IGC/Compiler/CISACodeGen/GeometryShaderLowering.cpp | 648 | +++ b/IGC/Compiler/CISACodeGen/GeometryShaderLowering.cpp |
628 | @@ -33,6 +33,7 @@ IN THE SOFTWARE. | 649 | @@ -17,6 +17,7 @@ SPDX-License-Identifier: MIT |
629 | #include <llvm/IR/PassManager.h> | 650 | #include <llvm/IR/PassManager.h> |
630 | #include <llvm/IR/IRBuilder.h> | 651 | #include <llvm/IR/IRBuilder.h> |
631 | #include <llvm/IR/Verifier.h> | 652 | #include <llvm/IR/Verifier.h> |
@@ -633,7 +654,7 @@ index e00e8469..a60b9892 100644 | |||
633 | #include "common/LLVMWarningsPop.hpp" | 654 | #include "common/LLVMWarningsPop.hpp" |
634 | #include "Compiler/InitializePasses.h" | 655 | #include "Compiler/InitializePasses.h" |
635 | #include "Probe/Assertion.h" | 656 | #include "Probe/Assertion.h" |
636 | @@ -507,7 +508,7 @@ void GeometryShaderLowering::AddURBRead( | 657 | @@ -444,7 +445,7 @@ void GeometryShaderLowering::AddURBRead( |
637 | { | 658 | { |
638 | Value* vec = UndefValue::get(inst->getType()); | 659 | Value* vec = UndefValue::get(inst->getType()); |
639 | IRBuilder<> builder(inst); | 660 | IRBuilder<> builder(inst); |
@@ -643,10 +664,10 @@ index e00e8469..a60b9892 100644 | |||
643 | Value* vecElement = builder.CreateExtractElement(urbRead, builder.getInt32(i)); | 664 | Value* vecElement = builder.CreateExtractElement(urbRead, builder.getInt32(i)); |
644 | vec = builder.CreateInsertElement(vec, vecElement, builder.getInt32(i)); | 665 | vec = builder.CreateInsertElement(vec, vecElement, builder.getInt32(i)); |
645 | diff --git a/IGC/Compiler/CISACodeGen/LdShrink.cpp b/IGC/Compiler/CISACodeGen/LdShrink.cpp | 666 | diff --git a/IGC/Compiler/CISACodeGen/LdShrink.cpp b/IGC/Compiler/CISACodeGen/LdShrink.cpp |
646 | index d89fbd0e..2df06df2 100644 | 667 | index 91226eeb2..f7099d6fb 100644 |
647 | --- a/IGC/Compiler/CISACodeGen/LdShrink.cpp | 668 | --- a/IGC/Compiler/CISACodeGen/LdShrink.cpp |
648 | +++ b/IGC/Compiler/CISACodeGen/LdShrink.cpp | 669 | +++ b/IGC/Compiler/CISACodeGen/LdShrink.cpp |
649 | @@ -30,6 +30,7 @@ IN THE SOFTWARE. | 670 | @@ -14,6 +14,7 @@ SPDX-License-Identifier: MIT |
650 | #include <llvm/Support/Debug.h> | 671 | #include <llvm/Support/Debug.h> |
651 | #include <llvm/Support/MathExtras.h> | 672 | #include <llvm/Support/MathExtras.h> |
652 | #include <llvm/Support/raw_ostream.h> | 673 | #include <llvm/Support/raw_ostream.h> |
@@ -654,7 +675,7 @@ index d89fbd0e..2df06df2 100644 | |||
654 | #include "common/LLVMWarningsPop.hpp" | 675 | #include "common/LLVMWarningsPop.hpp" |
655 | #include "Compiler/CISACodeGen/ShaderCodeGen.hpp" | 676 | #include "Compiler/CISACodeGen/ShaderCodeGen.hpp" |
656 | #include "Compiler/IGCPassSupport.h" | 677 | #include "Compiler/IGCPassSupport.h" |
657 | @@ -79,7 +80,7 @@ IGC_INITIALIZE_PASS_BEGIN(LdShrink, PASS_FLAG, PASS_DESC, PASS_CFG_ONLY, PASS_AN | 678 | @@ -63,7 +64,7 @@ IGC_INITIALIZE_PASS_BEGIN(LdShrink, PASS_FLAG, PASS_DESC, PASS_CFG_ONLY, PASS_AN |
658 | IGC_INITIALIZE_PASS_END(LdShrink, PASS_FLAG, PASS_DESC, PASS_CFG_ONLY, PASS_ANALYSIS) | 679 | IGC_INITIALIZE_PASS_END(LdShrink, PASS_FLAG, PASS_DESC, PASS_CFG_ONLY, PASS_ANALYSIS) |
659 | 680 | ||
660 | unsigned LdShrink::getExtractIndexMask(LoadInst* LI) const { | 681 | unsigned LdShrink::getExtractIndexMask(LoadInst* LI) const { |
@@ -663,7 +684,7 @@ index d89fbd0e..2df06df2 100644 | |||
663 | // Skip non-vector loads. | 684 | // Skip non-vector loads. |
664 | if (!VTy) | 685 | if (!VTy) |
665 | return 0; | 686 | return 0; |
666 | @@ -91,9 +92,12 @@ unsigned LdShrink::getExtractIndexMask(LoadInst* LI) const { | 687 | @@ -75,9 +76,12 @@ unsigned LdShrink::getExtractIndexMask(LoadInst* LI) const { |
667 | Type* Ty = VTy->getScalarType(); | 688 | Type* Ty = VTy->getScalarType(); |
668 | // Skip non-BYTE addressable data types. So far, check integer types | 689 | // Skip non-BYTE addressable data types. So far, check integer types |
669 | // only. | 690 | // only. |
@@ -679,10 +700,10 @@ index d89fbd0e..2df06df2 100644 | |||
679 | unsigned Mask = 0; // Maxmimally 32 elements. | 700 | unsigned Mask = 0; // Maxmimally 32 elements. |
680 | 701 | ||
681 | diff --git a/IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.cpp b/IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.cpp | 702 | diff --git a/IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.cpp b/IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.cpp |
682 | index c12b20d5..5245117d 100644 | 703 | index ffd4bdc41..15d7f6d6d 100644 |
683 | --- a/IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.cpp | 704 | --- a/IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.cpp |
684 | +++ b/IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.cpp | 705 | +++ b/IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.cpp |
685 | @@ -634,7 +634,7 @@ void TransposeHelper::handleGEPInst( | 706 | @@ -603,7 +603,7 @@ void TransposeHelper::handleGEPInst( |
686 | } | 707 | } |
687 | else | 708 | else |
688 | { | 709 | { |
@@ -691,7 +712,7 @@ index c12b20d5..5245117d 100644 | |||
691 | } | 712 | } |
692 | T = cast<VectorType>(T)->getElementType(); | 713 | T = cast<VectorType>(T)->getElementType(); |
693 | } | 714 | } |
694 | @@ -656,7 +656,7 @@ void TransposeHelper::handleGEPInst( | 715 | @@ -625,7 +625,7 @@ void TransposeHelper::handleGEPInst( |
695 | } | 716 | } |
696 | else if (T->isVectorTy()) | 717 | else if (T->isVectorTy()) |
697 | { | 718 | { |
@@ -700,7 +721,7 @@ index c12b20d5..5245117d 100644 | |||
700 | T = cast<VectorType>(T)->getElementType(); | 721 | T = cast<VectorType>(T)->getElementType(); |
701 | } | 722 | } |
702 | else | 723 | else |
703 | @@ -716,7 +716,7 @@ void TransposeHelperPromote::handleLoadInst( | 724 | @@ -685,7 +685,7 @@ void TransposeHelperPromote::handleLoadInst( |
704 | IRBuilder<> IRB(pLoad); | 725 | IRBuilder<> IRB(pLoad); |
705 | IGC_ASSERT(nullptr != pLoad->getType()); | 726 | IGC_ASSERT(nullptr != pLoad->getType()); |
706 | unsigned N = pLoad->getType()->isVectorTy() | 727 | unsigned N = pLoad->getType()->isVectorTy() |
@@ -709,7 +730,7 @@ index c12b20d5..5245117d 100644 | |||
709 | : 1; | 730 | : 1; |
710 | Value* Val = loadEltsFromVecAlloca(N, pVecAlloca, pScalarizedIdx, IRB, pLoad->getType()->getScalarType()); | 731 | Value* Val = loadEltsFromVecAlloca(N, pVecAlloca, pScalarizedIdx, IRB, pLoad->getType()->getScalarType()); |
711 | pLoad->replaceAllUsesWith(Val); | 732 | pLoad->replaceAllUsesWith(Val); |
712 | @@ -748,7 +748,7 @@ void TransposeHelperPromote::handleStoreInst( | 733 | @@ -717,7 +717,7 @@ void TransposeHelperPromote::handleStoreInst( |
713 | // %v1 = extractelement <2 x float> %v, i32 1 | 734 | // %v1 = extractelement <2 x float> %v, i32 1 |
714 | // %w1 = insertelement <32 x float> %w0, float %v1, i32 %idx+1 | 735 | // %w1 = insertelement <32 x float> %w0, float %v1, i32 %idx+1 |
715 | // store <32 x float> %w1, <32 x float>* %ptr1 | 736 | // store <32 x float> %w1, <32 x float>* %ptr1 |
@@ -719,10 +740,10 @@ index c12b20d5..5245117d 100644 | |||
719 | Value* VectorIdx = ConstantInt::get(pScalarizedIdx->getType(), i); | 740 | Value* VectorIdx = ConstantInt::get(pScalarizedIdx->getType(), i); |
720 | auto Val = IRB.CreateExtractElement(pStoreVal, VectorIdx); | 741 | auto Val = IRB.CreateExtractElement(pStoreVal, VectorIdx); |
721 | diff --git a/IGC/Compiler/CISACodeGen/MemOpt.cpp b/IGC/Compiler/CISACodeGen/MemOpt.cpp | 742 | diff --git a/IGC/Compiler/CISACodeGen/MemOpt.cpp b/IGC/Compiler/CISACodeGen/MemOpt.cpp |
722 | index e4279cca..aa788b11 100644 | 743 | index 32f24ba1c..c85eb1238 100644 |
723 | --- a/IGC/Compiler/CISACodeGen/MemOpt.cpp | 744 | --- a/IGC/Compiler/CISACodeGen/MemOpt.cpp |
724 | +++ b/IGC/Compiler/CISACodeGen/MemOpt.cpp | 745 | +++ b/IGC/Compiler/CISACodeGen/MemOpt.cpp |
725 | @@ -125,7 +125,7 @@ namespace { | 746 | @@ -108,7 +108,7 @@ namespace { |
726 | MemRefListTy& MemRefs, TrivialMemRefListTy& ToOpt); | 747 | MemRefListTy& MemRefs, TrivialMemRefListTy& ToOpt); |
727 | 748 | ||
728 | unsigned getNumElements(Type* Ty) const { | 749 | unsigned getNumElements(Type* Ty) const { |
@@ -731,8 +752,8 @@ index e4279cca..aa788b11 100644 | |||
731 | } | 752 | } |
732 | 753 | ||
733 | MemoryLocation getLocation(Instruction* I) const { | 754 | MemoryLocation getLocation(Instruction* I) const { |
734 | @@ -797,7 +797,7 @@ bool MemOpt::mergeLoad(LoadInst* LeadingLoad, | 755 | @@ -922,7 +922,7 @@ bool MemOpt::mergeLoad(LoadInst* LeadingLoad, |
735 | Pos = unsigned((std::get<1>(I) - FirstOffset) / LdScalarSize); | 756 | Pos = unsigned((std::get<1>(I) - FirstOffset) / LdScalarSize); |
736 | 757 | ||
737 | if (Ty->isVectorTy()) { | 758 | if (Ty->isVectorTy()) { |
738 | - if (Pos + cast<VectorType>(Ty)->getNumElements() > NumElts) { | 759 | - if (Pos + cast<VectorType>(Ty)->getNumElements() > NumElts) { |
@@ -740,7 +761,7 @@ index e4279cca..aa788b11 100644 | |||
740 | // This implies we're trying to extract an element from our new load | 761 | // This implies we're trying to extract an element from our new load |
741 | // with an index > the size of the new load. If this happens, | 762 | // with an index > the size of the new load. If this happens, |
742 | // we'll generate correct code if it does since we don't remove the | 763 | // we'll generate correct code if it does since we don't remove the |
743 | @@ -805,7 +805,7 @@ bool MemOpt::mergeLoad(LoadInst* LeadingLoad, | 764 | @@ -930,7 +930,7 @@ bool MemOpt::mergeLoad(LoadInst* LeadingLoad, |
744 | continue; | 765 | continue; |
745 | } | 766 | } |
746 | Value* Val = UndefValue::get(Ty); | 767 | Value* Val = UndefValue::get(Ty); |
@@ -749,7 +770,7 @@ index e4279cca..aa788b11 100644 | |||
749 | Value* Ex = Builder.CreateExtractElement(NewLoad, Builder.getInt32(Pos + i)); | 770 | Value* Ex = Builder.CreateExtractElement(NewLoad, Builder.getInt32(Pos + i)); |
750 | Ex = createBitOrPointerCast(Ex, ScalarTy, Builder); | 771 | Ex = createBitOrPointerCast(Ex, ScalarTy, Builder); |
751 | Val = Builder.CreateInsertElement(Val, Ex, Builder.getInt32(i)); | 772 | Val = Builder.CreateInsertElement(Val, Ex, Builder.getInt32(i)); |
752 | @@ -1083,7 +1083,7 @@ bool MemOpt::mergeStore(StoreInst* LeadingStore, | 773 | @@ -1312,7 +1312,7 @@ bool MemOpt::mergeStore(StoreInst* LeadingStore, |
753 | IGC_ASSERT(hasSameSize(ScalarTy, LeadingStoreScalarType)); | 774 | IGC_ASSERT(hasSameSize(ScalarTy, LeadingStoreScalarType)); |
754 | 775 | ||
755 | if (Ty->isVectorTy()) { | 776 | if (Ty->isVectorTy()) { |
@@ -759,10 +780,10 @@ index e4279cca..aa788b11 100644 | |||
759 | Ex = createBitOrPointerCast(Ex, LeadingStoreScalarType, Builder); | 780 | Ex = createBitOrPointerCast(Ex, LeadingStoreScalarType, Builder); |
760 | NewStoreVal = Builder.CreateInsertElement(NewStoreVal, Ex, | 781 | NewStoreVal = Builder.CreateInsertElement(NewStoreVal, Ex, |
761 | diff --git a/IGC/Compiler/CISACodeGen/MemOpt2.cpp b/IGC/Compiler/CISACodeGen/MemOpt2.cpp | 782 | diff --git a/IGC/Compiler/CISACodeGen/MemOpt2.cpp b/IGC/Compiler/CISACodeGen/MemOpt2.cpp |
762 | index 4b2628d2..20115c73 100644 | 783 | index 989727107..96e1ec303 100644 |
763 | --- a/IGC/Compiler/CISACodeGen/MemOpt2.cpp | 784 | --- a/IGC/Compiler/CISACodeGen/MemOpt2.cpp |
764 | +++ b/IGC/Compiler/CISACodeGen/MemOpt2.cpp | 785 | +++ b/IGC/Compiler/CISACodeGen/MemOpt2.cpp |
765 | @@ -32,6 +32,7 @@ IN THE SOFTWARE. | 786 | @@ -16,6 +16,7 @@ SPDX-License-Identifier: MIT |
766 | #include <llvm/Support/Debug.h> | 787 | #include <llvm/Support/Debug.h> |
767 | #include <llvm/Support/raw_ostream.h> | 788 | #include <llvm/Support/raw_ostream.h> |
768 | #include <llvm/Transforms/Utils/Local.h> | 789 | #include <llvm/Transforms/Utils/Local.h> |
@@ -770,7 +791,7 @@ index 4b2628d2..20115c73 100644 | |||
770 | #include "common/LLVMWarningsPop.hpp" | 791 | #include "common/LLVMWarningsPop.hpp" |
771 | #include "GenISAIntrinsics/GenIntrinsics.h" | 792 | #include "GenISAIntrinsics/GenIntrinsics.h" |
772 | 793 | ||
773 | @@ -124,7 +125,7 @@ namespace { | 794 | @@ -108,7 +109,7 @@ namespace { |
774 | if (!Ty->isSingleValueType()) | 795 | if (!Ty->isSingleValueType()) |
775 | return UINT_MAX; | 796 | return UINT_MAX; |
776 | // Simply return 1 so far for scalar types. | 797 | // Simply return 1 so far for scalar types. |
@@ -779,7 +800,7 @@ index 4b2628d2..20115c73 100644 | |||
779 | if (!VecTy) | 800 | if (!VecTy) |
780 | return 1; | 801 | return 1; |
781 | // Check how that vector is used. | 802 | // Check how that vector is used. |
782 | @@ -153,7 +154,7 @@ namespace { | 803 | @@ -137,7 +138,7 @@ namespace { |
783 | return UINT_MAX; | 804 | return UINT_MAX; |
784 | unsigned EltByte = (Ty->getScalarSizeInBits() + 7) / 8; | 805 | unsigned EltByte = (Ty->getScalarSizeInBits() + 7) / 8; |
785 | // Simply return 1 so far for scalar types. | 806 | // Simply return 1 so far for scalar types. |
@@ -789,10 +810,10 @@ index 4b2628d2..20115c73 100644 | |||
789 | return EltByte; | 810 | return EltByte; |
790 | // Check how that vector is used. | 811 | // Check how that vector is used. |
791 | diff --git a/IGC/Compiler/CISACodeGen/OpenCLKernelCodeGen.cpp b/IGC/Compiler/CISACodeGen/OpenCLKernelCodeGen.cpp | 812 | diff --git a/IGC/Compiler/CISACodeGen/OpenCLKernelCodeGen.cpp b/IGC/Compiler/CISACodeGen/OpenCLKernelCodeGen.cpp |
792 | index 5baff64f..032d127a 100644 | 813 | index 01272b03b..8dc187ebf 100644 |
793 | --- a/IGC/Compiler/CISACodeGen/OpenCLKernelCodeGen.cpp | 814 | --- a/IGC/Compiler/CISACodeGen/OpenCLKernelCodeGen.cpp |
794 | +++ b/IGC/Compiler/CISACodeGen/OpenCLKernelCodeGen.cpp | 815 | +++ b/IGC/Compiler/CISACodeGen/OpenCLKernelCodeGen.cpp |
795 | @@ -410,7 +410,7 @@ namespace IGC | 816 | @@ -409,7 +409,7 @@ namespace IGC |
796 | unsigned int numElements = 1; | 817 | unsigned int numElements = 1; |
797 | if (baseType->isVectorTy()) | 818 | if (baseType->isVectorTy()) |
798 | { | 819 | { |
@@ -802,10 +823,19 @@ index 5baff64f..032d127a 100644 | |||
802 | } | 823 | } |
803 | 824 | ||
804 | diff --git a/IGC/Compiler/CISACodeGen/PatternMatchPass.cpp b/IGC/Compiler/CISACodeGen/PatternMatchPass.cpp | 825 | diff --git a/IGC/Compiler/CISACodeGen/PatternMatchPass.cpp b/IGC/Compiler/CISACodeGen/PatternMatchPass.cpp |
805 | index 356f1141..52396304 100644 | 826 | index cd048136a..51632e3d9 100644 |
806 | --- a/IGC/Compiler/CISACodeGen/PatternMatchPass.cpp | 827 | --- a/IGC/Compiler/CISACodeGen/PatternMatchPass.cpp |
807 | +++ b/IGC/Compiler/CISACodeGen/PatternMatchPass.cpp | 828 | +++ b/IGC/Compiler/CISACodeGen/PatternMatchPass.cpp |
808 | @@ -2916,8 +2916,8 @@ namespace IGC | 829 | @@ -2438,7 +2438,7 @@ namespace IGC |
830 | // Store3d supports only types equal or less than 128 bits. | ||
831 | if (auto* storeInst = dyn_cast<StoreInst>(&I)) | ||
832 | { | ||
833 | - llvm::VectorType* vectorToStore = dyn_cast<llvm::VectorType>(storeInst->getValueOperand()->getType()); | ||
834 | + IGCLLVM::FixedVectorType* vectorToStore = dyn_cast<IGCLLVM::FixedVectorType>(storeInst->getValueOperand()->getType()); | ||
835 | |||
836 | // If stored value is a vector of pointers, the size must be calculated manually, | ||
837 | // because getPrimitiveSizeInBits returns 0 for pointers. | ||
838 | @@ -3072,8 +3072,8 @@ namespace IGC | ||
809 | llvm::Type* srcTy = bTInst->getOperand(0)->getType(); | 839 | llvm::Type* srcTy = bTInst->getOperand(0)->getType(); |
810 | llvm::Type* dstTy = bTInst->getType(); | 840 | llvm::Type* dstTy = bTInst->getType(); |
811 | 841 | ||
@@ -817,10 +847,10 @@ index 356f1141..52396304 100644 | |||
817 | if (srcNElts < dstNElts && srcTy->getScalarSizeInBits() < 64) | 847 | if (srcNElts < dstNElts && srcTy->getScalarSizeInBits() < 64) |
818 | { | 848 | { |
819 | diff --git a/IGC/Compiler/CISACodeGen/PushAnalysis.cpp b/IGC/Compiler/CISACodeGen/PushAnalysis.cpp | 849 | diff --git a/IGC/Compiler/CISACodeGen/PushAnalysis.cpp b/IGC/Compiler/CISACodeGen/PushAnalysis.cpp |
820 | index ba8c2770..8da82db0 100644 | 850 | index 4f905cc9a..0ee5ffbc0 100644 |
821 | --- a/IGC/Compiler/CISACodeGen/PushAnalysis.cpp | 851 | --- a/IGC/Compiler/CISACodeGen/PushAnalysis.cpp |
822 | +++ b/IGC/Compiler/CISACodeGen/PushAnalysis.cpp | 852 | +++ b/IGC/Compiler/CISACodeGen/PushAnalysis.cpp |
823 | @@ -899,7 +899,7 @@ namespace IGC | 853 | @@ -987,7 +987,7 @@ namespace IGC |
824 | 854 | ||
825 | if (pTypeToPush->isVectorTy()) | 855 | if (pTypeToPush->isVectorTy()) |
826 | { | 856 | { |
@@ -829,7 +859,7 @@ index ba8c2770..8da82db0 100644 | |||
829 | pTypeToPush = cast<VectorType>(pTypeToPush)->getElementType(); | 859 | pTypeToPush = cast<VectorType>(pTypeToPush)->getElementType(); |
830 | llvm::Type* pVecTy = IGCLLVM::FixedVectorType::get(pTypeToPush, num_elms); | 860 | llvm::Type* pVecTy = IGCLLVM::FixedVectorType::get(pTypeToPush, num_elms); |
831 | pReplacedInst = llvm::UndefValue::get(pVecTy); | 861 | pReplacedInst = llvm::UndefValue::get(pVecTy); |
832 | @@ -1102,7 +1102,7 @@ namespace IGC | 862 | @@ -1178,7 +1178,7 @@ namespace IGC |
833 | } | 863 | } |
834 | 864 | ||
835 | unsigned num_elms = | 865 | unsigned num_elms = |
@@ -839,10 +869,10 @@ index ba8c2770..8da82db0 100644 | |||
839 | llvm::Value* replaceVector = nullptr; | 869 | llvm::Value* replaceVector = nullptr; |
840 | unsigned int numberChannelReplaced = 0; | 870 | unsigned int numberChannelReplaced = 0; |
841 | diff --git a/IGC/Compiler/CISACodeGen/RegisterEstimator.cpp b/IGC/Compiler/CISACodeGen/RegisterEstimator.cpp | 871 | diff --git a/IGC/Compiler/CISACodeGen/RegisterEstimator.cpp b/IGC/Compiler/CISACodeGen/RegisterEstimator.cpp |
842 | index 95b937e7..1c6544a4 100644 | 872 | index dee494d73..ef33489c3 100644 |
843 | --- a/IGC/Compiler/CISACodeGen/RegisterEstimator.cpp | 873 | --- a/IGC/Compiler/CISACodeGen/RegisterEstimator.cpp |
844 | +++ b/IGC/Compiler/CISACodeGen/RegisterEstimator.cpp | 874 | +++ b/IGC/Compiler/CISACodeGen/RegisterEstimator.cpp |
845 | @@ -41,6 +41,7 @@ IN THE SOFTWARE. | 875 | @@ -25,6 +25,7 @@ SPDX-License-Identifier: MIT |
846 | #include <llvm/IR/IRBuilder.h> | 876 | #include <llvm/IR/IRBuilder.h> |
847 | #include <llvm/IR/InstIterator.h> | 877 | #include <llvm/IR/InstIterator.h> |
848 | #include <llvm/Support/MathExtras.h> | 878 | #include <llvm/Support/MathExtras.h> |
@@ -850,7 +880,7 @@ index 95b937e7..1c6544a4 100644 | |||
850 | #include "common/LLVMWarningsPop.hpp" | 880 | #include "common/LLVMWarningsPop.hpp" |
851 | #include "Probe/Assertion.h" | 881 | #include "Probe/Assertion.h" |
852 | 882 | ||
853 | @@ -146,7 +147,7 @@ RegUse RegisterEstimator::estimateNumOfRegs(Value* V) const | 883 | @@ -130,7 +131,7 @@ RegUse RegisterEstimator::estimateNumOfRegs(Value* V) const |
854 | Type* Ty = V->getType(); | 884 | Type* Ty = V->getType(); |
855 | if (!Ty->isVoidTy()) | 885 | if (!Ty->isVoidTy()) |
856 | { | 886 | { |
@@ -860,10 +890,10 @@ index 95b937e7..1c6544a4 100644 | |||
860 | uint32_t nelts = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; | 890 | uint32_t nelts = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; |
861 | uint32_t eltBits = (uint32_t)m_DL->getTypeSizeInBits(eltTy); | 891 | uint32_t eltBits = (uint32_t)m_DL->getTypeSizeInBits(eltTy); |
862 | diff --git a/IGC/Compiler/CISACodeGen/RegisterPressureEstimate.hpp b/IGC/Compiler/CISACodeGen/RegisterPressureEstimate.hpp | 892 | diff --git a/IGC/Compiler/CISACodeGen/RegisterPressureEstimate.hpp b/IGC/Compiler/CISACodeGen/RegisterPressureEstimate.hpp |
863 | index beac138d..ceeafa48 100644 | 893 | index 6148817e1..80898be35 100644 |
864 | --- a/IGC/Compiler/CISACodeGen/RegisterPressureEstimate.hpp | 894 | --- a/IGC/Compiler/CISACodeGen/RegisterPressureEstimate.hpp |
865 | +++ b/IGC/Compiler/CISACodeGen/RegisterPressureEstimate.hpp | 895 | +++ b/IGC/Compiler/CISACodeGen/RegisterPressureEstimate.hpp |
866 | @@ -42,6 +42,7 @@ See LRCENSE.TXT for details. | 896 | @@ -24,6 +24,7 @@ See LICENSE.TXT for details. |
