[meta-intel] [PATCH 4/6] llvm-spirv: update to sync with opencl-clang
Anuj Mittal
anuj.mittal at intel.com
Sun Aug 18 20:57:31 PDT 2019
Also include a patch to not build tests to prevent passing the build
flag to every component that is built out of tree.
Signed-off-by: Anuj Mittal <anuj.mittal at intel.com>
---
...rOpenCL-pass-to-handle-new-blocks-represn.patch | 715 +++++++++------------
.../clang/files/0001-skip-building-tests.patch | 50 ++
.../clang/files/0002-Remove-extra-semicolon.patch | 27 +
.../clang/llvm-project-source.bbappend | 4 +-
4 files changed, 371 insertions(+), 425 deletions(-)
create mode 100644 dynamic-layers/clang-layer/recipes-devtools/clang/files/0001-skip-building-tests.patch
create mode 100644 dynamic-layers/clang-layer/recipes-devtools/clang/files/0002-Remove-extra-semicolon.patch
diff --git a/dynamic-layers/clang-layer/recipes-devtools/clang/files/0001-Update-LowerOpenCL-pass-to-handle-new-blocks-represn.patch b/dynamic-layers/clang-layer/recipes-devtools/clang/files/0001-Update-LowerOpenCL-pass-to-handle-new-blocks-represn.patch
index 4a52867..2037421 100644
--- a/dynamic-layers/clang-layer/recipes-devtools/clang/files/0001-Update-LowerOpenCL-pass-to-handle-new-blocks-represn.patch
+++ b/dynamic-layers/clang-layer/recipes-devtools/clang/files/0001-Update-LowerOpenCL-pass-to-handle-new-blocks-represn.patch
@@ -1,370 +1,57 @@
-From 9ce0fe02fd6cda5fb29fbb0d5037a1798a810b8a Mon Sep 17 00:00:00 2001
+From 177cce531fd3665bb964a03db51890e0241e3e72 Mon Sep 17 00:00:00 2001
From: Alexey Sotkin <alexey.sotkin at intel.com>
Date: Thu, 21 Feb 2019 17:14:36 +0300
-Subject: [PATCH 1/3] Update LowerOpenCL pass to handle new blocks
- represntation in LLVM IR
+Subject: [PATCH] Update LowerOpenCL pass to handle new blocks represntation in
+ LLVM IR
-Upstream-Status: Backport
-[https://github.com/KhronosGroup/SPIRV-LLVM-Translator/commit/bd6ddfaf7232cd81c7f2fe9877e66f286731bd8e]
+Upstream-Status: Backport [https://github.com/KhronosGroup/SPIRV-LLVM-Translator/commit/bd6ddfaf7232cd81c7f2fe9877e66f286731bd8e]
Signed-off-by: Anuj Mittal <anuj.mittal at intel.com>
+