867 | #include "llvm/Pass.h" | 897 | #include "llvm/Pass.h" |
868 | #include <llvm/IR/InstVisitor.h> | 898 | #include <llvm/IR/InstVisitor.h> |
869 | #include "llvm/Analysis/LoopInfo.h" | 899 | #include "llvm/Analysis/LoopInfo.h" |
@@ -871,7 +901,7 @@ index beac138d..ceeafa48 100644 | |||
871 | #include "common/LLVMWarningsPop.hpp" | 901 | #include "common/LLVMWarningsPop.hpp" |
872 | #include "Compiler/IGCPassSupport.h" | 902 | #include "Compiler/IGCPassSupport.h" |
873 | #include "Compiler/CISACodeGen/WIAnalysis.hpp" | 903 | #include "Compiler/CISACodeGen/WIAnalysis.hpp" |
874 | @@ -240,7 +241,7 @@ namespace IGC | 904 | @@ -222,7 +223,7 @@ namespace IGC |
875 | auto Ty = V->getType(); | 905 | auto Ty = V->getType(); |
876 | if (Ty->isVoidTy()) | 906 | if (Ty->isVoidTy()) |
877 | return 0; | 907 | return 0; |
@@ -881,10 +911,10 @@ index beac138d..ceeafa48 100644 | |||
881 | uint32_t nelts = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; | 911 | uint32_t nelts = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; |
882 | uint32_t eltBits = (uint32_t)m_DL->getTypeSizeInBits(eltTy); | 912 | uint32_t eltBits = (uint32_t)m_DL->getTypeSizeInBits(eltTy); |
883 | diff --git a/IGC/Compiler/CISACodeGen/SLMConstProp.cpp b/IGC/Compiler/CISACodeGen/SLMConstProp.cpp | 913 | diff --git a/IGC/Compiler/CISACodeGen/SLMConstProp.cpp b/IGC/Compiler/CISACodeGen/SLMConstProp.cpp |
884 | index 962083f4..26a7b8be 100644 | 914 | index 84e481b7d..3acc4f305 100644 |
885 | --- a/IGC/Compiler/CISACodeGen/SLMConstProp.cpp | 915 | --- a/IGC/Compiler/CISACodeGen/SLMConstProp.cpp |
886 | +++ b/IGC/Compiler/CISACodeGen/SLMConstProp.cpp | 916 | +++ b/IGC/Compiler/CISACodeGen/SLMConstProp.cpp |
887 | @@ -39,6 +39,7 @@ IN THE SOFTWARE. | 917 | @@ -23,6 +23,7 @@ SPDX-License-Identifier: MIT |
888 | #include <llvm/IR/InstIterator.h> | 918 | #include <llvm/IR/InstIterator.h> |
889 | #include <llvm/Support/Debug.h> | 919 | #include <llvm/Support/Debug.h> |
890 | #include <llvm/Support/MathExtras.h> | 920 | #include <llvm/Support/MathExtras.h> |
@@ -892,7 +922,7 @@ index 962083f4..26a7b8be 100644 | |||
892 | #include "common/LLVMWarningsPop.hpp" | 922 | #include "common/LLVMWarningsPop.hpp" |
893 | #include <vector> | 923 | #include <vector> |
894 | #include "Probe/Assertion.h" | 924 | #include "Probe/Assertion.h" |
895 | @@ -629,7 +630,7 @@ bool SLMConstProp::isEqual(Constant* C0, Constant* C1) | 925 | @@ -613,7 +614,7 @@ bool SLMConstProp::isEqual(Constant* C0, Constant* C1) |
896 | 926 | ||
897 | bool SLMConstProp::isFloatType(Type* Ty) | 927 | bool SLMConstProp::isFloatType(Type* Ty) |
898 | { | 928 | { |
@@ -902,10 +932,10 @@ index 962083f4..26a7b8be 100644 | |||
902 | if (vTy->getNumElements() > 1) | 932 | if (vTy->getNumElements() > 1) |
903 | { | 933 | { |
904 | diff --git a/IGC/Compiler/CISACodeGen/ScalarizerCodeGen.cpp b/IGC/Compiler/CISACodeGen/ScalarizerCodeGen.cpp | 934 | diff --git a/IGC/Compiler/CISACodeGen/ScalarizerCodeGen.cpp b/IGC/Compiler/CISACodeGen/ScalarizerCodeGen.cpp |
905 | index ce816416..32ab79c1 100644 | 935 | index 33b711b94..0651d2c34 100644 |
906 | --- a/IGC/Compiler/CISACodeGen/ScalarizerCodeGen.cpp | 936 | --- a/IGC/Compiler/CISACodeGen/ScalarizerCodeGen.cpp |
907 | +++ b/IGC/Compiler/CISACodeGen/ScalarizerCodeGen.cpp | 937 | +++ b/IGC/Compiler/CISACodeGen/ScalarizerCodeGen.cpp |
908 | @@ -66,7 +66,7 @@ void ScalarizerCodeGen::visitBinaryOperator(llvm::BinaryOperator& I) | 938 | @@ -50,7 +50,7 @@ void ScalarizerCodeGen::visitBinaryOperator(llvm::BinaryOperator& I) |
909 | { | 939 | { |
910 | bool isNewTypeVector = false; | 940 | bool isNewTypeVector = false; |
911 | 941 | ||
@@ -914,7 +944,7 @@ index ce816416..32ab79c1 100644 | |||
914 | unsigned numElements = int_cast<unsigned>(instType->getNumElements()); | 944 | unsigned numElements = int_cast<unsigned>(instType->getNumElements()); |
915 | unsigned scalarSize = instType->getScalarSizeInBits(); | 945 | unsigned scalarSize = instType->getScalarSizeInBits(); |
916 | unsigned newScalarBits = numElements * scalarSize; | 946 | unsigned newScalarBits = numElements * scalarSize; |
917 | @@ -111,7 +111,7 @@ void ScalarizerCodeGen::visitBinaryOperator(llvm::BinaryOperator& I) | 947 | @@ -95,7 +95,7 @@ void ScalarizerCodeGen::visitBinaryOperator(llvm::BinaryOperator& I) |
918 | } | 948 | } |
919 | else | 949 | else |
920 | { | 950 | { |
@@ -924,10 +954,10 @@ index ce816416..32ab79c1 100644 | |||
924 | Value* ieLogicOp = UndefValue::get(newType); | 954 | Value* ieLogicOp = UndefValue::get(newType); |
925 | for (unsigned i = 0; i < newVecTypeNumEle; i++) | 955 | for (unsigned i = 0; i < newVecTypeNumEle; i++) |
926 | diff --git a/IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp b/IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp | 956 | diff --git a/IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp b/IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp |
927 | index d7baf806..e6fbdf23 100644 | 957 | index 14a68e0da..4f42a6770 100644 |
928 | --- a/IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp | 958 | --- a/IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp |
929 | +++ b/IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp | 959 | +++ b/IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp |
930 | @@ -589,12 +589,14 @@ static void AddLegalizationPasses(CodeGenContext& ctx, IGCPassManager& mpm, PSSi | 960 | @@ -639,12 +639,14 @@ static void AddLegalizationPasses(CodeGenContext& ctx, IGCPassManager& mpm, PSSi |
931 | { | 961 | { |
932 | mpm.add(createPruneUnusedArgumentsPass()); | 962 | mpm.add(createPruneUnusedArgumentsPass()); |
933 | 963 | ||
@@ -942,7 +972,7 @@ index d7baf806..e6fbdf23 100644 | |||
942 | mpm.add(createDeadCodeEliminationPass()); | 972 | mpm.add(createDeadCodeEliminationPass()); |
943 | mpm.add(createCFGSimplificationPass()); | 973 | mpm.add(createCFGSimplificationPass()); |
944 | } | 974 | } |
945 | @@ -1475,8 +1477,10 @@ void OptimizeIR(CodeGenContext* const pContext) | 975 | @@ -1549,8 +1551,10 @@ void OptimizeIR(CodeGenContext* const pContext) |
946 | // possible which potentially allows late stage code sinking of | 976 | // possible which potentially allows late stage code sinking of |
947 | // those calls by the instruction combiner. | 977 | // those calls by the instruction combiner. |
948 | mpm.add(createPostOrderFunctionAttrsLegacyPass()); | 978 | mpm.add(createPostOrderFunctionAttrsLegacyPass()); |
@@ -954,18 +984,21 @@ index d7baf806..e6fbdf23 100644 | |||
954 | 984 | ||
955 | // enable this only when Pooled EU is not supported | 985 | // enable this only when Pooled EU is not supported |
956 | diff --git a/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp b/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp | 986 | diff --git a/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp b/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp |
957 | index 87ba2751..c1f4a419 100644 | 987 | index e1d1a8e60..97dbc0e74 100644 |
958 | --- a/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp | 988 | --- a/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp |
959 | +++ b/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp | 989 | +++ b/IGC/Compiler/CISACodeGen/Simd32Profitability.cpp |
960 | @@ -30,6 +30,7 @@ IN THE SOFTWARE. | 990 | @@ -12,8 +12,10 @@ SPDX-License-Identifier: MIT |
991 | #include "Compiler/CISACodeGen/Platform.hpp" | ||
992 | #include "common/LLVMWarningsPush.hpp" | ||
961 | #include <llvmWrapper/IR/DerivedTypes.h> | 993 | #include <llvmWrapper/IR/DerivedTypes.h> |
994 | +#include <llvmWrapper/Transforms/Utils/LoopUtils.h> | ||
962 | #include <llvm/IR/InstIterator.h> | 995 | #include <llvm/IR/InstIterator.h> |
963 | #include <llvm/IR/Operator.h> | 996 | #include <llvm/IR/Operator.h> |
964 | +#include <llvmWrapper/IR/DerivedTypes.h> | 997 | +#include <llvmWrapper/IR/DerivedTypes.h> |
965 | #include "common/LLVMWarningsPop.hpp" | 998 | #include "common/LLVMWarningsPop.hpp" |
966 | #include "GenISAIntrinsics/GenIntrinsics.h" | 999 | #include "GenISAIntrinsics/GenIntrinsics.h" |
967 | #include "GenISAIntrinsics/GenIntrinsicInst.h" | 1000 | #include "GenISAIntrinsics/GenIntrinsicInst.h" |
968 | @@ -577,7 +578,7 @@ static bool isPayloadHeader(Value* V) { | 1001 | @@ -561,7 +563,7 @@ static bool isPayloadHeader(Value* V) { |
969 | Argument* Arg = dyn_cast<Argument>(V); | 1002 | Argument* Arg = dyn_cast<Argument>(V); |
970 | if (!Arg || !Arg->hasName()) | 1003 | if (!Arg || !Arg->hasName()) |
971 | return false; | 1004 | return false; |
@@ -974,7 +1007,7 @@ index 87ba2751..c1f4a419 100644 | |||
974 | if (!VTy || VTy->getNumElements() != 8 || | 1007 | if (!VTy || VTy->getNumElements() != 8 || |
975 | !VTy->getElementType()->isIntegerTy(32)) | 1008 | !VTy->getElementType()->isIntegerTy(32)) |
976 | return false; | 1009 | return false; |
977 | @@ -588,7 +589,7 @@ static bool isR0(Value* V) { | 1010 | @@ -572,7 +574,7 @@ static bool isR0(Value* V) { |
978 | Argument* Arg = dyn_cast<Argument>(V); | 1011 | Argument* Arg = dyn_cast<Argument>(V); |
979 | if (!Arg || !Arg->hasName()) | 1012 | if (!Arg || !Arg->hasName()) |
980 | return false; | 1013 | return false; |
@@ -983,7 +1016,7 @@ index 87ba2751..c1f4a419 100644 | |||
983 | if (!VTy || VTy->getNumElements() != 8 || | 1016 | if (!VTy || VTy->getNumElements() != 8 || |
984 | !VTy->getElementType()->isIntegerTy(32)) | 1017 | !VTy->getElementType()->isIntegerTy(32)) |
985 | return false; | 1018 | return false; |
986 | @@ -599,7 +600,7 @@ static bool isEnqueuedLocalSize(Value* V) { | 1019 | @@ -583,7 +585,7 @@ static bool isEnqueuedLocalSize(Value* V) { |
987 | Argument* Arg = dyn_cast<Argument>(V); | 1020 | Argument* Arg = dyn_cast<Argument>(V); |
988 | if (!Arg || !Arg->hasName()) | 1021 | if (!Arg || !Arg->hasName()) |
989 | return false; | 1022 | return false; |
@@ -992,23 +1025,20 @@ index 87ba2751..c1f4a419 100644 | |||
992 | if (!VTy || VTy->getNumElements() != 3 || | 1025 | if (!VTy || VTy->getNumElements() != 3 || |
993 | !VTy->getElementType()->isIntegerTy(32)) | 1026 | !VTy->getElementType()->isIntegerTy(32)) |
994 | return false; | 1027 | return false; |
995 | @@ -994,7 +995,11 @@ static bool hasLongStridedLdStInLoop(Function* F, LoopInfo* LI, WIAnalysis* WI) | 1028 | @@ -982,7 +984,7 @@ static bool hasLongStridedLdStInLoop(Function* F, LoopInfo* LI, WIAnalysis* WI) |
996 | // Collect innermost simple loop. | 1029 | // Collect innermost simple loop. |
997 | for (auto I = LI->begin(), E = LI->end(); I != E; ++I) { | 1030 | for (auto I = LI->begin(), E = LI->end(); I != E; ++I) { |
998 | auto L = *I; | 1031 | auto L = *I; |
999 | +#if LLVM_VERSION_MAJOR >= 12 | 1032 | - if (!L->empty()) |
1000 | + if (!L->isInnermost()) | 1033 | + if (!IGCLLVM::isInnermost(L)) |
1001 | +#else | ||
1002 | if (!L->empty()) | ||
1003 | +#endif | ||
1004 | continue; | 1034 | continue; |
1005 | if (L->getNumBlocks() != 2) | 1035 | if (L->getNumBlocks() != 2) |
1006 | continue; | 1036 | continue; |
1007 | diff --git a/IGC/Compiler/CISACodeGen/SimplifyConstant.cpp b/IGC/Compiler/CISACodeGen/SimplifyConstant.cpp | 1037 | diff --git a/IGC/Compiler/CISACodeGen/SimplifyConstant.cpp b/IGC/Compiler/CISACodeGen/SimplifyConstant.cpp |
1008 | index 73f472b5..0069c4b1 100644 | 1038 | index c42e8f373..a4f758a5a 100644 |
1009 | --- a/IGC/Compiler/CISACodeGen/SimplifyConstant.cpp | 1039 | --- a/IGC/Compiler/CISACodeGen/SimplifyConstant.cpp |
1010 | +++ b/IGC/Compiler/CISACodeGen/SimplifyConstant.cpp | 1040 | +++ b/IGC/Compiler/CISACodeGen/SimplifyConstant.cpp |
1011 | @@ -352,7 +352,7 @@ static unsigned getLegalVectorSize(unsigned N) { | 1041 | @@ -336,7 +336,7 @@ static unsigned getLegalVectorSize(unsigned N) { |
1012 | // Check vector size. We may demote the data type if all values can fit into | 1042 | // Check vector size. We may demote the data type if all values can fit into |
1013 | // smaller data type. | 1043 | // smaller data type. |
1014 | // | 1044 | // |
@@ -1017,7 +1047,7 @@ index 73f472b5..0069c4b1 100644 | |||
1017 | bool& IsSigned) { | 1047 | bool& IsSigned) { |
1018 | Constant* Init = GV->getInitializer(); | 1048 | Constant* Init = GV->getInitializer(); |
1019 | IGC_ASSERT(isa<ArrayType>(Init->getType())); | 1049 | IGC_ASSERT(isa<ArrayType>(Init->getType())); |
1020 | @@ -360,7 +360,7 @@ static bool checkSize(GlobalVariable* GV, VectorType*& DataType, | 1050 | @@ -344,7 +344,7 @@ static bool checkSize(GlobalVariable* GV, VectorType*& DataType, |
1021 | unsigned N = (unsigned)ArrayTy->getArrayNumElements(); | 1051 | unsigned N = (unsigned)ArrayTy->getArrayNumElements(); |
1022 | Type* BaseTy = ArrayTy->getArrayElementType(); | 1052 | Type* BaseTy = ArrayTy->getArrayElementType(); |
1023 | unsigned VectorSize = 1; | 1053 | unsigned VectorSize = 1; |
@@ -1026,7 +1056,7 @@ index 73f472b5..0069c4b1 100644 | |||
1026 | BaseTy = VT->getElementType(); | 1056 | BaseTy = VT->getElementType(); |
1027 | VectorSize = int_cast<unsigned>(VT->getNumElements()); | 1057 | VectorSize = int_cast<unsigned>(VT->getNumElements()); |
1028 | N *= VectorSize; | 1058 | N *= VectorSize; |
1029 | @@ -483,7 +483,7 @@ static Constant* getConstantVal(Type* VEltTy, Constant* V, bool IsSigned) { | 1059 | @@ -467,7 +467,7 @@ static Constant* getConstantVal(Type* VEltTy, Constant* V, bool IsSigned) { |
1030 | return ConstantInt::get(VEltTy, IVal, IsSigned); | 1060 | return ConstantInt::get(VEltTy, IVal, IsSigned); |
1031 | } | 1061 | } |
1032 | 1062 | ||
@@ -1035,7 +1065,7 @@ index 73f472b5..0069c4b1 100644 | |||
1035 | Function* F) { | 1065 | Function* F) { |
1036 | // Build the constant vector from constant array. | 1066 | // Build the constant vector from constant array. |
1037 | unsigned VS = int_cast<unsigned>(AllocaType->getNumElements()); | 1067 | unsigned VS = int_cast<unsigned>(AllocaType->getNumElements()); |
1038 | @@ -507,7 +507,7 @@ static void promote(GlobalVariable* GV, VectorType* AllocaType, bool IsSigned, | 1068 | @@ -491,7 +491,7 @@ static void promote(GlobalVariable* GV, VectorType* AllocaType, bool IsSigned, |
1039 | Constant* const Elt = CA->getAggregateElement(i); | 1069 | Constant* const Elt = CA->getAggregateElement(i); |
1040 | IGC_ASSERT_MESSAGE(nullptr != Elt, "Null AggregateElement"); | 1070 | IGC_ASSERT_MESSAGE(nullptr != Elt, "Null AggregateElement"); |
1041 | if (auto EltTy = dyn_cast<VectorType>(Elt->getType())) { | 1071 | if (auto EltTy = dyn_cast<VectorType>(Elt->getType())) { |
@@ -1044,7 +1074,7 @@ index 73f472b5..0069c4b1 100644 | |||
1044 | for (unsigned j = 0; j < VectorSize; ++j) { | 1074 | for (unsigned j = 0; j < VectorSize; ++j) { |
1045 | Constant* V = Elt->getAggregateElement(j); | 1075 | Constant* V = Elt->getAggregateElement(j); |
1046 | Vals[i * VectorSize + j] = getConstantVal(VEltTy, V, IsSigned); | 1076 | Vals[i * VectorSize + j] = getConstantVal(VEltTy, V, IsSigned); |
1047 | @@ -549,7 +549,7 @@ static void promote(GlobalVariable* GV, VectorType* AllocaType, bool IsSigned, | 1077 | @@ -533,7 +533,7 @@ static void promote(GlobalVariable* GV, VectorType* AllocaType, bool IsSigned, |
1048 | unsigned N = 1; | 1078 | unsigned N = 1; |
1049 | Value* Offset = Index; | 1079 | Value* Offset = Index; |
1050 | if (Ty->isVectorTy()) { | 1080 | if (Ty->isVectorTy()) { |
@@ -1053,7 +1083,7 @@ index 73f472b5..0069c4b1 100644 | |||
1053 | Offset = Builder.CreateMul(Offset, ConstantInt::get(Offset->getType(), N)); | 1083 | Offset = Builder.CreateMul(Offset, ConstantInt::get(Offset->getType(), N)); |
1054 | } | 1084 | } |
1055 | Value* Val = extractNElts(N, VectorData, Offset, Builder); | 1085 | Value* Val = extractNElts(N, VectorData, Offset, Builder); |
1056 | @@ -684,7 +684,7 @@ bool PromoteConstant::runOnFunction(Function& F) { | 1086 | @@ -668,7 +668,7 @@ bool PromoteConstant::runOnFunction(Function& F) { |
1057 | 1087 | ||
1058 | // If possible demote the data into smaller type. Uses of value will be | 1088 | // If possible demote the data into smaller type. Uses of value will be |
1059 | // promoted back with ZExt or SExt. | 1089 | // promoted back with ZExt or SExt. |
@@ -1063,10 +1093,10 @@ index 73f472b5..0069c4b1 100644 | |||
1063 | if (!checkSize(GV, AllocaType, IsSigned)) | 1093 | if (!checkSize(GV, AllocaType, IsSigned)) |
1064 | continue; | 1094 | continue; |
1065 | diff --git a/IGC/Compiler/CISACodeGen/TypeDemote.cpp b/IGC/Compiler/CISACodeGen/TypeDemote.cpp | 1095 | diff --git a/IGC/Compiler/CISACodeGen/TypeDemote.cpp b/IGC/Compiler/CISACodeGen/TypeDemote.cpp |
1066 | index eb7a7b64..b04a0f6e 100644 | 1096 | index cd8f2eff0..e539bd6ed 100644 |
1067 | --- a/IGC/Compiler/CISACodeGen/TypeDemote.cpp | 1097 | --- a/IGC/Compiler/CISACodeGen/TypeDemote.cpp |
1068 | +++ b/IGC/Compiler/CISACodeGen/TypeDemote.cpp | 1098 | +++ b/IGC/Compiler/CISACodeGen/TypeDemote.cpp |
1069 | @@ -32,6 +32,7 @@ IN THE SOFTWARE. | 1099 | @@ -16,6 +16,7 @@ SPDX-License-Identifier: MIT |
1070 | #include <llvm/ADT/PostOrderIterator.h> | 1100 | #include <llvm/ADT/PostOrderIterator.h> |
1071 | #include <llvm/IR/IRBuilder.h> | 1101 | #include <llvm/IR/IRBuilder.h> |
1072 | #include <llvm/Pass.h> | 1102 | #include <llvm/Pass.h> |
@@ -1074,7 +1104,7 @@ index eb7a7b64..b04a0f6e 100644 | |||
1074 | #include "common/LLVMWarningsPop.hpp" | 1104 | #include "common/LLVMWarningsPop.hpp" |
1075 | #include "GenISAIntrinsics/GenIntrinsics.h" | 1105 | #include "GenISAIntrinsics/GenIntrinsics.h" |
1076 | #include "Probe/Assertion.h" | 1106 | #include "Probe/Assertion.h" |
1077 | @@ -288,7 +289,7 @@ bool TypeDemote::demoteOnBasicBlock(BasicBlock* BB) const { | 1107 | @@ -272,7 +273,7 @@ bool TypeDemote::demoteOnBasicBlock(BasicBlock* BB) const { |
1078 | CastInst* CI = dyn_cast<CastInst>(Index); | 1108 | CastInst* CI = dyn_cast<CastInst>(Index); |
1079 | if (CI && (CI->getOpcode() == Instruction::ZExt || | 1109 | if (CI && (CI->getOpcode() == Instruction::ZExt || |
1080 | CI->getOpcode() == Instruction::SExt)) { | 1110 | CI->getOpcode() == Instruction::SExt)) { |
@@ -1084,10 +1114,10 @@ index eb7a7b64..b04a0f6e 100644 | |||
1084 | unsigned Bound = (N < 32) ? (1U << N) : UINT32_MAX; | 1114 | unsigned Bound = (N < 32) ? (1U << N) : UINT32_MAX; |
1085 | if (VS <= Bound) { | 1115 | if (VS <= Bound) { |
1086 | diff --git a/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.cpp b/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.cpp | 1116 | diff --git a/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.cpp b/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.cpp |
1087 | index 1cac7b12..12c4734e 100644 | 1117 | index 666eb62e3..8b44cfef3 100644 |
1088 | --- a/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.cpp | 1118 | --- a/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.cpp |
1089 | +++ b/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.cpp | 1119 | +++ b/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.cpp |
1090 | @@ -28,6 +28,7 @@ IN THE SOFTWARE. | 1120 | @@ -12,6 +12,7 @@ SPDX-License-Identifier: MIT |
1091 | #include "Compiler/CodeGenPublic.h" | 1121 | #include "Compiler/CodeGenPublic.h" |
1092 | #include "common/LLVMWarningsPush.hpp" | 1122 | #include "common/LLVMWarningsPush.hpp" |
1093 | #include <llvm/Support/Debug.h> | 1123 | #include <llvm/Support/Debug.h> |
@@ -1095,7 +1125,7 @@ index 1cac7b12..12c4734e 100644 | |||
1095 | #include "common/LLVMWarningsPop.hpp" | 1125 | #include "common/LLVMWarningsPop.hpp" |
1096 | #include <algorithm> | 1126 | #include <algorithm> |
1097 | #include "Probe/Assertion.h" | 1127 | #include "Probe/Assertion.h" |
1098 | @@ -40,13 +41,13 @@ namespace | 1128 | @@ -24,13 +25,13 @@ namespace |
1099 | // If V is scalar, return 1. | 1129 | // If V is scalar, return 1. |
1100 | // if V is vector, return the number of elements. | 1130 | // if V is vector, return the number of elements. |
1101 | inline int getNumElts(Value* V) { | 1131 | inline int getNumElts(Value* V) { |
@@ -1112,10 +1142,10 @@ index 1cac7b12..12c4734e 100644 | |||
1112 | } | 1142 | } |
1113 | } | 1143 | } |
1114 | diff --git a/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.hpp b/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.hpp | 1144 | diff --git a/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.hpp b/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.hpp |
1115 | index 56dac17e..42081dca 100644 | 1145 | index ca523f239..d16d286fa 100644 |
1116 | --- a/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.hpp | 1146 | --- a/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.hpp |
1117 | +++ b/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.hpp | 1147 | +++ b/IGC/Compiler/CISACodeGen/VariableReuseAnalysis.hpp |
1118 | @@ -37,6 +37,7 @@ IN THE SOFTWARE. | 1148 | @@ -21,6 +21,7 @@ SPDX-License-Identifier: MIT |
1119 | #include <llvm/IR/InstIterator.h> | 1149 | #include <llvm/IR/InstIterator.h> |
1120 | #include <llvm/IR/InstVisitor.h> | 1150 | #include <llvm/IR/InstVisitor.h> |
1121 | #include "llvm/Pass.h" | 1151 | #include "llvm/Pass.h" |
@@ -1123,7 +1153,7 @@ index 56dac17e..42081dca 100644 | |||
1123 | #include "llvm/Support/raw_ostream.h" | 1153 | #include "llvm/Support/raw_ostream.h" |
1124 | #include "common/LLVMWarningsPop.hpp" | 1154 | #include "common/LLVMWarningsPop.hpp" |
1125 | #include "Compiler/CISACodeGen/RegisterEstimator.hpp" | 1155 | #include "Compiler/CISACodeGen/RegisterEstimator.hpp" |
1126 | @@ -79,7 +80,7 @@ namespace IGC { | 1156 | @@ -63,7 +64,7 @@ namespace IGC { |
1127 | SSubVecDesc(llvm::Value* V) | 1157 | SSubVecDesc(llvm::Value* V) |
1128 | : Aliaser(V), BaseVector(V), StartElementOffset(0) | 1158 | : Aliaser(V), BaseVector(V), StartElementOffset(0) |
1129 | { | 1159 | { |
@@ -1133,10 +1163,10 @@ index 56dac17e..42081dca 100644 | |||
1133 | } | 1163 | } |
1134 | 1164 | ||
1135 | diff --git a/IGC/Compiler/CISACodeGen/VectorPreProcess.cpp b/IGC/Compiler/CISACodeGen/VectorPreProcess.cpp | 1165 | diff --git a/IGC/Compiler/CISACodeGen/VectorPreProcess.cpp b/IGC/Compiler/CISACodeGen/VectorPreProcess.cpp |
1136 | index 4c2b0644..4cd6a2db 100644 | 1166 | index 204b2f29a..2ea975555 100644 |
1137 | --- a/IGC/Compiler/CISACodeGen/VectorPreProcess.cpp | 1167 | --- a/IGC/Compiler/CISACodeGen/VectorPreProcess.cpp |
1138 | +++ b/IGC/Compiler/CISACodeGen/VectorPreProcess.cpp | 1168 | +++ b/IGC/Compiler/CISACodeGen/VectorPreProcess.cpp |
1139 | @@ -500,7 +500,7 @@ void VectorPreProcess::replaceAllVectorUsesWithScalars(Instruction* VI, ValVecto | 1169 | @@ -485,7 +485,7 @@ void VectorPreProcess::replaceAllVectorUsesWithScalars(Instruction* VI, ValVecto |
1140 | I = VI->getParent()->getFirstNonPHI(); | 1170 | I = VI->getParent()->getFirstNonPHI(); |
1141 | } | 1171 | } |
1142 | IRBuilder<> Builder(I); | 1172 | IRBuilder<> Builder(I); |
@@ -1145,7 +1175,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1145 | Value* newVec = UndefValue::get(VTy); | 1175 | Value* newVec = UndefValue::get(VTy); |
1146 | for (uint32_t i = 0, e = int_cast<uint32_t>(VTy->getNumElements()); i < e; ++i) | 1176 | for (uint32_t i = 0, e = int_cast<uint32_t>(VTy->getNumElements()); i < e; ++i) |
1147 | { | 1177 | { |
1148 | @@ -619,7 +619,7 @@ bool VectorPreProcess::splitStore( | 1178 | @@ -626,7 +626,7 @@ bool VectorPreProcess::splitStore( |
1149 | { | 1179 | { |
1150 | Instruction* SI = ASI.getInst(); | 1180 | Instruction* SI = ASI.getInst(); |
1151 | Value* StoredVal = ASI.getValueOperand(); | 1181 | Value* StoredVal = ASI.getValueOperand(); |
@@ -1154,7 +1184,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1154 | Type* ETy = VTy->getElementType(); | 1184 | Type* ETy = VTy->getElementType(); |
1155 | uint32_t nelts = int_cast<uint32_t>(VTy->getNumElements()); | 1185 | uint32_t nelts = int_cast<uint32_t>(VTy->getNumElements()); |
1156 | 1186 | ||
1157 | @@ -705,7 +705,7 @@ bool VectorPreProcess::splitStore( | 1187 | @@ -712,7 +712,7 @@ bool VectorPreProcess::splitStore( |
1158 | { | 1188 | { |
1159 | Type* Ty1 = splitInfo[i].first; | 1189 | Type* Ty1 = splitInfo[i].first; |
1160 | uint32_t len1 = splitInfo[i].second; | 1190 | uint32_t len1 = splitInfo[i].second; |
@@ -1163,7 +1193,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1163 | for (uint32_t j = 0; j < len1; ++j) | 1193 | for (uint32_t j = 0; j < len1; ++j) |
1164 | { | 1194 | { |
1165 | Value* subVec; | 1195 | Value* subVec; |
1166 | @@ -742,7 +742,7 @@ bool VectorPreProcess::splitStore( | 1196 | @@ -749,7 +749,7 @@ bool VectorPreProcess::splitStore( |
1167 | { | 1197 | { |
1168 | Type* Ty1 = splitInfo[i].first; | 1198 | Type* Ty1 = splitInfo[i].first; |
1169 | uint32_t len1 = splitInfo[i].second; | 1199 | uint32_t len1 = splitInfo[i].second; |
@@ -1172,7 +1202,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1172 | for (uint32_t j = 0; j < len1; ++j) | 1202 | for (uint32_t j = 0; j < len1; ++j) |
1173 | { | 1203 | { |
1174 | uint32_t vAlign = (uint32_t)MinAlign(Align, (uint32_t)eOffset * EBytes); | 1204 | uint32_t vAlign = (uint32_t)MinAlign(Align, (uint32_t)eOffset * EBytes); |
1175 | @@ -786,7 +786,7 @@ bool VectorPreProcess::splitLoad( | 1205 | @@ -793,7 +793,7 @@ bool VectorPreProcess::splitLoad( |
1176 | { | 1206 | { |
1177 | Instruction* LI = ALI.getInst(); | 1207 | Instruction* LI = ALI.getInst(); |
1178 | bool isLdRaw = isa<LdRawIntrinsic>(LI); | 1208 | bool isLdRaw = isa<LdRawIntrinsic>(LI); |
@@ -1181,7 +1211,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1181 | Type* ETy = VTy->getElementType(); | 1211 | Type* ETy = VTy->getElementType(); |
1182 | uint32_t nelts = int_cast<uint32_t>(VTy->getNumElements()); | 1212 | uint32_t nelts = int_cast<uint32_t>(VTy->getNumElements()); |
1183 | 1213 | ||
1184 | @@ -848,7 +848,7 @@ bool VectorPreProcess::splitLoad( | 1214 | @@ -855,7 +855,7 @@ bool VectorPreProcess::splitLoad( |
1185 | { | 1215 | { |
1186 | Type* Ty1 = splitInfo[i].first; | 1216 | Type* Ty1 = splitInfo[i].first; |
1187 | uint32_t len1 = splitInfo[i].second; | 1217 | uint32_t len1 = splitInfo[i].second; |
@@ -1190,7 +1220,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1190 | for (uint32_t j = 0; j < len1; ++j) | 1220 | for (uint32_t j = 0; j < len1; ++j) |
1191 | { | 1221 | { |
1192 | uint32_t vAlign = (uint32_t)MinAlign(Align, eOffset * EBytes); | 1222 | uint32_t vAlign = (uint32_t)MinAlign(Align, eOffset * EBytes); |
1193 | @@ -916,7 +916,7 @@ bool VectorPreProcess::splitLoadStore( | 1223 | @@ -923,7 +923,7 @@ bool VectorPreProcess::splitLoadStore( |
1194 | Optional<AbstractStoreInst> ASI = AbstractStoreInst::get(Inst); | 1224 | Optional<AbstractStoreInst> ASI = AbstractStoreInst::get(Inst); |
1195 | IGC_ASSERT_MESSAGE((ALI || ASI), "Inst should be either load or store"); | 1225 | IGC_ASSERT_MESSAGE((ALI || ASI), "Inst should be either load or store"); |
1196 | Type* Ty = ALI ? ALI->getInst()->getType() : ASI->getValueOperand()->getType(); | 1226 | Type* Ty = ALI ? ALI->getInst()->getType() : ASI->getValueOperand()->getType(); |
@@ -1199,7 +1229,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1199 | if (!VTy) | 1229 | if (!VTy) |
1200 | { | 1230 | { |
1201 | return false; | 1231 | return false; |
1202 | @@ -1147,7 +1147,7 @@ void VectorPreProcess::getOrGenScalarValues( | 1232 | @@ -1154,7 +1154,7 @@ void VectorPreProcess::getOrGenScalarValues( |
1203 | { | 1233 | { |
1204 | availBeforeInst = nullptr; | 1234 | availBeforeInst = nullptr; |
1205 | 1235 | ||
@@ -1208,7 +1238,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1208 | if (!VTy) | 1238 | if (!VTy) |
1209 | { | 1239 | { |
1210 | scalars[0] = VecVal; | 1240 | scalars[0] = VecVal; |
1211 | @@ -1298,7 +1298,7 @@ Instruction* VectorPreProcess::simplifyLoadStore(Instruction* Inst) | 1241 | @@ -1305,7 +1305,7 @@ Instruction* VectorPreProcess::simplifyLoadStore(Instruction* Inst) |
1212 | // | 1242 | // |
1213 | // TODO: further optimize this load into a message with channel masks | 1243 | // TODO: further optimize this load into a message with channel masks |
1214 | // for cases in which use indices are sparse like {0, 2}. | 1244 | // for cases in which use indices are sparse like {0, 2}. |
@@ -1217,7 +1247,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1217 | if (N == MaxIndex + 1) | 1247 | if (N == MaxIndex + 1) |
1218 | return Inst; | 1248 | return Inst; |
1219 | 1249 | ||
1220 | @@ -1355,7 +1355,7 @@ Instruction* VectorPreProcess::simplifyLoadStore(Instruction* Inst) | 1250 | @@ -1420,7 +1420,7 @@ Instruction* VectorPreProcess::simplifyLoadStore(Instruction* Inst) |
1221 | if (NBits < 32) | 1251 | if (NBits < 32) |
1222 | return Inst; | 1252 | return Inst; |
1223 | 1253 | ||
@@ -1226,7 +1256,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1226 | if (auto CV = dyn_cast<ConstantVector>(Val)) | 1256 | if (auto CV = dyn_cast<ConstantVector>(Val)) |
1227 | { | 1257 | { |
1228 | unsigned MaxIndex = 0; | 1258 | unsigned MaxIndex = 0; |
1229 | @@ -1524,7 +1524,7 @@ bool VectorPreProcess::runOnFunction(Function& F) | 1259 | @@ -1589,7 +1589,7 @@ bool VectorPreProcess::runOnFunction(Function& F) |
1230 | for (uint32_t j = 0; j < svals.size(); ++j) | 1260 | for (uint32_t j = 0; j < svals.size(); ++j) |
1231 | { | 1261 | { |
1232 | Type* Ty1 = svals[j]->getType(); | 1262 | Type* Ty1 = svals[j]->getType(); |
@@ -1235,7 +1265,7 @@ index 4c2b0644..4cd6a2db 100644 | |||
1235 | if (VTy1) { | 1265 | if (VTy1) { |
1236 | for (uint32_t k = 0; k < VTy1->getNumElements(); ++k) | 1266 | for (uint32_t k = 0; k < VTy1->getNumElements(); ++k) |
1237 | { | 1267 | { |
1238 | @@ -1573,7 +1573,7 @@ bool VectorPreProcess::runOnFunction(Function& F) | 1268 | @@ -1638,7 +1638,7 @@ bool VectorPreProcess::runOnFunction(Function& F) |
1239 | // If this is a 3-element vector load, remove it | 1269 | // If this is a 3-element vector load, remove it |
1240 | // from m_Vector3List as well. | 1270 | // from m_Vector3List as well. |
1241 | if (isAbstractLoadInst(tInst) && tInst->getType()->isVectorTy() && | 1271 | if (isAbstractLoadInst(tInst) && tInst->getType()->isVectorTy() && |
@@ -1245,10 +1275,10 @@ index 4c2b0644..4cd6a2db 100644 | |||
1245 | InstWorkVector::iterator | 1275 | InstWorkVector::iterator |
1246 | tI = m_Vector3List.begin(), | 1276 | tI = m_Vector3List.begin(), |
1247 | diff --git a/IGC/Compiler/CISACodeGen/VectorProcess.cpp b/IGC/Compiler/CISACodeGen/VectorProcess.cpp | 1277 | diff --git a/IGC/Compiler/CISACodeGen/VectorProcess.cpp b/IGC/Compiler/CISACodeGen/VectorProcess.cpp |
1248 | index 9f39cdc0..cfe45321 100644 | 1278 | index c30a07834..8046b2e65 100644 |
1249 | --- a/IGC/Compiler/CISACodeGen/VectorProcess.cpp | 1279 | --- a/IGC/Compiler/CISACodeGen/VectorProcess.cpp |
1250 | +++ b/IGC/Compiler/CISACodeGen/VectorProcess.cpp | 1280 | +++ b/IGC/Compiler/CISACodeGen/VectorProcess.cpp |
1251 | @@ -240,7 +240,7 @@ bool VectorProcess::reLayoutLoadStore(Instruction* Inst) | 1281 | @@ -224,7 +224,7 @@ bool VectorProcess::reLayoutLoadStore(Instruction* Inst) |
1252 | IGC_ASSERT(nullptr != Ptr); | 1282 | IGC_ASSERT(nullptr != Ptr); |
1253 | IGC_ASSERT(nullptr != Ty); | 1283 | IGC_ASSERT(nullptr != Ty); |
1254 | 1284 | ||
@@ -1257,7 +1287,7 @@ index 9f39cdc0..cfe45321 100644 | |||
1257 | 1287 | ||
1258 | // Treat a scalar as 1-element vector | 1288 | // Treat a scalar as 1-element vector |
1259 | uint32_t nelts = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; | 1289 | uint32_t nelts = VTy ? int_cast<uint32_t>(VTy->getNumElements()) : 1; |
1260 | @@ -654,7 +654,7 @@ void VectorMessage::getInfo(Type* Ty, uint32_t Align, bool useA32, | 1290 | @@ -638,7 +638,7 @@ void VectorMessage::getInfo(Type* Ty, uint32_t Align, bool useA32, |
1261 | VectorType* VTy = dyn_cast<VectorType>(Ty); | 1291 | VectorType* VTy = dyn_cast<VectorType>(Ty); |
1262 | Type* eTy = VTy ? cast<VectorType>(VTy)->getElementType() : Ty; | 1292 | Type* eTy = VTy ? cast<VectorType>(VTy)->getElementType() : Ty; |
1263 | unsigned eltSize = m_emitter->GetScalarTypeSizeInRegister(eTy); | 1293 | unsigned eltSize = m_emitter->GetScalarTypeSizeInRegister(eTy); |
@@ -1267,10 +1297,10 @@ index 9f39cdc0..cfe45321 100644 | |||
1267 | const unsigned TBytes = nElts * eltSize; | 1297 | const unsigned TBytes = nElts * eltSize; |
1268 | 1298 | ||
1269 | diff --git a/IGC/Compiler/CISACodeGen/helper.cpp b/IGC/Compiler/CISACodeGen/helper.cpp | 1299 | diff --git a/IGC/Compiler/CISACodeGen/helper.cpp b/IGC/Compiler/CISACodeGen/helper.cpp |
1270 | index 784affaa..e5507283 100644 | 1300 | index f3da862c6..288a69a79 100644 |
1271 | --- a/IGC/Compiler/CISACodeGen/helper.cpp | 1301 | --- a/IGC/Compiler/CISACodeGen/helper.cpp |
1272 | +++ b/IGC/Compiler/CISACodeGen/helper.cpp | 1302 | +++ b/IGC/Compiler/CISACodeGen/helper.cpp |
1273 | @@ -1634,7 +1634,7 @@ namespace IGC | 1303 | @@ -1744,7 +1744,7 @@ namespace IGC |
1274 | { | 1304 | { |
1275 | instList[i] = builder.CreateExtractElement(val, static_cast<uint64_t>(0)); | 1305 | instList[i] = builder.CreateExtractElement(val, static_cast<uint64_t>(0)); |
1276 | size_t iOld = i; | 1306 | size_t iOld = i; |
@@ -1279,7 +1309,7 @@ index 784affaa..e5507283 100644 | |||
1279 | { | 1309 | { |
1280 | instList.insert(instList.begin()+ iOld +j, builder.CreateExtractElement(val, j)); | 1310 | instList.insert(instList.begin()+ iOld +j, builder.CreateExtractElement(val, j)); |
1281 | i++; | 1311 | i++; |
1282 | @@ -1667,7 +1667,7 @@ namespace IGC | 1312 | @@ -1777,7 +1777,7 @@ namespace IGC |
1283 | } | 1313 | } |
1284 | break; | 1314 | break; |
1285 | case IGCLLVM::VectorTyID: | 1315 | case IGCLLVM::VectorTyID: |
@@ -1288,7 +1318,7 @@ index 784affaa..e5507283 100644 | |||
1288 | for (unsigned i = 0; i < num; i++) | 1318 | for (unsigned i = 0; i < num; i++) |
1289 | { | 1319 | { |
1290 | ScalarizeAggregateMembers(builder, builder.CreateExtractElement(val, i), instList); | 1320 | ScalarizeAggregateMembers(builder, builder.CreateExtractElement(val, i), instList); |
1291 | @@ -1707,7 +1707,7 @@ namespace IGC | 1321 | @@ -1817,7 +1817,7 @@ namespace IGC |
1292 | } | 1322 | } |
1293 | break; | 1323 | break; |
1294 | case IGCLLVM::VectorTyID: | 1324 | case IGCLLVM::VectorTyID: |
@@ -1297,7 +1327,7 @@ index 784affaa..e5507283 100644 | |||
1297 | for (unsigned i = 0; i < num; i++) | 1327 | for (unsigned i = 0; i < num; i++) |
1298 | { | 1328 | { |
1299 | indices.push_back(builder.getInt32(i)); | 1329 | indices.push_back(builder.getInt32(i)); |
1300 | @@ -1972,8 +1972,8 @@ namespace IGC | 1330 | @@ -2082,8 +2082,8 @@ namespace IGC |
1301 | return false; | 1331 | return false; |
1302 | } | 1332 | } |
1303 | 1333 | ||
@@ -1309,58 +1339,30 @@ index 784affaa..e5507283 100644 | |||
1309 | int s_nelts = sVTy ? (int)sVTy->getNumElements() : 1; | 1339 | int s_nelts = sVTy ? (int)sVTy->getNumElements() : 1; |
1310 | if (d_nelts != s_nelts) { | 1340 | if (d_nelts != s_nelts) { |
1311 | diff --git a/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp b/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp | 1341 | diff --git a/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp b/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp |
1312 | index cd7fc66e..adf992cb 100644 | 1342 | index 75089a77e..efb27ae4e 100644 |
1313 | --- a/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp | 1343 | --- a/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp |
1314 | +++ b/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp | 1344 | +++ b/IGC/Compiler/ConvertMSAAPayloadTo16Bit.cpp |
1315 | @@ -153,15 +153,24 @@ void ConvertMSAAPayloadTo16Bit::visitCallInst(CallInst& I) | 1345 | @@ -138,12 +138,11 @@ void ConvertMSAAPayloadTo16Bit::visitCallInst(CallInst& I) |
1316 | // In OGL there are uses of ldmcs other then ldms, using vec4float type. | 1346 | // There are uses of ldmcs other then ldms, using vector of int32 type. |
1317 | // Fix them to use newly created 16bit ldmcs. | 1347 | // Fix them to use newly created 16bit ldmcs. |
1318 | if (ldmcs->getType()->isVectorTy() && | 1348 | if (ldmcs->getType()->isVectorTy() && |
1319 | +#if LLVM_VERSION_MAJOR >= 12 | 1349 | - cast<VectorType>(ldmcs->getType())->getElementType() == m_builder->getInt32Ty()) |
1320 | + ldmcs->getType()->getScalarType()->isFloatTy()) | 1350 | + cast<IGCLLVM::FixedVectorType>(ldmcs->getType())->getElementType()->isFloatTy()) |
1321 | +#else | ||
1322 | ldmcs->getType()->getVectorElementType()->isFloatTy()) | ||
1323 | +#endif | ||
1324 | { | 1351 | { |
1325 | m_builder->SetInsertPoint(ldmcs); | 1352 | m_builder->SetInsertPoint(ldmcs); |
1326 | 1353 | - | |
1327 | +#if LLVM_VERSION_MAJOR >= 12 | 1354 | - uint32_t ldmcsNumOfElements = (uint32_t)cast<VectorType>(ldmcs->getType())->getNumElements(); |
1328 | + uint ldmcsNumOfElements = cast<IGCLLVM::FixedVectorType>(ldmcs->getType())->getNumElements(); | 1355 | - uint32_t newLdmcsNumOfElements = (uint32_t)cast<VectorType>(new_mcs_call->getType())->getNumElements(); |
1329 | + uint new_mcs_callNumOfElements = cast<IGCLLVM::FixedVectorType>(new_mcs_call->getType())->getNumElements(); | 1356 | + uint32_t ldmcsNumOfElements = (uint32_t)cast<IGCLLVM::FixedVectorType>(ldmcs->getType())->getNumElements(); |
1330 | +#else | 1357 | + uint32_t newLdmcsNumOfElements = (uint32_t)cast<IGCLLVM::FixedVectorType>(new_mcs_call->getType())->getNumElements(); |
1331 | uint ldmcsNumOfElements = ldmcs->getType()->getVectorNumElements(); | ||
1332 | uint new_mcs_callNumOfElements = new_mcs_call->getType()->getVectorNumElements(); | ||
1333 | +#endif | ||
1334 | 1358 | ||
1335 | // vec of 16bit ints to vec of 32bit ints | 1359 | // vec of 16bit ints to vec of 32bit ints |
1336 | - Type* new_mcs_callVecType = VectorType::get(m_builder->getInt32Ty(), new_mcs_callNumOfElements); | 1360 | Type* newLdmcsVecType = IGCLLVM::FixedVectorType::get(m_builder->getInt32Ty(), newLdmcsNumOfElements); |
1337 | + Type* new_mcs_callVecType = IGCLLVM::FixedVectorType::get(m_builder->getInt32Ty(), new_mcs_callNumOfElements); | ||
1338 | Value* ldmcsExtendedToInt32 = m_builder->CreateSExt(new_mcs_call, new_mcs_callVecType); | ||
1339 | |||
1340 | // if new ldmcs has fewer elements than ldmcs, extend vector | ||
1341 | @@ -175,7 +184,7 @@ void ConvertMSAAPayloadTo16Bit::visitCallInst(CallInst& I) | ||
1342 | } | ||
1343 | auto* pMask = ConstantDataVector::get(I.getContext(), maskVals); | ||
1344 | |||
1345 | - newLdmcsSizedVector = m_builder->CreateShuffleVector(ldmcsExtendedToInt32, UndefValue::get(VectorType::get(m_builder->getInt32Ty(), ldmcsNumOfElements)), pMask); | ||
1346 | + newLdmcsSizedVector = m_builder->CreateShuffleVector(ldmcsExtendedToInt32, UndefValue::get(IGCLLVM::FixedVectorType::get(m_builder->getInt32Ty(), ldmcsNumOfElements)), pMask); | ||
1347 | } | ||
1348 | else | ||
1349 | { | ||
1350 | @@ -183,7 +192,7 @@ void ConvertMSAAPayloadTo16Bit::visitCallInst(CallInst& I) | ||
1351 | } | ||
1352 | IGC_ASSERT(newLdmcsSizedVector); | ||
1353 | |||
1354 | - Type* ldmcsFloatVecType = VectorType::get(m_builder->getFloatTy(), ldmcsNumOfElements); | ||
1355 | + Type* ldmcsFloatVecType = IGCLLVM::FixedVectorType::get(m_builder->getFloatTy(), ldmcsNumOfElements); | ||
1356 | Value* ldmcsBitcastedToFloat = m_builder->CreateBitCast(ldmcsExtendedToInt32, ldmcsFloatVecType); | ||
1357 | ldmcs->replaceAllUsesWith(ldmcsBitcastedToFloat); | ||
1358 | } | ||
1359 | diff --git a/IGC/Compiler/CustomSafeOptPass.cpp b/IGC/Compiler/CustomSafeOptPass.cpp | 1361 | diff --git a/IGC/Compiler/CustomSafeOptPass.cpp b/IGC/Compiler/CustomSafeOptPass.cpp |
1360 | index 707bbfbe..72859a91 100644 | 1362 | index 3fe49f460..c65a466cc 100644 |
1361 | --- a/IGC/Compiler/CustomSafeOptPass.cpp | 1363 | --- a/IGC/Compiler/CustomSafeOptPass.cpp |
1362 | +++ b/IGC/Compiler/CustomSafeOptPass.cpp | 1364 | +++ b/IGC/Compiler/CustomSafeOptPass.cpp |
1363 | @@ -1372,7 +1372,7 @@ void IGC::CustomSafeOptPass::visitSampleBptr(llvm::SampleIntrinsic* sampleInst) | 1365 | @@ -1544,7 +1544,7 @@ void IGC::CustomSafeOptPass::visitSampleBptr(llvm::SampleIntrinsic* sampleInst) |
1364 | bool CustomSafeOptPass::isIdentityMatrix(ExtractElementInst& I) | 1366 | bool CustomSafeOptPass::isIdentityMatrix(ExtractElementInst& I) |
1365 | { | 1367 | { |
1366 | bool found = false; | 1368 | bool found = false; |
@@ -1369,7 +1371,7 @@ index 707bbfbe..72859a91 100644 | |||
1369 | auto extractTypeVecSize = (uint32_t)extractType->getNumElements(); | 1371 | auto extractTypeVecSize = (uint32_t)extractType->getNumElements(); |
1370 | if (extractTypeVecSize == 20 || | 1372 | if (extractTypeVecSize == 20 || |
1371 | extractTypeVecSize == 16) | 1373 | extractTypeVecSize == 16) |
1372 | @@ -1601,7 +1601,7 @@ void CustomSafeOptPass::visitExtractElementInst(ExtractElementInst& I) | 1374 | @@ -1773,7 +1773,7 @@ void CustomSafeOptPass::visitExtractElementInst(ExtractElementInst& I) |
1373 | int elOffset = (int)(bitShift / eltSize); | 1375 | int elOffset = (int)(bitShift / eltSize); |
1374 | elOffset = rightShift ? elOffset : -elOffset; | 1376 | elOffset = rightShift ? elOffset : -elOffset; |
1375 | unsigned int newIndex = (unsigned int)((int)cstIndex->getZExtValue() + elOffset); | 1377 | unsigned int newIndex = (unsigned int)((int)cstIndex->getZExtValue() + elOffset); |
@@ -1378,7 +1380,7 @@ index 707bbfbe..72859a91 100644 | |||
1378 | { | 1380 | { |
1379 | IRBuilder<> builder(&I); | 1381 | IRBuilder<> builder(&I); |
1380 | Value* newBitCast = builder.CreateBitCast(binOp->getOperand(0), vecType); | 1382 | Value* newBitCast = builder.CreateBitCast(binOp->getOperand(0), vecType); |
1381 | @@ -2001,7 +2001,7 @@ void GenSpecificPattern::createBitcastExtractInsertPattern(BinaryOperator& I, Va | 1383 | @@ -2173,7 +2173,7 @@ void GenSpecificPattern::createBitcastExtractInsertPattern(BinaryOperator& I, Va |
1382 | else if (auto IEIInst = dyn_cast<InsertElementInst>(Op)) | 1384 | else if (auto IEIInst = dyn_cast<InsertElementInst>(Op)) |
1383 | { | 1385 | { |
1384 | auto opType = IEIInst->getType(); | 1386 | auto opType = IEIInst->getType(); |
@@ -1387,7 +1389,7 @@ index 707bbfbe..72859a91 100644 | |||
1387 | { | 1389 | { |
1388 | elem = IEIInst->getOperand(1); | 1390 | elem = IEIInst->getOperand(1); |
1389 | } | 1391 | } |
1390 | @@ -2064,7 +2064,7 @@ void GenSpecificPattern::visitBinaryOperator(BinaryOperator& I) | 1392 | @@ -2236,7 +2236,7 @@ void GenSpecificPattern::visitBinaryOperator(BinaryOperator& I) |
1391 | else if (match(&I, pattern2) && AndOp2->getType()->isIntegerTy(64)) | 1393 | else if (match(&I, pattern2) && AndOp2->getType()->isIntegerTy(64)) |
1392 | { | 1394 | { |
1393 | ConstantVector* cVec = dyn_cast<ConstantVector>(VecOp); | 1395 | ConstantVector* cVec = dyn_cast<ConstantVector>(VecOp); |
@@ -1396,7 +1398,7 @@ index 707bbfbe..72859a91 100644 | |||
1396 | if (cVec && vector_type && | 1398 | if (cVec && vector_type && |
1397 | isa<ConstantInt>(cVec->getOperand(0)) && | 1399 | isa<ConstantInt>(cVec->getOperand(0)) && |
1398 | cast<ConstantInt>(cVec->getOperand(0))->isZero() && | 1400 | cast<ConstantInt>(cVec->getOperand(0))->isZero() && |
1399 | @@ -2210,7 +2210,7 @@ void GenSpecificPattern::visitBinaryOperator(BinaryOperator& I) | 1401 | @@ -2382,7 +2382,7 @@ void GenSpecificPattern::visitBinaryOperator(BinaryOperator& I) |
1400 | BitCastInst* opBC = cast<BitCastInst>(op); | 1402 | BitCastInst* opBC = cast<BitCastInst>(op); |
1401 | 1403 | ||
1402 | auto opType = opBC->getType(); | 1404 | auto opType = opBC->getType(); |
@@ -1405,7 +1407,7 @@ index 707bbfbe..72859a91 100644 | |||
1405 | return nullptr; | 1407 | return nullptr; |
1406 | 1408 | ||
1407 | if (opBC->getSrcTy()->isDoubleTy()) | 1409 | if (opBC->getSrcTy()->isDoubleTy()) |
1408 | @@ -2630,8 +2630,8 @@ void GenSpecificPattern::visitBitCastInst(BitCastInst& I) | 1410 | @@ -2808,8 +2808,8 @@ void GenSpecificPattern::visitBitCastInst(BitCastInst& I) |
1409 | if (zExtInst->getOperand(0)->getType()->isIntegerTy(32) && | 1411 | if (zExtInst->getOperand(0)->getType()->isIntegerTy(32) && |
1410 | isa<InsertElementInst>(bitCastInst->getOperand(0)) && | 1412 | isa<InsertElementInst>(bitCastInst->getOperand(0)) && |
1411 | bitCastInst->getOperand(0)->getType()->isVectorTy() && | 1413 | bitCastInst->getOperand(0)->getType()->isVectorTy() && |
@@ -1416,7 +1418,7 @@ index 707bbfbe..72859a91 100644 | |||
1416 | { | 1418 | { |
1417 | InsertElementInst* insertElementInst = cast<InsertElementInst>(bitCastInst->getOperand(0)); | 1419 | InsertElementInst* insertElementInst = cast<InsertElementInst>(bitCastInst->getOperand(0)); |
1418 | 1420 | ||
1419 | @@ -2731,7 +2731,7 @@ void GenSpecificPattern::visitFNeg(llvm::UnaryOperator& I) | 1421 | @@ -2979,7 +2979,7 @@ void GenSpecificPattern::visitFNeg(llvm::UnaryOperator& I) |
1420 | } | 1422 | } |
1421 | else | 1423 | else |
1422 | { | 1424 | { |
@@ -1425,7 +1427,7 @@ index 707bbfbe..72859a91 100644 | |||
1425 | fsub = llvm::UndefValue::get(I.getType()); | 1427 | fsub = llvm::UndefValue::get(I.getType()); |
1426 | 1428 | ||
1427 | for (uint32_t i = 0; i < vectorSize; ++i) | 1429 | for (uint32_t i = 0; i < vectorSize; ++i) |
1428 | @@ -2845,7 +2845,7 @@ Constant* IGCConstProp::replaceShaderConstant(LoadInst* inst) | 1430 | @@ -3056,7 +3056,7 @@ Constant* IGCConstProp::replaceShaderConstant(Instruction* inst) |
1429 | if (inst->getType()->isVectorTy()) | 1431 | if (inst->getType()->isVectorTy()) |
1430 | { | 1432 | { |
1431 | Type* srcEltTy = cast<VectorType>(inst->getType())->getElementType(); | 1433 | Type* srcEltTy = cast<VectorType>(inst->getType())->getElementType(); |
@@ -1434,7 +1436,7 @@ index 707bbfbe..72859a91 100644 | |||
1434 | uint32_t eltSize_in_bytes = (unsigned int)srcEltTy->getPrimitiveSizeInBits() / 8; | 1436 | uint32_t eltSize_in_bytes = (unsigned int)srcEltTy->getPrimitiveSizeInBits() / 8; |
1435 | IRBuilder<> builder(inst); | 1437 | IRBuilder<> builder(inst); |
1436 | Value* vectorValue = UndefValue::get(inst->getType()); | 1438 | Value* vectorValue = UndefValue::get(inst->getType()); |
1437 | @@ -3122,7 +3122,7 @@ Constant* IGCConstProp::ConstantFoldCmpInst(CmpInst* CI) | 1439 | @@ -3341,7 +3341,7 @@ Constant* IGCConstProp::ConstantFoldCmpInst(CmpInst* CI) |
1438 | { | 1440 | { |
1439 | bool AllTrue = true, AllFalse = true; | 1441 | bool AllTrue = true, AllFalse = true; |
1440 | auto VecOpnd = cast<Constant>(EEI->getVectorOperand()); | 1442 | auto VecOpnd = cast<Constant>(EEI->getVectorOperand()); |
@@ -1443,7 +1445,7 @@ index 707bbfbe..72859a91 100644 | |||
1443 | for (unsigned i = 0; i < N; ++i) | 1445 | for (unsigned i = 0; i < N; ++i) |
1444 | { | 1446 | { |
1445 | Constant* const Opnd = VecOpnd->getAggregateElement(i); | 1447 | Constant* const Opnd = VecOpnd->getAggregateElement(i); |
1446 | @@ -3918,8 +3918,8 @@ namespace IGC | 1448 | @@ -4137,8 +4137,8 @@ namespace IGC |
1447 | BitCastInst* BC = dyn_cast<BitCastInst>(&*BI++); | 1449 | BitCastInst* BC = dyn_cast<BitCastInst>(&*BI++); |
1448 | if (!BC) continue; | 1450 | if (!BC) continue; |
1449 | // Skip non-element-wise bitcast. | 1451 | // Skip non-element-wise bitcast. |
@@ -1454,7 +1456,7 @@ index 707bbfbe..72859a91 100644 | |||
1454 | if (!DstVTy || !SrcVTy || DstVTy->getNumElements() != SrcVTy->getNumElements()) | 1456 | if (!DstVTy || !SrcVTy || DstVTy->getNumElements() != SrcVTy->getNumElements()) |
1455 | continue; | 1457 | continue; |
1456 | // Skip if it's not used only all extractelement. | 1458 | // Skip if it's not used only all extractelement. |
1457 | @@ -4689,7 +4689,7 @@ void SplitIndirectEEtoSel::visitExtractElementInst(llvm::ExtractElementInst& I) | 1459 | @@ -4928,7 +4928,7 @@ void SplitIndirectEEtoSel::visitExtractElementInst(llvm::ExtractElementInst& I) |
1458 | { | 1460 | { |
1459 | using namespace llvm::PatternMatch; | 1461 | using namespace llvm::PatternMatch; |
1460 | 1462 | ||
@@ -1464,10 +1466,10 @@ index 707bbfbe..72859a91 100644 | |||
1464 | Type* eleType = vecTy->getElementType(); | 1466 | Type* eleType = vecTy->getElementType(); |
1465 | 1467 | ||
1466 | diff --git a/IGC/Compiler/DebugInfo/ScalarVISAModule.cpp b/IGC/Compiler/DebugInfo/ScalarVISAModule.cpp | 1468 | diff --git a/IGC/Compiler/DebugInfo/ScalarVISAModule.cpp b/IGC/Compiler/DebugInfo/ScalarVISAModule.cpp |
1467 | index 885ad65a..923f296e 100644 | 1469 | index d094f36e7..b66f4585a 100644 |
1468 | --- a/IGC/Compiler/DebugInfo/ScalarVISAModule.cpp | 1470 | --- a/IGC/Compiler/DebugInfo/ScalarVISAModule.cpp |
1469 | +++ b/IGC/Compiler/DebugInfo/ScalarVISAModule.cpp | 1471 | +++ b/IGC/Compiler/DebugInfo/ScalarVISAModule.cpp |
1470 | @@ -32,6 +32,7 @@ IN THE SOFTWARE. | 1472 | @@ -15,6 +15,7 @@ SPDX-License-Identifier: MIT |
1471 | #include "DebugInfo/DebugInfoUtils.hpp" | 1473 | #include "DebugInfo/DebugInfoUtils.hpp" |
1472 | 1474 | ||
1473 | #include "common/LLVMWarningsPush.hpp" | 1475 | #include "common/LLVMWarningsPush.hpp" |
@@ -1475,7 +1477,7 @@ index 885ad65a..923f296e 100644 | |||
1475 | #include "llvm/IR/Function.h" | 1477 | #include "llvm/IR/Function.h" |
1476 | #include "llvm/IR/Module.h" | 1478 | #include "llvm/IR/Module.h" |
1477 | #include "common/LLVMWarningsPop.hpp" | 1479 | #include "common/LLVMWarningsPop.hpp" |
1478 | @@ -581,7 +582,7 @@ ScalarVisaModule::GetVariableLocation(const llvm::Instruction* pInst) const | 1480 | @@ -565,7 +566,7 @@ ScalarVisaModule::GetVariableLocation(const llvm::Instruction* pInst) const |
1479 | 1481 | ||
1480 | if (pType->isVectorTy()) | 1482 | if (pType->isVectorTy()) |
1481 | { | 1483 | { |
@@ -1485,29 +1487,31 @@ index 885ad65a..923f296e 100644 | |||
1485 | else if (!pVar->IsUniform()) | 1487 | else if (!pVar->IsUniform()) |
1486 | { | 1488 | { |
1487 | diff --git a/IGC/Compiler/GenTTI.cpp b/IGC/Compiler/GenTTI.cpp | 1489 | diff --git a/IGC/Compiler/GenTTI.cpp b/IGC/Compiler/GenTTI.cpp |
1488 | index 34b21c73..9e4d2f26 100644 | 1490 | index d51396080..8a74ec287 100644 |
1489 | --- a/IGC/Compiler/GenTTI.cpp | 1491 | --- a/IGC/Compiler/GenTTI.cpp |
1490 | +++ b/IGC/Compiler/GenTTI.cpp | 1492 | +++ b/IGC/Compiler/GenTTI.cpp |
1491 | @@ -216,7 +216,13 @@ namespace llvm { | 1493 | @@ -21,6 +21,7 @@ SPDX-License-Identifier: MIT |
1494 | #include "llvm/Analysis/LoopInfo.h" | ||
1495 | #include "llvm/Analysis/ScalarEvolution.h" | ||
1496 | #include "llvm/Support/raw_ostream.h" | ||
1497 | +#include "llvmWrapper/Transforms/Utils/LoopUtils.h" | ||
1498 | #include "common/LLVMWarningsPop.hpp" | ||
1499 | |||
1500 | using namespace llvm; | ||
1501 | @@ -201,7 +202,7 @@ namespace llvm { | ||
1492 | 1502 | ||
1493 | // Skip non-simple loop. | 1503 | // Skip non-simple loop. |
1494 | if (L->getNumBlocks() != 1) { | 1504 | if (L->getNumBlocks() != 1) { |
1495 | - if (IGC_IS_FLAG_ENABLED(EnableAdvRuntimeUnroll) && L->empty()) { | 1505 | - if (IGC_IS_FLAG_ENABLED(EnableAdvRuntimeUnroll) && L->empty()) { |
1496 | + if (IGC_IS_FLAG_ENABLED(EnableAdvRuntimeUnroll) && | 1506 | + if (IGC_IS_FLAG_ENABLED(EnableAdvRuntimeUnroll) && IGCLLVM::isInnermost(L)) { |
1497 | +#if LLVM_VERSION_MAJOR >= 12 | ||
1498 | + L->isInnermost() | ||
1499 | +#else | ||
1500 | + L->empty() | ||
1501 | +#endif | ||
1502 | + ) { | ||
1503 | auto countNonPHI = [](BasicBlock* BB) { | 1507 | auto countNonPHI = [](BasicBlock* BB) { |
1504 | unsigned Total = BB->size(); | 1508 | unsigned Total = BB->size(); |
1505 | unsigned PHIs = 0; | 1509 | unsigned PHIs = 0; |
1506 | diff --git a/IGC/Compiler/GenUpdateCB.cpp b/IGC/Compiler/GenUpdateCB.cpp | 1510 | diff --git a/IGC/Compiler/GenUpdateCB.cpp b/IGC/Compiler/GenUpdateCB.cpp |
1507 | index 89b414b6..9034066c 100644 | 1511 | index f15738472..d83ced93e 100644 |
1508 | --- a/IGC/Compiler/GenUpdateCB.cpp | 1512 | --- a/IGC/Compiler/GenUpdateCB.cpp |
1509 | +++ b/IGC/Compiler/GenUpdateCB.cpp | 1513 | +++ b/IGC/Compiler/GenUpdateCB.cpp |
1510 | @@ -24,6 +24,7 @@ IN THE SOFTWARE. | 1514 | @@ -8,6 +8,7 @@ SPDX-License-Identifier: MIT |
1511 | 1515 | ||
1512 | #include "common/LLVMWarningsPush.hpp" | 1516 | #include "common/LLVMWarningsPush.hpp" |
1513 | #include "llvmWrapper/Bitcode/BitcodeWriter.h" | 1517 | #include "llvmWrapper/Bitcode/BitcodeWriter.h" |
@@ -1515,7 +1519,7 @@ index 89b414b6..9034066c 100644 | |||
1515 | #include <llvm/Support/ScaledNumber.h> | 1519 | #include <llvm/Support/ScaledNumber.h> |
1516 | #include <llvm/Bitcode/BitcodeReader.h> | 1520 | #include <llvm/Bitcode/BitcodeReader.h> |
1517 | #include "llvm/IR/DebugInfo.h" | 1521 | #include "llvm/IR/DebugInfo.h" |
1518 | @@ -384,7 +385,7 @@ bool GenUpdateCB::runOnFunction(Function& F) | 1522 | @@ -368,7 +369,7 @@ bool GenUpdateCB::runOnFunction(Function& F) |
1519 | isResInfo(dyn_cast<GenIntrinsicInst>(inst, | 1523 | isResInfo(dyn_cast<GenIntrinsicInst>(inst, |
1520 | GenISAIntrinsic::GenISA_resinfoptr), texId, lod, isUAV)) | 1524 | GenISAIntrinsic::GenISA_resinfoptr), texId, lod, isUAV)) |
1521 | { | 1525 | { |
@@ -1525,10 +1529,10 @@ index 89b414b6..9034066c 100644 | |||
1525 | if (VectorUsedByConstExtractOnly(inst, extracts)) | 1529 | if (VectorUsedByConstExtractOnly(inst, extracts)) |
1526 | { | 1530 | { |
1527 | diff --git a/IGC/Compiler/HandleFRemInstructions.cpp b/IGC/Compiler/HandleFRemInstructions.cpp | 1531 | diff --git a/IGC/Compiler/HandleFRemInstructions.cpp b/IGC/Compiler/HandleFRemInstructions.cpp |
1528 | index 94995a56..4523f3f0 100644 | 1532 | index 6b48f0b28..8025667f1 100644 |
1529 | --- a/IGC/Compiler/HandleFRemInstructions.cpp | 1533 | --- a/IGC/Compiler/HandleFRemInstructions.cpp |
1530 | +++ b/IGC/Compiler/HandleFRemInstructions.cpp | 1534 | +++ b/IGC/Compiler/HandleFRemInstructions.cpp |
1531 | @@ -28,6 +28,7 @@ IN THE SOFTWARE. | 1535 | @@ -12,6 +12,7 @@ SPDX-License-Identifier: MIT |
1532 | #include <llvm/IR/Instructions.h> | 1536 | #include <llvm/IR/Instructions.h> |
1533 | #include <llvm/IR/InstIterator.h> | 1537 | #include <llvm/IR/InstIterator.h> |
1534 | #include <llvm/Transforms/Utils/BasicBlockUtils.h> | 1538 | #include <llvm/Transforms/Utils/BasicBlockUtils.h> |
@@ -1536,7 +1540,7 @@ index 94995a56..4523f3f0 100644 | |||
1536 | #include "common/LLVMWarningsPop.hpp" | 1540 | #include "common/LLVMWarningsPop.hpp" |
1537 | #include "Compiler/IGCPassSupport.h" | 1541 | #include "Compiler/IGCPassSupport.h" |
1538 | #include "Probe/Assertion.h" | 1542 | #include "Probe/Assertion.h" |
1539 | @@ -77,7 +78,7 @@ void HandleFRemInstructions::visitFRem(llvm::BinaryOperator& I) | 1543 | @@ -61,7 +62,7 @@ void HandleFRemInstructions::visitFRem(llvm::BinaryOperator& I) |
1540 | 1544 | ||
1541 | if (ValType->isVectorTy()) | 1545 | if (ValType->isVectorTy()) |
1542 | { | 1546 | { |
@@ -1546,10 +1550,10 @@ index 94995a56..4523f3f0 100644 | |||
1546 | { | 1550 | { |
1547 | VecStr = "v" + std::to_string(VecCount); | 1551 | VecStr = "v" + std::to_string(VecCount); |
1548 | diff --git a/IGC/Compiler/HandleLoadStoreInstructions.cpp b/IGC/Compiler/HandleLoadStoreInstructions.cpp | 1552 | diff --git a/IGC/Compiler/HandleLoadStoreInstructions.cpp b/IGC/Compiler/HandleLoadStoreInstructions.cpp |
1549 | index d18eec73..c5535701 100644 | 1553 | index 40e9930b5..8c2e60018 100644 |
1550 | --- a/IGC/Compiler/HandleLoadStoreInstructions.cpp | 1554 | --- a/IGC/Compiler/HandleLoadStoreInstructions.cpp |
1551 | +++ b/IGC/Compiler/HandleLoadStoreInstructions.cpp | 1555 | +++ b/IGC/Compiler/HandleLoadStoreInstructions.cpp |
1552 | @@ -77,7 +77,7 @@ void HandleLoadStoreInstructions::visitLoadInst(llvm::LoadInst& I) | 1556 | @@ -61,7 +61,7 @@ void HandleLoadStoreInstructions::visitLoadInst(llvm::LoadInst& I) |
1553 | 1557 | ||
1554 | if (I.getType()->isVectorTy()) | 1558 | if (I.getType()->isVectorTy()) |
1555 | { | 1559 | { |
@@ -1558,7 +1562,7 @@ index d18eec73..c5535701 100644 | |||
1558 | doubleDstType = IGCLLVM::FixedVectorType::get(builder.getDoubleTy(), numVectorElements); | 1562 | doubleDstType = IGCLLVM::FixedVectorType::get(builder.getDoubleTy(), numVectorElements); |
1559 | } | 1563 | } |
1560 | uint as = ptrv->getType()->getPointerAddressSpace(); | 1564 | uint as = ptrv->getType()->getPointerAddressSpace(); |
1561 | @@ -163,7 +163,7 @@ void HandleLoadStoreInstructions::visitStoreInst(llvm::StoreInst& I) | 1565 | @@ -134,7 +134,7 @@ void HandleLoadStoreInstructions::visitStoreInst(llvm::StoreInst& I) |
1562 | 1566 | ||
1563 | if (I.getValueOperand()->getType()->isVectorTy()) | 1567 | if (I.getValueOperand()->getType()->isVectorTy()) |
1564 | { | 1568 | { |
@@ -1568,10 +1572,10 @@ index d18eec73..c5535701 100644 | |||
1568 | 1572 | ||
1569 | 1573 | ||
1570 | diff --git a/IGC/Compiler/LegalizationPass.cpp b/IGC/Compiler/LegalizationPass.cpp | 1574 | diff --git a/IGC/Compiler/LegalizationPass.cpp b/IGC/Compiler/LegalizationPass.cpp |
1571 | index b651350b..9524a495 100644 | 1575 | index bc03b251d..aa63e2b92 100644 |
1572 | --- a/IGC/Compiler/LegalizationPass.cpp | 1576 | --- a/IGC/Compiler/LegalizationPass.cpp |
1573 | +++ b/IGC/Compiler/LegalizationPass.cpp | 1577 | +++ b/IGC/Compiler/LegalizationPass.cpp |
1574 | @@ -670,7 +670,7 @@ void Legalization::visitBitCastInst(llvm::BitCastInst& I) | 1578 | @@ -671,7 +671,7 @@ void Legalization::visitBitCastInst(llvm::BitCastInst& I) |
1575 | if (!isa<TruncInst>(pZ->getOperand(0))) | 1579 | if (!isa<TruncInst>(pZ->getOperand(0))) |
1576 | return; | 1580 | return; |
1577 | 1581 | ||
@@ -1580,7 +1584,7 @@ index b651350b..9524a495 100644 | |||
1580 | if (pVecTy->getNumElements() != 3) | 1584 | if (pVecTy->getNumElements() != 3) |
1581 | return; | 1585 | return; |
1582 | 1586 | ||
1583 | @@ -765,7 +765,7 @@ void Legalization::visitSelectInst(SelectInst& I) | 1587 | @@ -766,7 +766,7 @@ void Legalization::visitSelectInst(SelectInst& I) |
1584 | } | 1588 | } |
1585 | else if (I.getType()->isVectorTy()) | 1589 | else if (I.getType()->isVectorTy()) |
1586 | { | 1590 | { |
@@ -1589,7 +1593,7 @@ index b651350b..9524a495 100644 | |||
1589 | Value* newVec = UndefValue::get(I.getType()); | 1593 | Value* newVec = UndefValue::get(I.getType()); |
1590 | m_builder->SetInsertPoint(&I); | 1594 | m_builder->SetInsertPoint(&I); |
1591 | for (unsigned int i = 0; i < vecSize; i++) | 1595 | for (unsigned int i = 0; i < vecSize; i++) |
1592 | @@ -1147,7 +1147,7 @@ void Legalization::visitStoreInst(StoreInst& I) | 1596 | @@ -1148,7 +1148,7 @@ void Legalization::visitStoreInst(StoreInst& I) |
1593 | if (ConstantDataVector * vec = dyn_cast<ConstantDataVector>(I.getOperand(0))) | 1597 | if (ConstantDataVector * vec = dyn_cast<ConstantDataVector>(I.getOperand(0))) |
1594 | { | 1598 | { |
1595 | Value* newVec = UndefValue::get(vec->getType()); | 1599 | Value* newVec = UndefValue::get(vec->getType()); |
@@ -1607,7 +1611,7 @@ index b651350b..9524a495 100644 | |||
1607 | for (unsigned int i = 0; i < nbElement; i++) | 1611 | for (unsigned int i = 0; i < nbElement; i++) |
1608 | { | 1612 | { |
1609 | Constant* cst = vec->getOperand(i); | 1613 | Constant* cst = vec->getOperand(i); |
1610 | @@ -1189,7 +1189,7 @@ void Legalization::visitStoreInst(StoreInst& I) | 1614 | @@ -1188,7 +1188,7 @@ void Legalization::visitStoreInst(StoreInst& I) |
1611 | else if (ConstantAggregateZero * vec = dyn_cast<ConstantAggregateZero>(I.getOperand(0))) | 1615 | else if (ConstantAggregateZero * vec = dyn_cast<ConstantAggregateZero>(I.getOperand(0))) |
1612 | { | 1616 | { |
1613 | Value* newVec = UndefValue::get(vec->getType()); | 1617 | Value* newVec = UndefValue::get(vec->getType()); |
@@ -1616,7 +1620,7 @@ index b651350b..9524a495 100644 | |||
1616 | for (unsigned int i = 0; i < nbElement; i++) | 1620 | for (unsigned int i = 0; i < nbElement; i++) |
1617 | { | 1621 | { |
1618 | Constant* cst = vec->getElementValue(i); | 1622 | Constant* cst = vec->getElementValue(i); |
1619 | @@ -1340,7 +1340,7 @@ void Legalization::visitInsertElementInst(InsertElementInst& I) | 1623 | @@ -1338,7 +1338,7 @@ void Legalization::visitInsertElementInst(InsertElementInst& I) |
1620 | if (ConstantDataVector * vec = dyn_cast<ConstantDataVector>(I.getOperand(0))) | 1624 | if (ConstantDataVector * vec = dyn_cast<ConstantDataVector>(I.getOperand(0))) |
1621 | { | 1625 | { |
1622 | Value* newVec = UndefValue::get(vec->getType()); | 1626 | Value* newVec = UndefValue::get(vec->getType()); |
@@ -1625,7 +1629,7 @@ index b651350b..9524a495 100644 | |||
1625 | for (unsigned int i = 0; i < nbElement; i++) | 1629 | for (unsigned int i = 0; i < nbElement; i++) |
1626 | { | 1630 | { |
1627 | Constant* cst = vec->getElementAsConstant(i); | 1631 | Constant* cst = vec->getElementAsConstant(i); |
1628 | @@ -1360,7 +1360,7 @@ void Legalization::visitInsertElementInst(InsertElementInst& I) | 1632 | @@ -1358,7 +1358,7 @@ void Legalization::visitInsertElementInst(InsertElementInst& I) |
1629 | else if (ConstantVector * vec = dyn_cast<ConstantVector>(I.getOperand(0))) | 1633 | else if (ConstantVector * vec = dyn_cast<ConstantVector>(I.getOperand(0))) |
1630 | { | 1634 | { |
1631 | Value* newVec = UndefValue::get(I.getType()); | 1635 | Value* newVec = UndefValue::get(I.getType()); |
@@ -1634,7 +1638,7 @@ index b651350b..9524a495 100644 | |||
1634 | for (unsigned int i = 0; i < nbElement; i++) | 1638 | for (unsigned int i = 0; i < nbElement; i++) |
1635 | { | 1639 | { |
1636 | Constant* cst = vec->getOperand(i); | 1640 | Constant* cst = vec->getOperand(i); |
1637 | @@ -1380,7 +1380,7 @@ void Legalization::visitInsertElementInst(InsertElementInst& I) | 1641 | @@ -1378,7 +1378,7 @@ void Legalization::visitInsertElementInst(InsertElementInst& I) |
1638 | else if (ConstantAggregateZero * vec = dyn_cast<ConstantAggregateZero>(I.getOperand(0))) | 1642 | else if (ConstantAggregateZero * vec = dyn_cast<ConstantAggregateZero>(I.getOperand(0))) |
1639 | { | 1643 | { |
1640 | Value* newVec = UndefValue::get(I.getType()); | 1644 | Value* newVec = UndefValue::get(I.getType()); |
@@ -1643,7 +1647,7 @@ index b651350b..9524a495 100644 | |||
1643 | for (unsigned int i = 0; i < nbElement; i++) | 1647 | for (unsigned int i = 0; i < nbElement; i++) |
1644 | { | 1648 | { |
1645 | Constant* cst = vec->getElementValue(i); | 1649 | Constant* cst = vec->getElementValue(i); |
1646 | @@ -1397,7 +1397,7 @@ void Legalization::visitInsertElementInst(InsertElementInst& I) | 1650 | @@ -1395,7 +1395,7 @@ void Legalization::visitInsertElementInst(InsertElementInst& I) |
1647 | else if (I.getOperand(1)->getType()->isIntegerTy(1)) | 1651 | else if (I.getOperand(1)->getType()->isIntegerTy(1)) |
1648 | { | 1652 | { |
1649 | // This promotes i1 insertelement to i32 | 1653 | // This promotes i1 insertelement to i32 |
@@ -1652,7 +1656,7 @@ index b651350b..9524a495 100644 | |||
1652 | Value* newVec = UndefValue::get(IGCLLVM::FixedVectorType::get(m_builder->getInt32Ty(), nbElement)); | 1656 | Value* newVec = UndefValue::get(IGCLLVM::FixedVectorType::get(m_builder->getInt32Ty(), nbElement)); |
1653 | PromoteInsertElement(&I, newVec); | 1657 | PromoteInsertElement(&I, newVec); |
1654 | } | 1658 | } |
1655 | @@ -1410,7 +1410,7 @@ void Legalization::visitShuffleVectorInst(ShuffleVectorInst& I) | 1659 | @@ -1408,7 +1408,7 @@ void Legalization::visitShuffleVectorInst(ShuffleVectorInst& I) |
1656 | // If the original vector is a constant, just use the scalar constant, | 1660 | // If the original vector is a constant, just use the scalar constant, |
1657 | // otherwise extract from the original vector. | 1661 | // otherwise extract from the original vector. |
1658 | 1662 | ||
@@ -1661,7 +1665,7 @@ index b651350b..9524a495 100644 | |||
1661 | Value* newVec = UndefValue::get(resType); | 1665 | Value* newVec = UndefValue::get(resType); |
1662 | Value* src0 = I.getOperand(0); | 1666 | Value* src0 = I.getOperand(0); |
1663 | Value* src1 = I.getOperand(1); | 1667 | Value* src1 = I.getOperand(1); |
1664 | @@ -1421,7 +1421,7 @@ void Legalization::visitShuffleVectorInst(ShuffleVectorInst& I) | 1668 | @@ -1419,7 +1419,7 @@ void Legalization::visitShuffleVectorInst(ShuffleVectorInst& I) |
1665 | Constant* mask = I.getShuffleMaskForBitcode(); | 1669 | Constant* mask = I.getShuffleMaskForBitcode(); |
1666 | #endif | 1670 | #endif |
1667 | // The two inputs are guaranteed to be of the same type | 1671 | // The two inputs are guaranteed to be of the same type |
@@ -1670,7 +1674,7 @@ index b651350b..9524a495 100644 | |||
1670 | int inCount = int_cast<int>(inType->getNumElements()); | 1674 | int inCount = int_cast<int>(inType->getNumElements()); |
1671 | int inBase = 2; // 2 means using undef | 1675 | int inBase = 2; // 2 means using undef |
1672 | // if inType == resType, use src0/src1 as the input | 1676 | // if inType == resType, use src0/src1 as the input |
1673 | @@ -1637,7 +1637,7 @@ Type* Legalization::LegalAllocaType(Type* type) const | 1677 | @@ -1635,7 +1635,7 @@ Type* Legalization::LegalAllocaType(Type* type) const |
1674 | #endif | 1678 | #endif |
1675 | legalType = IGCLLVM::FixedVectorType::get( | 1679 | legalType = IGCLLVM::FixedVectorType::get( |
1676 | LegalAllocaType(cast<VectorType>(type)->getElementType()), | 1680 | LegalAllocaType(cast<VectorType>(type)->getElementType()), |
@@ -1679,7 +1683,7 @@ index b651350b..9524a495 100644 | |||
1679 | break; | 1683 | break; |
1680 | case Type::StructTyID: | 1684 | case Type::StructTyID: |
1681 | return LegalStructAllocaType(type); | 1685 | return LegalStructAllocaType(type); |
1682 | @@ -1937,7 +1937,7 @@ void Legalization::visitIntrinsicInst(llvm::IntrinsicInst& I) | 1686 | @@ -1976,7 +1976,7 @@ void Legalization::visitIntrinsicInst(llvm::IntrinsicInst& I) |
1683 | Value* newValue = nullptr; | 1687 | Value* newValue = nullptr; |
1684 | if (srcType->isVectorTy()) | 1688 | if (srcType->isVectorTy()) |
1685 | { | 1689 | { |
@@ -1688,7 +1692,7 @@ index b651350b..9524a495 100644 | |||
1688 | const unsigned int numElements = (uint32_t)sourceVT->getNumElements(); | 1692 | const unsigned int numElements = (uint32_t)sourceVT->getNumElements(); |
1689 | Value* dstVec = UndefValue::get(srcType); | 1693 | Value* dstVec = UndefValue::get(srcType); |
1690 | for (unsigned int i = 0; i < numElements; ++i) | 1694 | for (unsigned int i = 0; i < numElements; ++i) |
1691 | @@ -2081,7 +2081,7 @@ void Legalization::visitTruncInst(llvm::TruncInst& I) { | 1695 | @@ -2119,7 +2119,7 @@ void Legalization::visitTruncInst(llvm::TruncInst& I) { |
1692 | } | 1696 | } |
1693 | 1697 | ||
1694 | Src = BC->getOperand(0); | 1698 | Src = BC->getOperand(0); |
@@ -1698,41 +1702,23 @@ index b651350b..9524a495 100644 | |||
1698 | if (!VTy || VTy->getNumElements() != 3 || !VTy->getElementType()->isIntegerTy(16)) | 1702 | if (!VTy || VTy->getNumElements() != 3 || !VTy->getElementType()->isIntegerTy(16)) |
1699 | return; | 1703 | return; |
1700 | diff --git a/IGC/Compiler/Legalizer/InstPromoter.cpp b/IGC/Compiler/Legalizer/InstPromoter.cpp | 1704 | diff --git a/IGC/Compiler/Legalizer/InstPromoter.cpp b/IGC/Compiler/Legalizer/InstPromoter.cpp |
1701 | index bba1d96d..8fadf89f 100644 | 1705 | index 56c0b9981..6d1778b61 100644 |
1702 | --- a/IGC/Compiler/Legalizer/InstPromoter.cpp | 1706 | --- a/IGC/Compiler/Legalizer/InstPromoter.cpp |
1703 | +++ b/IGC/Compiler/Legalizer/InstPromoter.cpp | 1707 | +++ b/IGC/Compiler/Legalizer/InstPromoter.cpp |
1704 | @@ -25,6 +25,7 @@ IN THE SOFTWARE. | 1708 | @@ -382,7 +382,7 @@ bool InstPromoter::visitBitCastInst(BitCastInst& I) { |
1705 | #define DEBUG_TYPE "type-legalizer" | 1709 | IRB->CreateBitCast(Val, IGCLLVM::FixedVectorType::get(DestTy->getScalarType(), N)); |
1706 | #include "TypeLegalizer.h" | ||
1707 | #include "InstPromoter.h" | ||
1708 | +#include "llvmWrapper/IR/DerivedTypes.h" | ||
1709 | #include "common/LLVMWarningsPush.hpp" | ||
1710 | #include "llvm/Support/Debug.h" | ||
1711 | #include "llvm/Support/MathExtras.h" | ||
1712 | @@ -394,10 +395,16 @@ bool InstPromoter::visitBitCastInst(BitCastInst& I) { | ||
1713 | unsigned N = | ||
1714 | Val->getType()->getScalarSizeInBits() / DestTy->getScalarSizeInBits(); | ||
1715 | Value* BC = | ||
1716 | - IRB->CreateBitCast(Val, VectorType::get(DestTy->getScalarType(), N)); | ||
1717 | + IRB->CreateBitCast(Val, IGCLLVM::FixedVectorType::get(DestTy->getScalarType(), N)); | ||
1718 | 1710 | ||
1719 | std::vector<Constant*> Vals; | 1711 | std::vector<Constant*> Vals; |
1720 | - for (unsigned i = 0; i < DestTy->getVectorNumElements(); i++) | 1712 | - for (unsigned i = 0; i < (unsigned)cast<VectorType>(DestTy)->getNumElements(); i++) |
1721 | + for (unsigned i = 0; | 1713 | + for (unsigned i = 0; i < cast<IGCLLVM::FixedVectorType>(DestTy)->getNumElements(); i++) |
1722 | +#if LLVM_VERSION_MAJOR >= 12 | ||
1723 | + i < cast<IGCLLVM::FixedVectorType>(DestTy)->getNumElements(); | ||
1724 | +#else | ||
1725 | + i < DestTy->getVectorNumElements(); | ||
1726 | +#endif | ||
1727 | + i++) | ||
1728 | Vals.push_back(IRB->getInt32(i)); | 1714 | Vals.push_back(IRB->getInt32(i)); |
1729 | 1715 | ||
1730 | Value* Mask = ConstantVector::get(Vals); | 1716 | Value* Mask = ConstantVector::get(Vals); |
1731 | diff --git a/IGC/Compiler/Legalizer/InstScalarizer.cpp b/IGC/Compiler/Legalizer/InstScalarizer.cpp | 1717 | diff --git a/IGC/Compiler/Legalizer/InstScalarizer.cpp b/IGC/Compiler/Legalizer/InstScalarizer.cpp |
1732 | index 3902f740..c8e18cc5 100644 | 1718 | index c2b3f7ba7..2e519eef0 100644 |
1733 | --- a/IGC/Compiler/Legalizer/InstScalarizer.cpp | 1719 | --- a/IGC/Compiler/Legalizer/InstScalarizer.cpp |
1734 | +++ b/IGC/Compiler/Legalizer/InstScalarizer.cpp | 1720 | +++ b/IGC/Compiler/Legalizer/InstScalarizer.cpp |
1735 | @@ -130,7 +130,7 @@ bool InstScalarizer::visitLoadInst(LoadInst& I) { | 1721 | @@ -114,7 +114,7 @@ bool InstScalarizer::visitLoadInst(LoadInst& I) { |
1736 | // otherwise, it's broken. | 1722 | // otherwise, it's broken. |
1737 | IGC_ASSERT(TL->getTypeSizeInBits(EltTy) == TL->getTypeStoreSizeInBits(EltTy)); | 1723 | IGC_ASSERT(TL->getTypeSizeInBits(EltTy) == TL->getTypeStoreSizeInBits(EltTy)); |
1738 | 1724 | ||
@@ -1741,7 +1727,7 @@ index 3902f740..c8e18cc5 100644 | |||
1741 | unsigned Elt = 0; | 1727 | unsigned Elt = 0; |
1742 | 1728 | ||
1743 | Type* NewPtrTy = PointerType::get(EltTy, AS); | 1729 | Type* NewPtrTy = PointerType::get(EltTy, AS); |
1744 | @@ -268,7 +268,7 @@ bool InstScalarizer::visitStoreInst(StoreInst& I) { | 1730 | @@ -252,7 +252,7 @@ bool InstScalarizer::visitStoreInst(StoreInst& I) { |
1745 | // otherwise, it's broken. | 1731 | // otherwise, it's broken. |
1746 | IGC_ASSERT(TL->getTypeSizeInBits(EltTy) == TL->getTypeStoreSizeInBits(EltTy)); | 1732 | IGC_ASSERT(TL->getTypeSizeInBits(EltTy) == TL->getTypeStoreSizeInBits(EltTy)); |
1747 | 1733 | ||
@@ -1750,7 +1736,7 @@ index 3902f740..c8e18cc5 100644 | |||
1750 | unsigned Elt = 0; | 1736 | unsigned Elt = 0; |
1751 | 1737 | ||
1752 | Type* NewPtrTy = PointerType::get(EltTy, AS); | 1738 | Type* NewPtrTy = PointerType::get(EltTy, AS); |
1753 | @@ -498,7 +498,7 @@ bool InstScalarizer::visitInsertElementInst(InsertElementInst& I) { | 1739 | @@ -482,7 +482,7 @@ bool InstScalarizer::visitInsertElementInst(InsertElementInst& I) { |
1754 | IGC_ASSERT(EltSeq->size()); | 1740 | IGC_ASSERT(EltSeq->size()); |
1755 | IGC_ASSERT(VecSeqCopy.size() % EltSeq->size() == 0); | 1741 | IGC_ASSERT(VecSeqCopy.size() % EltSeq->size() == 0); |
1756 | 1742 | ||
@@ -1760,10 +1746,10 @@ index 3902f740..c8e18cc5 100644 | |||
1760 | for (unsigned Elt = 0; Elt != NumElts; ++Elt) { | 1746 | for (unsigned Elt = 0; Elt != NumElts; ++Elt) { |
1761 | if (Elt == Idx) { | 1747 | if (Elt == Idx) { |
1762 | diff --git a/IGC/Compiler/Legalizer/PeepholeTypeLegalizer.cpp b/IGC/Compiler/Legalizer/PeepholeTypeLegalizer.cpp | 1748 | diff --git a/IGC/Compiler/Legalizer/PeepholeTypeLegalizer.cpp b/IGC/Compiler/Legalizer/PeepholeTypeLegalizer.cpp |
1763 | index 28f8ce84..778132c4 100644 | 1749 | index 5257de424..7d93665dc 100644 |
1764 | --- a/IGC/Compiler/Legalizer/PeepholeTypeLegalizer.cpp | 1750 | --- a/IGC/Compiler/Legalizer/PeepholeTypeLegalizer.cpp |
1765 | +++ b/IGC/Compiler/Legalizer/PeepholeTypeLegalizer.cpp | 1751 | +++ b/IGC/Compiler/Legalizer/PeepholeTypeLegalizer.cpp |
1766 | @@ -162,7 +162,7 @@ void PeepholeTypeLegalizer::legalizePhiInstruction(Instruction& I) | 1752 | @@ -146,7 +146,7 @@ void PeepholeTypeLegalizer::legalizePhiInstruction(Instruction& I) |
1767 | 1753 | ||
1768 | if (quotient > 1) | 1754 | if (quotient > 1) |
1769 | { | 1755 | { |
@@ -1772,7 +1758,7 @@ index 28f8ce84..778132c4 100644 | |||
1772 | Type* newType = IGCLLVM::FixedVectorType::get(Type::getIntNTy(I.getContext(), promoteToInt), quotient * numElements); | 1758 | Type* newType = IGCLLVM::FixedVectorType::get(Type::getIntNTy(I.getContext(), promoteToInt), quotient * numElements); |
1773 | 1759 | ||
1774 | PHINode* newPhi = m_builder->CreatePHI(newType, oldPhi->getNumIncomingValues()); | 1760 | PHINode* newPhi = m_builder->CreatePHI(newType, oldPhi->getNumIncomingValues()); |
1775 | @@ -223,7 +223,7 @@ void PeepholeTypeLegalizer::legalizeExtractElement(Instruction& I) | 1761 | @@ -207,7 +207,7 @@ void PeepholeTypeLegalizer::legalizeExtractElement(Instruction& I) |
1776 | unsigned elementWidth = extract->getType()->getScalarSizeInBits(); | 1762 | unsigned elementWidth = extract->getType()->getScalarSizeInBits(); |
1777 | if (!isLegalInteger(elementWidth) && extract->getType()->isIntOrIntVectorTy()) | 1763 | if (!isLegalInteger(elementWidth) && extract->getType()->isIntOrIntVectorTy()) |
1778 | { | 1764 | { |
@@ -1781,7 +1767,7 @@ index 28f8ce84..778132c4 100644 | |||
1781 | unsigned quotient, promoteToInt; | 1767 | unsigned quotient, promoteToInt; |
1782 | promoteInt(elementWidth, quotient, promoteToInt, DL->getLargestLegalIntTypeSizeInBits()); | 1768 | promoteInt(elementWidth, quotient, promoteToInt, DL->getLargestLegalIntTypeSizeInBits()); |
1783 | 1769 | ||
1784 | @@ -850,7 +850,7 @@ void PeepholeTypeLegalizer::cleanupZExtInst(Instruction& I) { | 1770 | @@ -834,7 +834,7 @@ void PeepholeTypeLegalizer::cleanupZExtInst(Instruction& I) { |
1785 | } | 1771 | } |
1786 | 1772 | ||
1787 | unsigned ipElmtSize = prevInst->getOperand(0)->getType()->getScalarSizeInBits(); | 1773 | unsigned ipElmtSize = prevInst->getOperand(0)->getType()->getScalarSizeInBits(); |
@@ -1791,10 +1777,10 @@ index 28f8ce84..778132c4 100644 | |||
1791 | 1777 | ||
1792 | Value* vecRes = UndefValue::get(IGCLLVM::FixedVectorType::get(llvm::Type::getIntNTy(I.getContext(), promoteToInt), quotient)); | 1778 | Value* vecRes = UndefValue::get(IGCLLVM::FixedVectorType::get(llvm::Type::getIntNTy(I.getContext(), promoteToInt), quotient)); |
1793 | diff --git a/IGC/Compiler/Legalizer/TypeLegalizer.cpp b/IGC/Compiler/Legalizer/TypeLegalizer.cpp | 1779 | diff --git a/IGC/Compiler/Legalizer/TypeLegalizer.cpp b/IGC/Compiler/Legalizer/TypeLegalizer.cpp |
1794 | index f8f71d71..a46f42bd 100644 | 1780 | index 6dfdae07e..03f866f14 100644 |
1795 | --- a/IGC/Compiler/Legalizer/TypeLegalizer.cpp | 1781 | --- a/IGC/Compiler/Legalizer/TypeLegalizer.cpp |
1796 | +++ b/IGC/Compiler/Legalizer/TypeLegalizer.cpp | 1782 | +++ b/IGC/Compiler/Legalizer/TypeLegalizer.cpp |
1797 | @@ -32,6 +32,7 @@ IN THE SOFTWARE. | 1783 | @@ -16,6 +16,7 @@ SPDX-License-Identifier: MIT |
1798 | #include "InstScalarizer.h" | 1784 | #include "InstScalarizer.h" |
1799 | #include "InstElementizer.h" | 1785 | #include "InstElementizer.h" |
1800 | #include "common/LLVMWarningsPush.hpp" | 1786 | #include "common/LLVMWarningsPush.hpp" |
@@ -1802,7 +1788,7 @@ index f8f71d71..a46f42bd 100644 | |||
1802 | #include "llvm/ADT/PostOrderIterator.h" | 1788 | #include "llvm/ADT/PostOrderIterator.h" |
1803 | #include "llvm/IR/CFG.h" | 1789 | #include "llvm/IR/CFG.h" |
1804 | #include "llvm/Support/Debug.h" | 1790 | #include "llvm/Support/Debug.h" |
1805 | @@ -290,7 +291,7 @@ TypeSeq* TypeLegalizer::getScalarizedTypeSeq(Type* Ty) { | 1791 | @@ -274,7 +275,7 @@ TypeSeq* TypeLegalizer::getScalarizedTypeSeq(Type* Ty) { |
1806 | } | 1792 | } |
1807 | 1793 | ||
1808 | Type* EltTy = cast<VectorType>(Ty)->getElementType(); | 1794 | Type* EltTy = cast<VectorType>(Ty)->getElementType(); |
@@ -1812,10 +1798,10 @@ index f8f71d71..a46f42bd 100644 | |||
1812 | 1798 | ||
1813 | return &TMI->second; | 1799 | return &TMI->second; |
1814 | diff --git a/IGC/Compiler/Legalizer/TypeLegalizer.h b/IGC/Compiler/Legalizer/TypeLegalizer.h | 1800 | diff --git a/IGC/Compiler/Legalizer/TypeLegalizer.h b/IGC/Compiler/Legalizer/TypeLegalizer.h |
1815 | index f3ce264c..a6910d08 100644 | 1801 | index b51f00ac7..ee6d41992 100644 |
1816 | --- a/IGC/Compiler/Legalizer/TypeLegalizer.h | 1802 | --- a/IGC/Compiler/Legalizer/TypeLegalizer.h |
1817 | +++ b/IGC/Compiler/Legalizer/TypeLegalizer.h | 1803 | +++ b/IGC/Compiler/Legalizer/TypeLegalizer.h |
1818 | @@ -28,6 +28,7 @@ IN THE SOFTWARE. | 1804 | @@ -12,6 +12,7 @@ SPDX-License-Identifier: MIT |
1819 | #include "llvmWrapper/IR/Instructions.h" | 1805 | #include "llvmWrapper/IR/Instructions.h" |
1820 | #include "llvmWrapper/Analysis/InlineCost.h" | 1806 | #include "llvmWrapper/Analysis/InlineCost.h" |
1821 | #include "llvmWrapper/IR/InstrTypes.h" | 1807 | #include "llvmWrapper/IR/InstrTypes.h" |
@@ -1823,7 +1809,7 @@ index f3ce264c..a6910d08 100644 | |||
1823 | #include "llvmWrapper/Support/Alignment.h" | 1809 | #include "llvmWrapper/Support/Alignment.h" |
1824 | #include "llvm/ADT/ArrayRef.h" | 1810 | #include "llvm/ADT/ArrayRef.h" |
1825 | #include "llvm/ADT/DenseMap.h" | 1811 | #include "llvm/ADT/DenseMap.h" |
1826 | @@ -287,7 +288,7 @@ namespace IGC { | 1812 | @@ -271,7 +272,7 @@ namespace IGC { |
1827 | if (!Ty->isVectorTy()) | 1813 | if (!Ty->isVectorTy()) |
1828 | return false; | 1814 | return false; |
1829 | 1815 | ||
@@ -1833,52 +1819,35 @@ index f3ce264c..a6910d08 100644 | |||
1833 | const auto& ProfitLengths = getProfitLoadVectorLength(EltTy); | 1819 | const auto& ProfitLengths = getProfitLoadVectorLength(EltTy); |
1834 | 1820 | ||
1835 | diff --git a/IGC/Compiler/LowPrecisionOptPass.cpp b/IGC/Compiler/LowPrecisionOptPass.cpp | 1821 | diff --git a/IGC/Compiler/LowPrecisionOptPass.cpp b/IGC/Compiler/LowPrecisionOptPass.cpp |
1836 | index 64912719..1d1db949 100644 | 1822 | index b7a1474f4..e3d65bc18 100644 |
1837 | --- a/IGC/Compiler/LowPrecisionOptPass.cpp | 1823 | --- a/IGC/Compiler/LowPrecisionOptPass.cpp |
1838 | +++ b/IGC/Compiler/LowPrecisionOptPass.cpp | 1824 | +++ b/IGC/Compiler/LowPrecisionOptPass.cpp |
1839 | @@ -252,7 +252,7 @@ bool LowPrecisionOpt::propagateSamplerType(llvm::GenIntrinsicInst& I) | 1825 | @@ -253,7 +253,7 @@ bool LowPrecisionOpt::propagateSamplerType(llvm::GenIntrinsicInst& I) |
1840 | return false; | 1826 | |
1827 | if (I.getType()->isVectorTy()) | ||
1828 | { | ||
1829 | - numberOfElements = int_cast<unsigned int>(cast<VectorType>(I.getType())->getNumElements()); | ||
1830 | + numberOfElements = int_cast<unsigned int>(cast<IGCLLVM::FixedVectorType>(I.getType())->getNumElements()); | ||
1841 | } | 1831 | } |
1842 | 1832 | ||
1843 | - VectorType* oldTy = cast<VectorType>(I.getType()); | ||
1844 | + IGCLLVM::FixedVectorType* oldTy = cast<IGCLLVM::FixedVectorType>(I.getType()); | ||
1845 | llvm::SmallVector<llvm::Type*, 4> overloadTys; | 1833 | llvm::SmallVector<llvm::Type*, 4> overloadTys; |
1846 | auto retTy = IGCLLVM::FixedVectorType::get(newDstType, int_cast<unsigned int>(oldTy->getNumElements())); | ||
1847 | overloadTys.push_back(retTy); | ||
1848 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp | 1834 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp |
1849 | index 393b4b0a..e9c07b34 100644 | 1835 | index b22e14bd5..a2b5927ea 100644 |
1850 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp | 1836 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp |
1851 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp | 1837 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.cpp |
1852 | @@ -23,6 +23,8 @@ IN THE SOFTWARE. | 1838 | @@ -7,6 +7,7 @@ SPDX-License-Identifier: MIT |
1853 | ============================= end_copyright_notice ===========================*/ | 1839 | ============================= end_copyright_notice ===========================*/ |
1854 | 1840 | ||
1855 | #include "llvm/Config/llvm-config.h" | 1841 | #include "llvm/Config/llvm-config.h" |
1856 | +#include "llvmWrapper/IR/DerivedTypes.h" | 1842 | +#include <llvm/Analysis/TargetLibraryInfo.h> |
1857 | +#include "llvmWrapper/Analysis/TargetLibraryInfo.h" | ||
1858 | #include "Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.h" | 1843 | #include "Compiler/Optimizer/OpenCLPasses/AddressSpaceAliasAnalysis/AddressSpaceAliasAnalysis.h" |
1859 | #include "Compiler/CodeGenPublic.h" | 1844 | #include "Compiler/CodeGenPublic.h" |
1860 | #include "Compiler/IGCPassSupport.h" | 1845 | #include "Compiler/IGCPassSupport.h" |
1861 | @@ -178,12 +180,11 @@ namespace { | ||
1862 | bool doInitialization(Module& M) override { | ||
1863 | if(M.size() > 0) | ||
1864 | { | ||
1865 | - Result.reset(new AddressSpaceAAResult( | ||
1866 | - getAnalysis<TargetLibraryInfoWrapperPass>().getTLI( | ||
1867 | #if LLVM_VERSION_MAJOR >= 10 | ||
1868 | - *M.begin() | ||
1869 | + Function &F = *M.begin(); | ||
1870 | #endif | ||
1871 | - ), | ||
1872 | + Result.reset(new AddressSpaceAAResult( | ||
1873 | + getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(), | ||
1874 | *getAnalysis<CodeGenContextWrapper>().getCodeGenContext())); | ||
1875 | } | ||
1876 | return false; | ||
1877 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/AggregateArguments/AggregateArguments.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/AggregateArguments/AggregateArguments.cpp | 1846 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/AggregateArguments/AggregateArguments.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/AggregateArguments/AggregateArguments.cpp |
1878 | index 67535cfc..3cdd2975 100644 | 1847 | index 0edacda24..7c0e7936d 100644 |
1879 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/AggregateArguments/AggregateArguments.cpp | 1848 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/AggregateArguments/AggregateArguments.cpp |
1880 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/AggregateArguments/AggregateArguments.cpp | 1849 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/AggregateArguments/AggregateArguments.cpp |
1881 | @@ -24,6 +24,7 @@ IN THE SOFTWARE. | 1850 | @@ -8,6 +8,7 @@ SPDX-License-Identifier: MIT |
1882 | 1851 | ||
1883 | #include "Compiler/Optimizer/OpenCLPasses/AggregateArguments/AggregateArguments.hpp" | 1852 | #include "Compiler/Optimizer/OpenCLPasses/AggregateArguments/AggregateArguments.hpp" |
1884 | #include "Compiler/IGCPassSupport.h" | 1853 | #include "Compiler/IGCPassSupport.h" |
@@ -1886,7 +1855,7 @@ index 67535cfc..3cdd2975 100644 | |||
1886 | #include "common/LLVMWarningsPush.hpp" | 1855 | #include "common/LLVMWarningsPush.hpp" |
1887 | #include "llvm/IR/Function.h" | 1856 | #include "llvm/IR/Function.h" |
1888 | #include "common/LLVMWarningsPop.hpp" | 1857 | #include "common/LLVMWarningsPop.hpp" |
1889 | @@ -130,7 +131,7 @@ static uint64_t getNumElements(Type* type) | 1858 | @@ -114,7 +115,7 @@ static uint64_t getNumElements(Type* type) |
1890 | { | 1859 | { |
1891 | return arrayType->getNumElements(); | 1860 | return arrayType->getNumElements(); |
1892 | } | 1861 | } |
@@ -1896,10 +1865,10 @@ index 67535cfc..3cdd2975 100644 | |||
1896 | return vectorType->getNumElements(); | 1865 | return vectorType->getNumElements(); |
1897 | } | 1866 | } |
1898 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/CorrectlyRoundedDivSqrt/CorrectlyRoundedDivSqrt.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/CorrectlyRoundedDivSqrt/CorrectlyRoundedDivSqrt.cpp | 1867 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/CorrectlyRoundedDivSqrt/CorrectlyRoundedDivSqrt.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/CorrectlyRoundedDivSqrt/CorrectlyRoundedDivSqrt.cpp |
1899 | index 2a9d7499..79ed0948 100644 | 1868 | index 68b188d0b..ecc1b648a 100644 |
1900 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/CorrectlyRoundedDivSqrt/CorrectlyRoundedDivSqrt.cpp | 1869 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/CorrectlyRoundedDivSqrt/CorrectlyRoundedDivSqrt.cpp |
1901 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/CorrectlyRoundedDivSqrt/CorrectlyRoundedDivSqrt.cpp | 1870 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/CorrectlyRoundedDivSqrt/CorrectlyRoundedDivSqrt.cpp |
1902 | @@ -27,6 +27,7 @@ IN THE SOFTWARE. | 1871 | @@ -11,6 +11,7 @@ SPDX-License-Identifier: MIT |
1903 | #include "Compiler/IGCPassSupport.h" | 1872 | #include "Compiler/IGCPassSupport.h" |
1904 | #include "GenISAIntrinsics/GenIntrinsicInst.h" | 1873 | #include "GenISAIntrinsics/GenIntrinsicInst.h" |
1905 | 1874 | ||
@@ -1907,7 +1876,7 @@ index 2a9d7499..79ed0948 100644 | |||
1907 | #include "common/LLVMWarningsPush.hpp" | 1876 | #include "common/LLVMWarningsPush.hpp" |
1908 | #include <llvm/IR/Function.h> | 1877 | #include <llvm/IR/Function.h> |
1909 | #include <llvm/IR/Instructions.h> | 1878 | #include <llvm/IR/Instructions.h> |
1910 | @@ -133,7 +134,7 @@ Value* CorrectlyRoundedDivSqrt::emitIEEEDivide(BinaryOperator* I, Value* Op0, Va | 1879 | @@ -117,7 +118,7 @@ Value* CorrectlyRoundedDivSqrt::emitIEEEDivide(BinaryOperator* I, Value* Op0, Va |
1911 | } | 1880 | } |
1912 | else | 1881 | else |
1913 | { | 1882 | { |
@@ -1917,22 +1886,19 @@ index 2a9d7499..79ed0948 100644 | |||
1917 | Divide = UndefValue::get(Ty); | 1886 | Divide = UndefValue::get(Ty); |
1918 | for (unsigned i = 0; i < VecLen; i++) | 1887 | for (unsigned i = 0; i < VecLen; i++) |
1919 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp | 1888 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp |
1920 | index c5ff3ac0..119520ed 100644 | 1889 | index 387350066..86673d224 100644 |
1921 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp | 1890 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp |
1922 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp | 1891 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/DeviceEnqueueFuncs/TransformBlocks.cpp |
1923 | @@ -952,7 +952,11 @@ namespace //Anonymous | 1892 | @@ -937,7 +937,7 @@ namespace //Anonymous |
1924 | { | 1893 | { |
1925 | auto ndrangeStructName = "struct.ndrange_t"; | 1894 | auto ndrangeStructName = "struct.ndrange_t"; |
1926 | auto module = _deviceExecCall->getModule(); | 1895 | auto module = _deviceExecCall->getModule(); |
1927 | +#if LLVM_VERSION_MAJOR >= 12 | 1896 | - auto ndrangeTy = module->getTypeByName(ndrangeStructName); |
1928 | + auto ndrangeTy = llvm::StructType::getTypeByName(module->getContext(), ndrangeStructName); | 1897 | + auto ndrangeTy = IGCLLVM::getTypeByName(module, ndrangeStructName); |
1929 | +#else | ||
1930 | auto ndrangeTy = module->getTypeByName(ndrangeStructName); | ||
1931 | +#endif | ||
1932 | if (ndrangeTy == nullptr) | 1898 | if (ndrangeTy == nullptr) |
1933 | { | 1899 | { |
1934 | //create struct type | 1900 | //create struct type |
1935 | @@ -1697,7 +1701,7 @@ namespace //Anonymous | 1901 | @@ -1682,7 +1682,7 @@ namespace //Anonymous |
1936 | // this generates <element_type><num_elements> string. Ie for char2 element_type is char and num_elements is 2 | 1902 | // this generates <element_type><num_elements> string. Ie for char2 element_type is char and num_elements is 2 |
1937 | // that is done by callin BaseTypeName on vector element type, this recursive call has only a depth of one since | 1903 | // that is done by callin BaseTypeName on vector element type, this recursive call has only a depth of one since |
1938 | // there are no compound vectors in OpenCL. | 1904 | // there are no compound vectors in OpenCL. |
@@ -1942,10 +1908,10 @@ index c5ff3ac0..119520ed 100644 | |||
1942 | } | 1908 | } |
1943 | default: | 1909 | default: |
1944 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/KernelArgs.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/KernelArgs.cpp | 1910 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/KernelArgs.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/KernelArgs.cpp |
1945 | index 6448463d..7df71079 100644 | 1911 | index 4a7ce70b4..0178f4652 100644 |
1946 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/KernelArgs.cpp | 1912 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/KernelArgs.cpp |
1947 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/KernelArgs.cpp | 1913 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/KernelArgs.cpp |
1948 | @@ -24,6 +24,7 @@ IN THE SOFTWARE. | 1914 | @@ -8,6 +8,7 @@ SPDX-License-Identifier: MIT |
1949 | 1915 | ||
1950 | #include "Compiler/Optimizer/OpenCLPasses/KernelArgs.hpp" | 1916 | #include "Compiler/Optimizer/OpenCLPasses/KernelArgs.hpp" |
1951 | #include "AdaptorCommon/ImplicitArgs.hpp" | 1917 | #include "AdaptorCommon/ImplicitArgs.hpp" |
@@ -1953,7 +1919,7 @@ index 6448463d..7df71079 100644 | |||
1953 | #include "common/LLVMWarningsPush.hpp" | 1919 | #include "common/LLVMWarningsPush.hpp" |
1954 | #include <llvm/IR/Argument.h> | 1920 | #include <llvm/IR/Argument.h> |
1955 | #include <llvm/IR/DataLayout.h> | 1921 | #include <llvm/IR/DataLayout.h> |
1956 | @@ -388,7 +389,7 @@ unsigned int KernelArg::calcAssociatedArgNo(const ImplicitArg& implicitArg, cons | 1922 | @@ -372,7 +373,7 @@ unsigned int KernelArg::calcAssociatedArgNo(const ImplicitArg& implicitArg, cons |
1957 | 1923 | ||
1958 | unsigned int KernelArg::getNumComponents() const | 1924 | unsigned int KernelArg::getNumComponents() const |
1959 | { | 1925 | { |
@@ -1963,10 +1929,10 @@ index 6448463d..7df71079 100644 | |||
1963 | // Vector | 1929 | // Vector |
1964 | return int_cast<unsigned int>(vecType->getNumElements()); | 1930 | return int_cast<unsigned int>(vecType->getNumElements()); |
1965 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/OpenCLPrintf/OpenCLPrintfResolution.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/OpenCLPrintf/OpenCLPrintfResolution.cpp | 1931 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/OpenCLPrintf/OpenCLPrintfResolution.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/OpenCLPrintf/OpenCLPrintfResolution.cpp |
1966 | index c7414b8e..74964510 100644 | 1932 | index 92c534a10..711fb6b01 100644 |
1967 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/OpenCLPrintf/OpenCLPrintfResolution.cpp | 1933 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/OpenCLPrintf/OpenCLPrintfResolution.cpp |
1968 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/OpenCLPrintf/OpenCLPrintfResolution.cpp | 1934 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/OpenCLPrintf/OpenCLPrintfResolution.cpp |
1969 | @@ -719,7 +719,7 @@ Value* OpenCLPrintfResolution::fixupPrintfArg(CallInst& printfCall, Value* arg, | 1935 | @@ -723,7 +723,7 @@ Value* OpenCLPrintfResolution::fixupPrintfArg(CallInst& printfCall, Value* arg, |
1970 | } | 1936 | } |
1971 | 1937 | ||
1972 | Type* newType = Type::getFloatTy(arg->getContext()); | 1938 | Type* newType = Type::getFloatTy(arg->getContext()); |
@@ -1975,7 +1941,7 @@ index c7414b8e..74964510 100644 | |||
1975 | { | 1941 | { |
1976 | newType = IGCLLVM::FixedVectorType::get(newType, (unsigned)argVT->getNumElements()); | 1942 | newType = IGCLLVM::FixedVectorType::get(newType, (unsigned)argVT->getNumElements()); |
1977 | } | 1943 | } |
1978 | @@ -749,7 +749,7 @@ void OpenCLPrintfResolution::preprocessPrintfArgs(CallInst& printfCall) | 1944 | @@ -753,7 +753,7 @@ void OpenCLPrintfResolution::preprocessPrintfArgs(CallInst& printfCall) |
1979 | IGC::SHADER_PRINTF_TYPE argDataType = getPrintfArgDataType(arg); | 1945 | IGC::SHADER_PRINTF_TYPE argDataType = getPrintfArgDataType(arg); |
1980 | arg = fixupPrintfArg(printfCall, arg, argDataType); | 1946 | arg = fixupPrintfArg(printfCall, arg, argDataType); |
1981 | uint vecSize = 0; | 1947 | uint vecSize = 0; |
@@ -1985,18 +1951,10 @@ index c7414b8e..74964510 100644 | |||
1985 | } | 1951 | } |
1986 | m_argDescriptors.push_back(SPrintfArgDescriptor(argDataType, arg, vecSize)); | 1952 | m_argDescriptors.push_back(SPrintfArgDescriptor(argDataType, arg, vecSize)); |
1987 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp | 1953 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp |
1988 | index 2ed5a2b6..07f85f4c 100644 | 1954 | index ac14b2d04..48c7e8637 100644 |
1989 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp | 1955 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp |
1990 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp | 1956 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/PrivateMemory/PrivateMemoryResolution.cpp |
1991 | @@ -29,6 +29,7 @@ IN THE SOFTWARE. | 1957 | @@ -618,7 +618,7 @@ public: |
1992 | #include "Compiler/IGCPassSupport.h" | ||
1993 | #include "Compiler/CISACodeGen/GenCodeGenModule.h" | ||
1994 | #include "Compiler/CISACodeGen/LowerGEPForPrivMem.hpp" | ||
1995 | +#include "llvmWrapper/IR/DerivedTypes.h" | ||
1996 | #include "common/LLVMWarningsPush.hpp" | ||
1997 | #include "llvm/Transforms/Utils/Local.h" | ||
1998 | #include "llvm/IR/DataLayout.h" | ||
1999 | @@ -604,7 +605,7 @@ public: | ||
2000 | Type* scalarptrTy = PointerType::get(scalarType, pLoad->getPointerAddressSpace()); | 1958 | Type* scalarptrTy = PointerType::get(scalarType, pLoad->getPointerAddressSpace()); |
2001 | IGC_ASSERT(scalarType->getPrimitiveSizeInBits() / 8 == elementSize); | 1959 | IGC_ASSERT(scalarType->getPrimitiveSizeInBits() / 8 == elementSize); |
2002 | Value* vec = UndefValue::get(pLoad->getType()); | 1960 | Value* vec = UndefValue::get(pLoad->getType()); |
@@ -2005,7 +1963,7 @@ index 2ed5a2b6..07f85f4c 100644 | |||
2005 | for (unsigned i = 0, e = (unsigned)pLoadVT->getNumElements(); i < e; ++i) | 1963 | for (unsigned i = 0, e = (unsigned)pLoadVT->getNumElements(); i < e; ++i) |
2006 | { | 1964 | { |
2007 | Value* ptr = IRB.CreateIntToPtr(address, scalarptrTy); | 1965 | Value* ptr = IRB.CreateIntToPtr(address, scalarptrTy); |
2008 | @@ -643,7 +644,7 @@ public: | 1966 | @@ -657,7 +657,7 @@ public: |
2009 | IGC_ASSERT(scalarType->getPrimitiveSizeInBits() / 8 == elementSize); | 1967 | IGC_ASSERT(scalarType->getPrimitiveSizeInBits() / 8 == elementSize); |
2010 | Value* vec = pStore->getValueOperand(); | 1968 | Value* vec = pStore->getValueOperand(); |
2011 | 1969 | ||
@@ -2014,7 +1972,7 @@ index 2ed5a2b6..07f85f4c 100644 | |||
2014 | for (unsigned i = 0; i < vecNumElts; ++i) | 1972 | for (unsigned i = 0; i < vecNumElts; ++i) |
2015 | { | 1973 | { |
2016 | Value* ptr = IRB.CreateIntToPtr(address, scalarptrTy); | 1974 | Value* ptr = IRB.CreateIntToPtr(address, scalarptrTy); |
2017 | @@ -707,7 +708,7 @@ bool PrivateMemoryResolution::testTransposedMemory(const Type* pTmpType, const T | 1975 | @@ -728,7 +728,7 @@ bool PrivateMemoryResolution::testTransposedMemory(const Type* pTmpType, const T |
2018 | } | 1976 | } |
2019 | else if(pTmpType->isVectorTy()) | 1977 | else if(pTmpType->isVectorTy()) |
2020 | { | 1978 | { |
@@ -2023,23 +1981,11 @@ index 2ed5a2b6..07f85f4c 100644 | |||
2023 | tmpAllocaSize *= pTmpVType->getNumElements(); | 1981 | tmpAllocaSize *= pTmpVType->getNumElements(); |
2024 | pTmpType = pTmpType->getContainedType(0); | 1982 | pTmpType = pTmpType->getContainedType(0); |
2025 | ok = (nullptr != pTmpType); | 1983 | ok = (nullptr != pTmpType); |
2026 | @@ -815,7 +816,11 @@ bool PrivateMemoryResolution::resolveAllocaInstructions(bool privateOnStack) | ||
2027 | // Construct an empty DebugLoc. | ||
2028 | IF_DEBUG_INFO(DebugLoc entryDebugLoc); | ||
2029 | // Assign with the function location if available. | ||
2030 | +#if LLVM_VERSION_MAJOR >= 12 | ||
2031 | + IF_DEBUG_INFO_IF(DISubprogram *subprogram = m_currFunction->getSubprogram(), entryDebugLoc = DILocation::get(subprogram->getContext(), subprogram->getLine(), 0, subprogram);); | ||
2032 | +#else | ||
2033 | IF_DEBUG_INFO_IF(DISubprogram *subprogram = m_currFunction->getSubprogram(), entryDebugLoc = DebugLoc::get(subprogram->getLine(), 0, subprogram);); | ||
2034 | +#endif | ||
2035 | IF_DEBUG_INFO(entryBuilder.SetCurrentDebugLocation(entryDebugLoc)); | ||
2036 | |||
2037 | if (privateOnStack) | ||
2038 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/ReplaceUnsupportedIntrinsics/ReplaceUnsupportedIntrinsics.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/ReplaceUnsupportedIntrinsics/ReplaceUnsupportedIntrinsics.cpp | 1984 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/ReplaceUnsupportedIntrinsics/ReplaceUnsupportedIntrinsics.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/ReplaceUnsupportedIntrinsics/ReplaceUnsupportedIntrinsics.cpp |
2039 | index c8414efa..68491d48 100644 | 1985 | index 88707c588..ae9c7e671 100644 |
2040 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/ReplaceUnsupportedIntrinsics/ReplaceUnsupportedIntrinsics.cpp | 1986 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/ReplaceUnsupportedIntrinsics/ReplaceUnsupportedIntrinsics.cpp |
2041 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/ReplaceUnsupportedIntrinsics/ReplaceUnsupportedIntrinsics.cpp | 1987 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/ReplaceUnsupportedIntrinsics/ReplaceUnsupportedIntrinsics.cpp |
2042 | @@ -254,7 +254,7 @@ Instruction* ReplaceUnsupportedIntrinsics::insertLoop(Instruction* Loc, Value* L | 1988 | @@ -244,7 +244,7 @@ Instruction* ReplaceUnsupportedIntrinsics::insertLoop(Instruction* Loc, Value* L |
2043 | Value* ReplaceUnsupportedIntrinsics::replicateScalar( | 1989 | Value* ReplaceUnsupportedIntrinsics::replicateScalar( |
2044 | Value* ScalarVal, Type* Ty, Instruction* InsertBefore) | 1990 | Value* ScalarVal, Type* Ty, Instruction* InsertBefore) |
2045 | { | 1991 | { |
@@ -2048,7 +1994,7 @@ index c8414efa..68491d48 100644 | |||
2048 | Type* ETy = VTy ? VTy->getElementType() : Ty; | 1994 | Type* ETy = VTy ? VTy->getElementType() : Ty; |
2049 | uint32_t sBits = (unsigned int)ScalarVal->getType()->getPrimitiveSizeInBits(); | 1995 | uint32_t sBits = (unsigned int)ScalarVal->getType()->getPrimitiveSizeInBits(); |
2050 | uint32_t nBits = (unsigned int)ETy->getPrimitiveSizeInBits(); | 1996 | uint32_t nBits = (unsigned int)ETy->getPrimitiveSizeInBits(); |
2051 | @@ -868,7 +868,7 @@ void ReplaceUnsupportedIntrinsics::replaceFunnelShift(IntrinsicInst* I) { | 1997 | @@ -905,7 +905,7 @@ void ReplaceUnsupportedIntrinsics::replaceFunnelShift(IntrinsicInst* I) { |
2052 | } | 1998 | } |
2053 | 1999 | ||
2054 | Value* numBits = Builder.getIntN(sizeInBits, sizeInBits); | 2000 | Value* numBits = Builder.getIntN(sizeInBits, sizeInBits); |
@@ -2058,26 +2004,23 @@ index c8414efa..68491d48 100644 | |||
2058 | } | 2004 | } |
2059 | auto shiftModulo = Builder.CreateURem(I->getArgOperand(2), numBits); | 2005 | auto shiftModulo = Builder.CreateURem(I->getArgOperand(2), numBits); |
2060 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp | 2006 | diff --git a/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp b/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp |
2061 | index 7188cc50..535d6268 100644 | 2007 | index a8e24680c..4de05effa 100644 |
2062 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp | 2008 | --- a/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp |
2063 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp | 2009 | +++ b/IGC/Compiler/Optimizer/OpenCLPasses/WIFuncs/WIFuncResolution.cpp |
2064 | @@ -303,7 +303,11 @@ static Value* BuildLoadInst(CallInst& CI, unsigned int Offset, Type* DataType) | 2010 | @@ -279,7 +279,7 @@ static Value* BuildLoadInst(CallInst& CI, unsigned int Offset, Type* DataType) |
2065 | auto Size = ElemByteSize; | 2011 | auto Size = ElemByteSize; |
2066 | if (DataType->isVectorTy()) | 2012 | if (auto DataVecType = dyn_cast<VectorType>(DataType)) |
2067 | { | 2013 | { |
2068 | +#if LLVM_VERSION_MAJOR >= 12 | 2014 | - Size *= (unsigned int)DataVecType->getNumElements(); |
2069 | + Size *= cast<IGCLLVM::FixedVectorType>(DataType)->getNumElements(); | 2015 | + Size *= cast<IGCLLVM::FixedVectorType>(DataType)->getNumElements(); |
2070 | +#else | ||
2071 | Size *= DataType->getVectorNumElements(); | ||
2072 | +#endif | ||
2073 | } | 2016 | } |
2074 | unsigned int AlignedOffset = (Offset / ElemByteSize) * ElemByteSize; | 2017 | unsigned int AlignedOffset = (Offset / ElemByteSize) * ElemByteSize; |
2075 | unsigned int LoadByteSize = (Offset == AlignedOffset) ? Size : Size * 2; | 2018 | unsigned int LoadByteSize = (Offset == AlignedOffset) ? Size : Size * 2; |
2076 | diff --git a/IGC/Compiler/Optimizer/PreCompiledFuncImport.cpp b/IGC/Compiler/Optimizer/PreCompiledFuncImport.cpp | 2019 | diff --git a/IGC/Compiler/Optimizer/PreCompiledFuncImport.cpp b/IGC/Compiler/Optimizer/PreCompiledFuncImport.cpp |
2077 | index 1fda21d5..63a22864 100644 | 2020 | index 248ab1039..423beef7b 100644 |
2078 | --- a/IGC/Compiler/Optimizer/PreCompiledFuncImport.cpp | 2021 | --- a/IGC/Compiler/Optimizer/PreCompiledFuncImport.cpp |
2079 | +++ b/IGC/Compiler/Optimizer/PreCompiledFuncImport.cpp | 2022 | +++ b/IGC/Compiler/Optimizer/PreCompiledFuncImport.cpp |
2080 | @@ -340,7 +340,7 @@ bool PreCompiledFuncImport::preProcessDouble() | 2023 | @@ -339,7 +339,7 @@ bool PreCompiledFuncImport::preProcessDouble() |
2081 | } | 2024 | } |
2082 | else | 2025 | else |
2083 | { | 2026 | { |
@@ -2086,7 +2029,7 @@ index 1fda21d5..63a22864 100644 | |||
2086 | fsub = llvm::UndefValue::get(Inst->getType()); | 2029 | fsub = llvm::UndefValue::get(Inst->getType()); |
2087 | 2030 | ||
2088 | for (uint32_t i = 0; i < vectorSize; ++i) | 2031 | for (uint32_t i = 0; i < vectorSize; ++i) |
2089 | @@ -936,7 +936,7 @@ void PreCompiledFuncImport::processDivide(BinaryOperator& inst, EmulatedFunction | 2032 | @@ -1002,7 +1002,7 @@ void PreCompiledFuncImport::processDivide(BinaryOperator& inst, EmulatedFunction |
2090 | 2033 | ||
2091 | Type* argumentType = inst.getOperand(0)->getType(); | 2034 | Type* argumentType = inst.getOperand(0)->getType(); |
2092 | 2035 | ||
@@ -2096,10 +2039,10 @@ index 1fda21d5..63a22864 100644 | |||
2096 | numElements = (unsigned)argumentVType->getNumElements(); | 2039 | numElements = (unsigned)argumentVType->getNumElements(); |
2097 | } | 2040 | } |
2098 | diff --git a/IGC/Compiler/Optimizer/Scalarizer.cpp b/IGC/Compiler/Optimizer/Scalarizer.cpp | 2041 | diff --git a/IGC/Compiler/Optimizer/Scalarizer.cpp b/IGC/Compiler/Optimizer/Scalarizer.cpp |
2099 | index 4f59eaf9..a4e73a6d 100644 | 2042 | index 740fb4d57..b98ba6693 100644 |
2100 | --- a/IGC/Compiler/Optimizer/Scalarizer.cpp | 2043 | --- a/IGC/Compiler/Optimizer/Scalarizer.cpp |
2101 | +++ b/IGC/Compiler/Optimizer/Scalarizer.cpp | 2044 | +++ b/IGC/Compiler/Optimizer/Scalarizer.cpp |
2102 | @@ -289,7 +289,7 @@ void ScalarizeFunction::scalarizeInstruction(BinaryOperator* BI) | 2045 | @@ -403,7 +403,7 @@ void ScalarizeFunction::scalarizeInstruction(BinaryOperator* BI) |
2103 | { | 2046 | { |
2104 | V_PRINT(scalarizer, "\t\tBinary instruction\n"); | 2047 | V_PRINT(scalarizer, "\t\tBinary instruction\n"); |
2105 | IGC_ASSERT_MESSAGE(BI, "instruction type dynamic cast failed"); | 2048 | IGC_ASSERT_MESSAGE(BI, "instruction type dynamic cast failed"); |
@@ -2108,7 +2051,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2108 | // Only need handling for vector binary ops | 2051 | // Only need handling for vector binary ops |
2109 | if (!instType) return; | 2052 | if (!instType) return; |
2110 | 2053 | ||
2111 | @@ -349,7 +349,7 @@ void ScalarizeFunction::scalarizeInstruction(CmpInst* CI) | 2054 | @@ -463,7 +463,7 @@ void ScalarizeFunction::scalarizeInstruction(CmpInst* CI) |
2112 | { | 2055 | { |
2113 | V_PRINT(scalarizer, "\t\tCompare instruction\n"); | 2056 | V_PRINT(scalarizer, "\t\tCompare instruction\n"); |
2114 | IGC_ASSERT_MESSAGE(CI, "instruction type dynamic cast failed"); | 2057 | IGC_ASSERT_MESSAGE(CI, "instruction type dynamic cast failed"); |
@@ -2117,7 +2060,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2117 | // Only need handling for vector compares | 2060 | // Only need handling for vector compares |
2118 | if (!instType) return; | 2061 | if (!instType) return; |
2119 | 2062 | ||
2120 | @@ -397,13 +397,13 @@ void ScalarizeFunction::scalarizeInstruction(CastInst* CI) | 2063 | @@ -511,13 +511,13 @@ void ScalarizeFunction::scalarizeInstruction(CastInst* CI) |
2121 | { | 2064 | { |
2122 | V_PRINT(scalarizer, "\t\tCast instruction\n"); | 2065 | V_PRINT(scalarizer, "\t\tCast instruction\n"); |
2123 | IGC_ASSERT_MESSAGE(CI, "instruction type dynamic cast failed"); | 2066 | IGC_ASSERT_MESSAGE(CI, "instruction type dynamic cast failed"); |
@@ -2133,7 +2076,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2133 | if (!srcType || (instType->getNumElements() != srcType->getNumElements())) | 2076 | if (!srcType || (instType->getNumElements() != srcType->getNumElements())) |
2134 | { | 2077 | { |
2135 | return recoverNonScalarizableInst(CI); | 2078 | return recoverNonScalarizableInst(CI); |
2136 | @@ -458,7 +458,7 @@ void ScalarizeFunction::scalarizeInstruction(PHINode* PI) | 2079 | @@ -572,7 +572,7 @@ void ScalarizeFunction::scalarizeInstruction(PHINode* PI) |
2137 | { | 2080 | { |
2138 | V_PRINT(scalarizer, "\t\tPHI instruction\n"); | 2081 | V_PRINT(scalarizer, "\t\tPHI instruction\n"); |
2139 | IGC_ASSERT_MESSAGE(PI, "instruction type dynamic cast failed"); | 2082 | IGC_ASSERT_MESSAGE(PI, "instruction type dynamic cast failed"); |
@@ -2142,7 +2085,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2142 | // Only need handling for vector PHI | 2085 | // Only need handling for vector PHI |
2143 | if (!instType) return; | 2086 | if (!instType) return; |
2144 | 2087 | ||
2145 | @@ -569,7 +569,7 @@ void ScalarizeFunction::scalarizeInstruction(SelectInst* SI) | 2088 | @@ -683,7 +683,7 @@ void ScalarizeFunction::scalarizeInstruction(SelectInst* SI) |
2146 | { | 2089 | { |
2147 | V_PRINT(scalarizer, "\t\tSelect instruction\n"); | 2090 | V_PRINT(scalarizer, "\t\tSelect instruction\n"); |
2148 | IGC_ASSERT_MESSAGE(SI, "instruction type dynamic cast failed"); | 2091 | IGC_ASSERT_MESSAGE(SI, "instruction type dynamic cast failed"); |
@@ -2151,7 +2094,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2151 | // Only need handling for vector select | 2094 | // Only need handling for vector select |
2152 | if (!instType) return; | 2095 | if (!instType) return; |
2153 | 2096 | ||
2154 | @@ -651,7 +651,7 @@ void ScalarizeFunction::scalarizeInstruction(ExtractElementInst* EI) | 2097 | @@ -765,7 +765,7 @@ void ScalarizeFunction::scalarizeInstruction(ExtractElementInst* EI) |
2155 | 2098 | ||
2156 | // Connect the "extracted" value to all its consumers | 2099 | // Connect the "extracted" value to all its consumers |
2157 | uint64_t scalarIndex = cast<ConstantInt>(scalarIndexVal)->getZExtValue(); | 2100 | uint64_t scalarIndex = cast<ConstantInt>(scalarIndexVal)->getZExtValue(); |
@@ -2160,7 +2103,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2160 | if (static_cast<unsigned int>(scalarIndex) < (unsigned)valueVType->getNumElements()) | 2103 | if (static_cast<unsigned int>(scalarIndex) < (unsigned)valueVType->getNumElements()) |
2161 | { | 2104 | { |
2162 | IGC_ASSERT_MESSAGE(NULL != operand[static_cast<unsigned int>(scalarIndex)], "SCM error"); | 2105 | IGC_ASSERT_MESSAGE(NULL != operand[static_cast<unsigned int>(scalarIndex)], "SCM error"); |
2163 | @@ -715,7 +715,7 @@ void ScalarizeFunction::scalarizeInstruction(InsertElementInst* II) | 2106 | @@ -829,7 +829,7 @@ void ScalarizeFunction::scalarizeInstruction(InsertElementInst* II) |
2164 | if (isa<UndefValue>(sourceVectorValue)) | 2107 | if (isa<UndefValue>(sourceVectorValue)) |
2165 | { | 2108 | { |
2166 | // Scalarize the undef value (generate a scalar undef) | 2109 | // Scalarize the undef value (generate a scalar undef) |
@@ -2169,7 +2112,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2169 | IGC_ASSERT_MESSAGE(inputVectorType, "expected vector argument"); | 2112 | IGC_ASSERT_MESSAGE(inputVectorType, "expected vector argument"); |
2170 | 2113 | ||
2171 | UndefValue* undefVal = UndefValue::get(inputVectorType->getElementType()); | 2114 | UndefValue* undefVal = UndefValue::get(inputVectorType->getElementType()); |
2172 | @@ -756,7 +756,7 @@ void ScalarizeFunction::scalarizeInstruction(ShuffleVectorInst* SI) | 2115 | @@ -870,7 +870,7 @@ void ScalarizeFunction::scalarizeInstruction(ShuffleVectorInst* SI) |
2173 | IGC_ASSERT(nullptr != sourceVector0Value); | 2116 | IGC_ASSERT(nullptr != sourceVector0Value); |
2174 | Value* sourceVector1Value = SI->getOperand(1); | 2117 | Value* sourceVector1Value = SI->getOperand(1); |
2175 | IGC_ASSERT(nullptr != sourceVector1Value); | 2118 | IGC_ASSERT(nullptr != sourceVector1Value); |
@@ -2178,16 +2121,16 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2178 | IGC_ASSERT_MESSAGE(nullptr != inputType, "vector input error"); | 2121 | IGC_ASSERT_MESSAGE(nullptr != inputType, "vector input error"); |
2179 | IGC_ASSERT_MESSAGE(inputType == sourceVector1Value->getType(), "vector input error"); | 2122 | IGC_ASSERT_MESSAGE(inputType == sourceVector1Value->getType(), "vector input error"); |
2180 | unsigned sourceVectorWidth = int_cast<unsigned>(inputType->getNumElements()); | 2123 | unsigned sourceVectorWidth = int_cast<unsigned>(inputType->getNumElements()); |
2181 | @@ -778,7 +778,7 @@ void ScalarizeFunction::scalarizeInstruction(ShuffleVectorInst* SI) | 2124 | @@ -892,7 +892,7 @@ void ScalarizeFunction::scalarizeInstruction(ShuffleVectorInst* SI) |
2182 | 2125 | ||
2183 | // Generate array for shuffled scalar values | 2126 | // Generate array for shuffled scalar values |
2184 | SmallVector<Value*, MAX_INPUT_VECTOR_WIDTH>newVector; | 2127 | SmallVector<Value*, MAX_INPUT_VECTOR_WIDTH>newVector; |
2185 | - unsigned width = int_cast<unsigned>(SI->getType()->getNumElements()); | 2128 | - unsigned width = int_cast<unsigned>(SI->getType()->getNumElements()); |
2186 | + unsigned width = int_cast<unsigned>(dyn_cast<IGCLLVM::FixedVectorType>(SI->getType())->getNumElements()); | 2129 | + unsigned width = int_cast<unsigned>(cast<IGCLLVM::FixedVectorType>(SI->getType())->getNumElements()); |
2187 | 2130 | ||
2188 | // Generate undef value, which may be needed as some scalar elements | 2131 | // Generate undef value, which may be needed as some scalar elements |
2189 | UndefValue* undef = UndefValue::get(inputType->getElementType()); | 2132 | UndefValue* undef = UndefValue::get(inputType->getElementType()); |
2190 | @@ -847,7 +847,7 @@ void ScalarizeFunction::scalarizeInstruction(GetElementPtrInst* GI) | 2133 | @@ -961,7 +961,7 @@ void ScalarizeFunction::scalarizeInstruction(GetElementPtrInst* GI) |
2191 | 2134 | ||
2192 | if (baseValue->getType()->isVectorTy()) | 2135 | if (baseValue->getType()->isVectorTy()) |
2193 | { | 2136 | { |
@@ -2196,7 +2139,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2196 | // Obtain the scalarized operands | 2139 | // Obtain the scalarized operands |
2197 | obtainScalarizedValues(operand1, NULL, baseValue, GI); | 2140 | obtainScalarizedValues(operand1, NULL, baseValue, GI); |
2198 | ptrTy = dyn_cast<VectorType>(baseValue->getType())->getElementType(); | 2141 | ptrTy = dyn_cast<VectorType>(baseValue->getType())->getElementType(); |
2199 | @@ -858,7 +858,7 @@ void ScalarizeFunction::scalarizeInstruction(GetElementPtrInst* GI) | 2142 | @@ -972,7 +972,7 @@ void ScalarizeFunction::scalarizeInstruction(GetElementPtrInst* GI) |
2200 | } | 2143 | } |
2201 | if (indexValue->getType()->isVectorTy()) | 2144 | if (indexValue->getType()->isVectorTy()) |
2202 | { | 2145 | { |
@@ -2205,25 +2148,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2205 | // Obtain the scalarized operands | 2148 | // Obtain the scalarized operands |
2206 | obtainScalarizedValues(operand2, NULL, indexValue, GI); | 2149 | obtainScalarizedValues(operand2, NULL, indexValue, GI); |
2207 | } | 2150 | } |
2208 | @@ -897,7 +897,7 @@ void ScalarizeFunction::scalarizeInstruction(LoadInst* LI) | 2151 | @@ -1011,7 +1011,7 @@ void ScalarizeFunction::obtainScalarizedValues(SmallVectorImpl<Value*>& retValue |
2209 | V_PRINT(scalarizer, "\t\tLoad instruction\n"); | ||
2210 | IGC_ASSERT_MESSAGE(LI, "instruction type dynamic cast failed"); | ||
2211 | |||
2212 | - VectorType* dataType = dyn_cast<VectorType>(LI->getType()); | ||
2213 | + IGCLLVM::FixedVectorType* dataType = dyn_cast<IGCLLVM::FixedVectorType>(LI->getType()); | ||
2214 | if (isScalarizableLoadStoreType(dataType) && m_pDL) | ||
2215 | { | ||
2216 | // Prepare empty SCM entry for the instruction | ||
2217 | @@ -969,7 +969,7 @@ void ScalarizeFunction::scalarizeInstruction(StoreInst* SI) | ||
2218 | |||
2219 | int indexPtr = SI->getPointerOperandIndex(); | ||
2220 | int indexData = 1 - indexPtr; | ||
2221 | - VectorType* dataType = dyn_cast<VectorType>(SI->getOperand(indexData)->getType()); | ||
2222 | + IGCLLVM::FixedVectorType* dataType = dyn_cast<IGCLLVM::FixedVectorType>(SI->getOperand(indexData)->getType()); | ||
2223 | if (isScalarizableLoadStoreType(dataType) && m_pDL) | ||
2224 | { | ||
2225 | // Get additional info from instruction | ||
2226 | @@ -1040,7 +1040,7 @@ void ScalarizeFunction::obtainScalarizedValues(SmallVectorImpl<Value*>& retValue | ||
2227 | { | 2152 | { |
2228 | V_PRINT(scalarizer, "\t\t\tObtaining scalar value... " << *origValue << "\n"); | 2153 | V_PRINT(scalarizer, "\t\t\tObtaining scalar value... " << *origValue << "\n"); |
2229 | 2154 | ||
@@ -2232,7 +2157,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2232 | IGC_ASSERT_MESSAGE(origType, "Value must have a vector type!"); | 2157 | IGC_ASSERT_MESSAGE(origType, "Value must have a vector type!"); |
2233 | unsigned width = int_cast<unsigned>(origType->getNumElements()); | 2158 | unsigned width = int_cast<unsigned>(origType->getNumElements()); |
2234 | 2159 | ||
2235 | @@ -1202,7 +1202,7 @@ void ScalarizeFunction::obtainVectorValueWhichMightBeScalarizedImpl(Value* vecto | 2160 | @@ -1170,7 +1170,7 @@ void ScalarizeFunction::obtainVectorValueWhichMightBeScalarizedImpl(Value* vecto |
2236 | } | 2161 | } |
2237 | 2162 | ||
2238 | Value* assembledVector = UndefValue::get(vectorVal->getType()); | 2163 | Value* assembledVector = UndefValue::get(vectorVal->getType()); |
@@ -2241,7 +2166,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2241 | for (unsigned i = 0; i < width; i++) | 2166 | for (unsigned i = 0; i < width; i++) |
2242 | { | 2167 | { |
2243 | IGC_ASSERT_MESSAGE(NULL != valueEntry->scalarValues[i], "SCM entry has NULL value"); | 2168 | IGC_ASSERT_MESSAGE(NULL != valueEntry->scalarValues[i], "SCM entry has NULL value"); |
2244 | @@ -1262,7 +1262,7 @@ void ScalarizeFunction::updateSCMEntryWithValues(ScalarizeFunction::SCMEntry* en | 2169 | @@ -1230,7 +1230,7 @@ void ScalarizeFunction::updateSCMEntryWithValues(ScalarizeFunction::SCMEntry* en |
2245 | bool matchDbgLoc) | 2170 | bool matchDbgLoc) |
2246 | { | 2171 | { |
2247 | IGC_ASSERT_MESSAGE((origValue->getType()->isArrayTy() || origValue->getType()->isVectorTy()), "only Vector values are supported"); | 2172 | IGC_ASSERT_MESSAGE((origValue->getType()->isArrayTy() || origValue->getType()->isVectorTy()), "only Vector values are supported"); |
@@ -2250,7 +2175,7 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2250 | 2175 | ||
2251 | entry->isOriginalVectorRemoved = isOrigValueRemoved; | 2176 | entry->isOriginalVectorRemoved = isOrigValueRemoved; |
2252 | 2177 | ||
2253 | @@ -1326,7 +1326,7 @@ void ScalarizeFunction::resolveDeferredInstructions() | 2178 | @@ -1301,7 +1301,7 @@ void ScalarizeFunction::resolveDeferredInstructions() |
2254 | Instruction* vectorInst = dyn_cast<Instruction>(current.unresolvedInst); | 2179 | Instruction* vectorInst = dyn_cast<Instruction>(current.unresolvedInst); |
2255 | IGC_ASSERT_MESSAGE(vectorInst, "DRL only handles unresolved instructions"); | 2180 | IGC_ASSERT_MESSAGE(vectorInst, "DRL only handles unresolved instructions"); |
2256 | 2181 | ||
@@ -2260,10 +2185,10 @@ index 4f59eaf9..a4e73a6d 100644 | |||
2260 | unsigned width = int_cast<unsigned>(currType->getNumElements()); | 2185 | unsigned width = int_cast<unsigned>(currType->getNumElements()); |
2261 | 2186 | ||
2262 | diff --git a/IGC/Compiler/Optimizer/ValueTracker.cpp b/IGC/Compiler/Optimizer/ValueTracker.cpp | 2187 | diff --git a/IGC/Compiler/Optimizer/ValueTracker.cpp b/IGC/Compiler/Optimizer/ValueTracker.cpp |
2263 | index c06f7312..d8f0d385 100644 | 2188 | index a2796daa5..a3d03d267 100644 |
2264 | --- a/IGC/Compiler/Optimizer/ValueTracker.cpp | 2189 | --- a/IGC/Compiler/Optimizer/ValueTracker.cpp |
2265 | +++ b/IGC/Compiler/Optimizer/ValueTracker.cpp | 2190 | +++ b/IGC/Compiler/Optimizer/ValueTracker.cpp |
2266 | @@ -178,7 +178,7 @@ Value* ValueTracker::handleExtractElement(ExtractElementInst* E) | 2191 | @@ -162,7 +162,7 @@ Value* ValueTracker::handleExtractElement(ExtractElementInst* E) |
2267 | { | 2192 | { |
2268 | auto mask = I->getShuffleMask(); | 2193 | auto mask = I->getShuffleMask(); |
2269 | uint shuffleidx = int_cast<uint>(mask[(uint)idx]); | 2194 | uint shuffleidx = int_cast<uint>(mask[(uint)idx]); |
@@ -2273,10 +2198,10 @@ index c06f7312..d8f0d385 100644 | |||
2273 | I->getOperand(0) : I->getOperand(1); | 2198 | I->getOperand(0) : I->getOperand(1); |
2274 | } | 2199 | } |
2275 | diff --git a/IGC/Compiler/VerificationPass.cpp b/IGC/Compiler/VerificationPass.cpp | 2200 | diff --git a/IGC/Compiler/VerificationPass.cpp b/IGC/Compiler/VerificationPass.cpp |
2276 | index 98437d79..40d50f5a 100644 | 2201 | index 7a9d69bec..df73419c0 100644 |
2277 | --- a/IGC/Compiler/VerificationPass.cpp | 2202 | --- a/IGC/Compiler/VerificationPass.cpp |
2278 | +++ b/IGC/Compiler/VerificationPass.cpp | 2203 | +++ b/IGC/Compiler/VerificationPass.cpp |
2279 | @@ -302,7 +302,7 @@ bool VerificationPass::verifyType(Type* type, Value* val) | 2204 | @@ -286,7 +286,7 @@ bool VerificationPass::verifyType(Type* type, Value* val) |
2280 | 2205 | ||
2281 | case IGCLLVM::VectorTyID: | 2206 | case IGCLLVM::VectorTyID: |
2282 | { | 2207 | { |
@@ -2286,48 +2211,39 @@ index 98437d79..40d50f5a 100644 | |||
2286 | if (!m_IGC_IR_spec.vectorTypeSizes.count(typeSize)) | 2211 | if (!m_IGC_IR_spec.vectorTypeSizes.count(typeSize)) |
2287 | { | 2212 | { |
2288 | diff --git a/IGC/DebugInfo/DebugInfoUtils.hpp b/IGC/DebugInfo/DebugInfoUtils.hpp | 2213 | diff --git a/IGC/DebugInfo/DebugInfoUtils.hpp b/IGC/DebugInfo/DebugInfoUtils.hpp |
2289 | index 7f2e288a..b77a550d 100644 | 2214 | index 28b7fcf73..d16d79a3d 100644 |
2290 | --- a/IGC/DebugInfo/DebugInfoUtils.hpp | 2215 | --- a/IGC/DebugInfo/DebugInfoUtils.hpp |
2291 | +++ b/IGC/DebugInfo/DebugInfoUtils.hpp | 2216 | +++ b/IGC/DebugInfo/DebugInfoUtils.hpp |
2292 | @@ -108,7 +108,11 @@ namespace IGC | 2217 | @@ -92,7 +92,7 @@ namespace IGC |
2293 | IGCLLVM::DIBuilder Builder(M); | 2218 | IGCLLVM::DIBuilder Builder(M); |
2294 | llvm::DIGlobalVariable* GV = GVs[j]->getVariable(); | 2219 | llvm::DIGlobalVariable* GV = GVs[j]->getVariable(); |
2295 | llvm::DIScope* scopeToUse = GV->getScope(); | 2220 | llvm::DIScope* scopeToUse = GV->getScope(); |
2296 | +#if LLVM_VERSION_MAJOR >= 12 | 2221 | - llvm::DILocation* locToUse = llvm::DebugLoc::get(GV->getLine(), 0, scopeToUse, loc); |
2297 | + llvm::DILocation* locToUse = llvm::DILocation::get(scopeToUse->getContext(), GV->getLine(), 0, scopeToUse, loc); | 2222 | + llvm::DILocation* locToUse = llvm::DILocation::get(scopeToUse->getContext(), GV->getLine(), 0, scopeToUse, loc); |
2298 | +#else | ||
2299 | llvm::DILocation* locToUse = llvm::DebugLoc::get(GV->getLine(), 0, scopeToUse, loc); | ||
2300 | +#endif | ||
2301 | if (llvm::isa<llvm::DICompileUnit>(GV->getScope())) | 2223 | if (llvm::isa<llvm::DICompileUnit>(GV->getScope())) |
2302 | { | 2224 | { |
2303 | // Function has no DebugLoc so it is either internal | 2225 | // Function has no DebugLoc so it is either internal |
2304 | diff --git a/IGC/DebugInfo/DwarfDebug.cpp b/IGC/DebugInfo/DwarfDebug.cpp | 2226 | diff --git a/IGC/DebugInfo/DwarfDebug.cpp b/IGC/DebugInfo/DwarfDebug.cpp |
2305 | index 683b83f7..bd9f17b7 100644 | 2227 | index 027a3a530..f57dd72cb 100644 |
2306 | --- a/IGC/DebugInfo/DwarfDebug.cpp | 2228 | --- a/IGC/DebugInfo/DwarfDebug.cpp |
2307 | +++ b/IGC/DebugInfo/DwarfDebug.cpp | 2229 | +++ b/IGC/DebugInfo/DwarfDebug.cpp |
2308 | @@ -2102,9 +2102,17 @@ static DebugLoc getFnDebugLoc(DebugLoc DL, const LLVMContext& Ctx) | 2230 | @@ -1993,9 +1993,9 @@ static DebugLoc getFnDebugLoc(DebugLoc DL, const LLVMContext& Ctx) |
2309 | // Check for number of operands since the compatibility is cheap here. | 2231 | // Check for number of operands since the compatibility is cheap here. |
2310 | if (SP->getNumOperands() > 19) | 2232 | if (SP->getNumOperands() > 19) |
2311 | { | 2233 | { |
2312 | +#if LLVM_VERSION_MAJOR >= 12 | 2234 | - return DebugLoc::get(SP->getScopeLine(), 0, SP); |
2313 | + return DILocation::get(SP->getContext(), SP->getScopeLine(), 0, SP); | 2235 | + return DILocation::get(SP->getContext(), SP->getScopeLine(), 0, SP); |
2314 | +#else | ||
2315 | return DebugLoc::get(SP->getScopeLine(), 0, SP); | ||
2316 | +#endif | ||
2317 | } | 2236 | } |
2318 | +#if LLVM_VERSION_MAJOR >= 12 | 2237 | - return DebugLoc::get(SP->getLine(), 0, SP); |
2319 | + return DILocation::get(SP->getContext(), SP->getLine(), 0, SP); | 2238 | + return DILocation::get(SP->getContext(), SP->getLine(), 0, SP); |
2320 | +#else | ||
2321 | return DebugLoc::get(SP->getLine(), 0, SP); | ||
2322 | +#endif | ||
2323 | } | 2239 | } |
2324 | 2240 | ||
2325 | return DebugLoc(); | 2241 | return DebugLoc(); |
2326 | diff --git a/IGC/GenISAIntrinsics/GenIntrinsics.cpp b/IGC/GenISAIntrinsics/GenIntrinsics.cpp | 2242 | diff --git a/IGC/GenISAIntrinsics/GenIntrinsics.cpp b/IGC/GenISAIntrinsics/GenIntrinsics.cpp |
2327 | index 708cfdaa..772155b2 100644 | 2243 | index a6d6f0ae6..c48271125 100644 |
2328 | --- a/IGC/GenISAIntrinsics/GenIntrinsics.cpp | 2244 | --- a/IGC/GenISAIntrinsics/GenIntrinsics.cpp |
2329 | +++ b/IGC/GenISAIntrinsics/GenIntrinsics.cpp | 2245 | +++ b/IGC/GenISAIntrinsics/GenIntrinsics.cpp |
2330 | @@ -204,7 +204,7 @@ static Type *DecodeFixedType(ArrayRef<GenISAIntrinsic::IITDescriptor> &Infos, | 2246 | @@ -188,7 +188,7 @@ static Type *DecodeFixedType(ArrayRef<GenISAIntrinsic::IITDescriptor> &Infos, |
2331 | case IITDescriptor::SameVecWidthArgument: { | 2247 | case IITDescriptor::SameVecWidthArgument: { |
2332 | Type *EltTy = DecodeFixedType(Infos, Tys, Context); | 2248 | Type *EltTy = DecodeFixedType(Infos, Tys, Context); |
2333 | Type *Ty = Tys[D.getArgumentNumber()]; | 2249 | Type *Ty = Tys[D.getArgumentNumber()]; |
@@ -2336,7 +2252,7 @@ index 708cfdaa..772155b2 100644 | |||
2336 | return IGCLLVM::FixedVectorType::get(EltTy, int_cast<unsigned int>(VTy->getNumElements())); | 2252 | return IGCLLVM::FixedVectorType::get(EltTy, int_cast<unsigned int>(VTy->getNumElements())); |
2337 | } | 2253 | } |
2338 | IGC_ASSERT_EXIT_MESSAGE(0, "unhandled"); | 2254 | IGC_ASSERT_EXIT_MESSAGE(0, "unhandled"); |
2339 | @@ -215,7 +215,7 @@ static Type *DecodeFixedType(ArrayRef<GenISAIntrinsic::IITDescriptor> &Infos, | 2255 | @@ -199,7 +199,7 @@ static Type *DecodeFixedType(ArrayRef<GenISAIntrinsic::IITDescriptor> &Infos, |
2340 | } | 2256 | } |
2341 | case IITDescriptor::VecOfPtrsToElt: { | 2257 | case IITDescriptor::VecOfPtrsToElt: { |
2342 | Type *Ty = Tys[D.getArgumentNumber()]; | 2258 | Type *Ty = Tys[D.getArgumentNumber()]; |
@@ -2345,7 +2261,7 @@ index 708cfdaa..772155b2 100644 | |||
2345 | if (!VTy) | 2261 | if (!VTy) |
2346 | IGC_ASSERT_EXIT_MESSAGE(0, "Expected an argument of Vector Type"); | 2262 | IGC_ASSERT_EXIT_MESSAGE(0, "Expected an argument of Vector Type"); |
2347 | Type *EltTy = cast<VectorType>(VTy)->getElementType(); | 2263 | Type *EltTy = cast<VectorType>(VTy)->getElementType(); |
2348 | @@ -434,7 +434,7 @@ static std::string getMangledTypeStr(Type* Ty) { | 2264 | @@ -418,7 +418,7 @@ static std::string getMangledTypeStr(Type* Ty) { |
2349 | Result += "f"; | 2265 | Result += "f"; |
2350 | } | 2266 | } |
2351 | else if (isa<VectorType>(Ty)) | 2267 | else if (isa<VectorType>(Ty)) |
@@ -2355,10 +2271,10 @@ index 708cfdaa..772155b2 100644 | |||
2355 | else if (Ty) | 2271 | else if (Ty) |
2356 | Result += EVT::getEVT(Ty).getEVTString(); | 2272 | Result += EVT::getEVT(Ty).getEVTString(); |
2357 | diff --git a/IGC/LLVM3DBuilder/BuiltinsFrontendDefinitions.hpp b/IGC/LLVM3DBuilder/BuiltinsFrontendDefinitions.hpp | 2273 | diff --git a/IGC/LLVM3DBuilder/BuiltinsFrontendDefinitions.hpp b/IGC/LLVM3DBuilder/BuiltinsFrontendDefinitions.hpp |
2358 | index a89d63b9..a1bc5d2c 100644 | 2274 | index 2306c2357..b5f66f048 100644 |
2359 | --- a/IGC/LLVM3DBuilder/BuiltinsFrontendDefinitions.hpp | 2275 | --- a/IGC/LLVM3DBuilder/BuiltinsFrontendDefinitions.hpp |
2360 | +++ b/IGC/LLVM3DBuilder/BuiltinsFrontendDefinitions.hpp | 2276 | +++ b/IGC/LLVM3DBuilder/BuiltinsFrontendDefinitions.hpp |
2361 | @@ -4929,7 +4929,7 @@ void LLVM3DBuilder<preserveNames, T, Inserter>::VectorToScalars( | 2277 | @@ -4993,7 +4993,7 @@ void LLVM3DBuilder<preserveNames, T, Inserter>::VectorToScalars( |
2362 | IGC_ASSERT(nullptr != vector->getType()); | 2278 | IGC_ASSERT(nullptr != vector->getType()); |
2363 | IGC_ASSERT(vector->getType()->isVectorTy()); | 2279 | IGC_ASSERT(vector->getType()->isVectorTy()); |
2364 | 2280 | ||
@@ -2367,11 +2283,39 @@ index a89d63b9..a1bc5d2c 100644 | |||
2367 | IGC_ASSERT(1 < count); | 2283 | IGC_ASSERT(1 < count); |
2368 | IGC_ASSERT(count <= 4); | 2284 | IGC_ASSERT(count <= 4); |
2369 | IGC_ASSERT(count <= maxSize); | 2285 | IGC_ASSERT(count <= maxSize); |
2286 | diff --git a/IGC/WrapperLLVM/include/llvmWrapper/IR/DerivedTypes.h b/IGC/WrapperLLVM/include/llvmWrapper/IR/DerivedTypes.h | ||
2287 | index a207f5d44..117dae814 100644 | ||
2288 | --- a/IGC/WrapperLLVM/include/llvmWrapper/IR/DerivedTypes.h | ||
2289 | +++ b/IGC/WrapperLLVM/include/llvmWrapper/IR/DerivedTypes.h | ||
2290 | @@ -11,6 +11,7 @@ SPDX-License-Identifier: MIT | ||
2291 | |||
2292 | #include "llvm/Config/llvm-config.h" | ||
2293 | #include "llvm/IR/DerivedTypes.h" | ||
2294 | +#include "llvm/IR/Module.h" | ||
2295 | |||
2296 | namespace IGCLLVM | ||
2297 | { | ||
2298 | @@ -44,6 +45,15 @@ namespace IGCLLVM | ||
2299 | return false; | ||
2300 | #endif | ||
2301 | } | ||
2302 | + | ||
2303 | + inline llvm::StructType *getTypeByName(llvm::Module *M, llvm::StringRef Name) { | ||
2304 | +#if LLVM_VERSION_MAJOR >= 12 | ||
2305 | + return llvm::StructType::getTypeByName(M->getContext(), Name); | ||
2306 | +#else | ||
2307 | + return M->getTypeByName(Name); | ||
2308 | +#endif | ||
2309 | + } | ||
2310 | + | ||
2311 | } | ||
2312 | |||
2313 | #endif | ||
2370 | diff --git a/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h b/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h | 2314 | diff --git a/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h b/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h |
2371 | index 08e5d6cf..30e29720 100644 | 2315 | index 19c5183eb..709f1bbbb 100644 |
2372 | --- a/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h | 2316 | --- a/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h |
2373 | +++ b/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h | 2317 | +++ b/IGC/WrapperLLVM/include/llvmWrapper/Support/TypeSize.h |
2374 | @@ -35,10 +35,16 @@ using namespace llvm; | 2318 | @@ -17,10 +17,14 @@ using namespace llvm; |
2375 | namespace IGCLLVM { | 2319 | namespace IGCLLVM { |
2376 | #if LLVM_VERSION_MAJOR < 11 | 2320 | #if LLVM_VERSION_MAJOR < 11 |
2377 | inline unsigned getElementCount(unsigned EC) { return EC; } | 2321 | inline unsigned getElementCount(unsigned EC) { return EC; } |
@@ -2380,20 +2324,18 @@ index 08e5d6cf..30e29720 100644 | |||
2380 | inline ElementCount getElementCount(unsigned EC) { | 2324 | inline ElementCount getElementCount(unsigned EC) { |
2381 | return ElementCount(EC, false); | 2325 | return ElementCount(EC, false); |
2382 | } | 2326 | } |
2383 | +#elif LLVM_VERSION_MAJOR == 12 | 2327 | +#else |
2384 | +inline ElementCount getElementCount(unsigned EC) { | 2328 | +inline ElementCount getElementCount(unsigned EC) { |
2385 | + return ElementCount::get(EC, false); | 2329 | + return ElementCount::get(EC, false); |
2386 | +} | 2330 | +} |
2387 | +#else | ||
2388 | +#error "unsupported LLVM version" | ||
2389 | #endif | 2331 | #endif |
2390 | } // namespace IGCLLVM | 2332 | } // namespace IGCLLVM |
2391 | 2333 | ||
2392 | diff --git a/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Scalar.h b/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Scalar.h | 2334 | diff --git a/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Scalar.h b/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Scalar.h |
2393 | index f6e54fb1..5266ee87 100644 | 2335 | index 74841ddc1..1df041372 100644 |
2394 | --- a/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Scalar.h | 2336 | --- a/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Scalar.h |
2395 | +++ b/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Scalar.h | 2337 | +++ b/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Scalar.h |
2396 | @@ -44,7 +44,7 @@ namespace IGCLLVM | 2338 | @@ -26,7 +26,7 @@ namespace IGCLLVM |
2397 | { | 2339 | { |
2398 | return llvm::createLoopUnrollPass(OptLevel, false, Threshold, Count, AllowPartial, Runtime, UpperBound, AllowPeeling); | 2340 | return llvm::createLoopUnrollPass(OptLevel, false, Threshold, Count, AllowPartial, Runtime, UpperBound, AllowPeeling); |
2399 | } | 2341 | } |
@@ -2402,7 +2344,7 @@ index f6e54fb1..5266ee87 100644 | |||
2402 | inline static llvm::Pass * createLoopUnrollPass( | 2344 | inline static llvm::Pass * createLoopUnrollPass( |
2403 | int OptLevel = 2, int Threshold = -1, int Count = -1, | 2345 | int OptLevel = 2, int Threshold = -1, int Count = -1, |
2404 | int AllowPartial = -1, int Runtime = -1, | 2346 | int AllowPartial = -1, int Runtime = -1, |
2405 | @@ -52,7 +52,7 @@ namespace IGCLLVM | 2347 | @@ -34,7 +34,7 @@ namespace IGCLLVM |
2406 | { | 2348 | { |
2407 | return llvm::createLoopUnrollPass(OptLevel, false, false, Threshold, Count, AllowPartial, Runtime, UpperBound, AllowPeeling); | 2349 | return llvm::createLoopUnrollPass(OptLevel, false, false, Threshold, Count, AllowPartial, Runtime, UpperBound, AllowPeeling); |
2408 | } | 2350 | } |
@@ -2411,23 +2353,42 @@ index f6e54fb1..5266ee87 100644 | |||
2411 | //DO NOT assume same function signature for all incoming llvm versions! Double check to upgrade! | 2353 | //DO NOT assume same function signature for all incoming llvm versions! Double check to upgrade! |
2412 | #error Not supported llvm version. | 2354 | #error Not supported llvm version. |
2413 | #endif | 2355 | #endif |
2356 | diff --git a/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Utils/LoopUtils.h b/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Utils/LoopUtils.h | ||
2357 | index 8cf0adf24..33d5b7c35 100644 | ||
2358 | --- a/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Utils/LoopUtils.h | ||
2359 | +++ b/IGC/WrapperLLVM/include/llvmWrapper/Transforms/Utils/LoopUtils.h | ||
2360 | @@ -23,6 +23,14 @@ namespace IGCLLVM | ||
2361 | return llvm::InsertPreheaderForLoop(L, DT, LI, nullptr, PreserveLCSSA); | ||
2362 | } | ||
2363 | #endif | ||
2364 | + | ||
2365 | + inline bool isInnermost(llvm::Loop *L) { | ||
2366 | +#if LLVM_VERSION_MAJOR >= 12 | ||
2367 | + return L->isInnermost(); | ||
2368 | +#else | ||
2369 | + return L->empty(); | ||
2370 | +#endif | ||
2371 | + } | ||
2372 | } | ||
2373 | |||
2374 | #endif | ||
2414 | diff --git a/IGC/common/igc_resourceDimTypes.h b/IGC/common/igc_resourceDimTypes.h | 2375 | diff --git a/IGC/common/igc_resourceDimTypes.h b/IGC/common/igc_resourceDimTypes.h |
2415 | index 6831b06c..d790330f 100644 | 2376 | index a9a6a1871..713d6f4bb 100644 |
2416 | --- a/IGC/common/igc_resourceDimTypes.h | 2377 | --- a/IGC/common/igc_resourceDimTypes.h |
2417 | +++ b/IGC/common/igc_resourceDimTypes.h | 2378 | +++ b/IGC/common/igc_resourceDimTypes.h |
2418 | @@ -66,6 +66,11 @@ namespace IGC | 2379 | @@ -50,6 +50,10 @@ namespace IGC |
2419 | resourceDimTypeId == DIM_2D_TYPE || resourceDimTypeId == DIM_2D_ARRAY_TYPE || | 2380 | resourceDimTypeId == DIM_2D_TYPE || resourceDimTypeId == DIM_2D_ARRAY_TYPE || |
2420 | resourceDimTypeId == DIM_3D_TYPE || resourceDimTypeId == DIM_CUBE_TYPE || resourceDimTypeId == DIM_CUBE_ARRAY_TYPE)); | 2381 | resourceDimTypeId == DIM_3D_TYPE || resourceDimTypeId == DIM_CUBE_TYPE || resourceDimTypeId == DIM_CUBE_ARRAY_TYPE)); |
2421 | 2382 | ||
2422 | +#if LLVM_VERSION_MAJOR >= 12 | 2383 | +#if LLVM_VERSION_MAJOR >= 12 |
2423 | + llvm::LLVMContext& llvmCtx = module.getContext(); | 2384 | + return llvm::StructType::getTypeByName(module.getContext(), ResourceDimensionTypeName[resourceDimTypeId]); |
2424 | + return llvm::StructType::getTypeByName(llvmCtx, ResourceDimensionTypeName[resourceDimTypeId]); | ||
2425 | +#else | 2385 | +#else |
2426 | return module.getTypeByName(ResourceDimensionTypeName[resourceDimTypeId]); | 2386 | return module.getTypeByName(ResourceDimensionTypeName[resourceDimTypeId]); |
2427 | +#endif | 2387 | +#endif |
2428 | } | 2388 | } |
2429 | } | 2389 | -} |
2430 | \ No newline at end of file | 2390 | \ No newline at end of file |
2391 | +} | ||
2431 | -- | 2392 | -- |
2432 | 2.17.1 | 2393 | 2.20.1 |
2433 | 2394 | ||
diff --git a/dynamic-layers/clang-layer/recipes-opencl/igc/files/link-to-LLVMGenXIntrinsics.patch b/dynamic-layers/clang-layer/recipes-opencl/igc/files/link-to-LLVMGenXIntrinsics.patch deleted file mode 100644 index aead2d1c..00000000 --- a/dynamic-layers/clang-layer/recipes-opencl/igc/files/link-to-LLVMGenXIntrinsics.patch +++ /dev/null | |||
@@ -1,71 +0,0 @@ | |||
1 | From 11b923c99cbe3580885ce40c322277fc823107a0 Mon Sep 17 00:00:00 2001 | ||
2 | From: Naveen Saini <naveen.kumar.saini@intel.com> | ||
3 | Date: Tue, 2 Feb 2021 13:39:53 +0800 | ||
4 | Subject: [PATCH] IGC/VectorCompiler/CMakeLists.txt: link to external | ||
5 | LLVMGenXIntrinsics | ||
6 | |||
7 | By default LLVMGenXIntrinsics is to be build In-tree, but we want to | ||
8 | link externally. | ||
9 | |||
10 | Fix llvm-tblgen path. | ||
11 | |||
12 | Upstream-Status: Inappropriate [configuration specific] | ||
13 | |||
14 | Signed-off-by: Naveen Saini <naveen.kumar.saini@intel.com> | ||
15 | --- | ||
16 | IGC/VectorCompiler/CMakeLists.txt | 32 ++----------------------------- | ||
17 | 1 file changed, 2 insertions(+), 30 deletions(-) | ||
18 | |||
19 | diff --git a/IGC/VectorCompiler/CMakeLists.txt b/IGC/VectorCompiler/CMakeLists.txt | ||
20 | index 86f343ee..b4268160 100644 | ||
21 | --- a/IGC/VectorCompiler/CMakeLists.txt | ||
22 | +++ b/IGC/VectorCompiler/CMakeLists.txt | ||
23 | @@ -53,10 +53,7 @@ if(IGC_BUILD__USING_SYSTEM_LLVM OR (WIN32 AND LLVM_USE_PREBUILT)) | ||
24 | message(STATUS "[VC] Using system llvm") | ||
25 | |||
26 | # Need to search for llvm-tblgen | ||
27 | - find_program(LLVM_TABLEGEN_EXE "llvm-tblgen" | ||
28 | - ${LLVM_TOOLS_BINARY_DIR} | ||
29 | - NO_DEFAULT_PATH | ||
30 | - ) | ||
31 | + find_program(LLVM_TABLEGEN_EXE "llvm-tblgen") | ||
32 | if(LLVM_TABLEGEN_EXE-NOTFOUND) | ||
33 | message(FATAL_ERROR "[VC] llvm-tblgen is not found") | ||
34 | endif() | ||
35 | @@ -140,32 +137,7 @@ if(LLVM_ON_WIN32) | ||
36 | add_compile_options(/wd4141) | ||
37 | endif() | ||
38 | |||
39 | -if(DEFINED VC_INTRINSICS_SRC) | ||
40 | - set(INTRSRC "${VC_INTRINSICS_SRC}/GenXIntrinsics") | ||
41 | -endif() | ||
42 | - | ||
43 | -if(NOT DEFINED INTRSRC) | ||
44 | - set(INTRSRC "${CMAKE_CURRENT_SOURCE_DIR}/../../../vc-intrinsics/GenXIntrinsics") | ||
45 | -endif() | ||
46 | - | ||
47 | -message(STATUS "[VC] Using vc-intrinsics source from: ${INTRSRC}") | ||
48 | - | ||
49 | -# We are using prebuilt SPIRV and building intrinsics. | ||
50 | -set(INTRBUILD "${CMAKE_CURRENT_BINARY_DIR}/intrbuild") | ||
51 | -add_subdirectory(${INTRSRC} ${INTRBUILD}) | ||
52 | -# Make separate target for intrinstics headers. | ||
53 | -# Ideally, this should be inside intrinsics, but | ||
54 | -# this breaks in-tree build of intrinsics with LLVM. | ||
55 | -add_library(VCIntrinsicsHeaders INTERFACE) | ||
56 | -target_include_directories(VCIntrinsicsHeaders | ||
57 | - INTERFACE | ||
58 | - ${INTRSRC}/include | ||
59 | - ${INTRBUILD}/include | ||
60 | - ) | ||
61 | -add_dependencies(VCIntrinsicsHeaders | ||
62 | - GenXIntrinsics | ||
63 | - ) | ||
64 | - | ||
65 | +find_package(LLVMGenXIntrinsics REQUIRED) | ||
66 | include(cmake/spirv.cmake) | ||
67 | include(${IGC_SOURCE_DIR}/cmake/utils.cmake) | ||
68 | |||
69 | -- | ||
70 | 2.17.1 | ||
71 | |||
diff --git a/dynamic-layers/clang-layer/recipes-opencl/igc/intel-graphics-compiler_1.0.6646.bb b/dynamic-layers/clang-layer/recipes-opencl/igc/intel-graphics-compiler_1.0.8365.bb index d546dc22..153f4bb4 100644 --- a/dynamic-layers/clang-layer/recipes-opencl/igc/intel-graphics-compiler_1.0.6646.bb +++ b/dynamic-layers/clang-layer/recipes-opencl/igc/intel-graphics-compiler_1.0.8365.bb | |||
@@ -5,19 +5,18 @@ hardware architecture." | |||
5 | 5 | ||
6 | LICENSE = "MIT & BSD-3-Clause" | 6 | LICENSE = "MIT & BSD-3-Clause" |
7 | LIC_FILES_CHKSUM = "file://IGC/BiFModule/Implementation/ExternalLibraries/libclc/LICENSE.TXT;md5=311cfc1a5b54bab8ed34a0b5fba4373e \ | 7 | LIC_FILES_CHKSUM = "file://IGC/BiFModule/Implementation/ExternalLibraries/libclc/LICENSE.TXT;md5=311cfc1a5b54bab8ed34a0b5fba4373e \ |
8 | file://IGC/Compiler/LegalizationPass.cpp;beginline=1;endline=23;md5=8b19c5999abc484f18232b0905367f9f \ | 8 | file://IGC/Compiler/LegalizationPass.cpp;beginline=1;endline=23;md5=4a985f2545dd5a846e205b1e60a51cd9 \ |
9 | file://NOTICES.txt;md5=b12e73994de4fbe0f688cf0bc91512a0" | 9 | file://NOTICES.txt;md5=db621145dfb627436bc90ad600386801" |
10 | 10 | ||
11 | SRC_URI = "git://github.com/intel/intel-graphics-compiler.git;protocol=https; \ | 11 | SRC_URI = "git://github.com/intel/intel-graphics-compiler.git;protocol=https; \ |
12 | file://0001-skip-execution-of-ElfPackager.patch \ | 12 | file://0001-skip-execution-of-ElfPackager.patch \ |
13 | file://link-to-LLVMGenXIntrinsics.patch \ | 13 | file://0002-IGC-VectorCompiler-CMakeLists.txt-link-to-external-L.patch \ |
14 | file://improve_src_package_reproducibility.patch \ | 14 | file://0003-Improve-Reproducibility-for-src-package.patch \ |
15 | file://0001-Fix-build-with-LLVM-12.patch \ | 15 | file://0004-find-external-llvm-tblgen.patch \ |
16 | file://0002-Review-fixes-for-LLVM-12-phase-1.patch \ | 16 | file://0005-Temporary-LLVM-12-compatiblity-fix.patch \ |
17 | file://0003-Review-fixes-for-LLVM-12-phase-2.patch \ | ||
18 | " | 17 | " |
19 | 18 | ||
20 | SRCREV = "535aaaef03ce338e05e6162118082e6e007e8c10" | 19 | SRCREV = "5d5672d6cc0c415dae76648390026f777004bd99" |
21 | 20 | ||
22 | # Used to replace with relative path in reproducibility patch | 21 | # Used to replace with relative path in reproducibility patch |
23 | export B | 22 | export B |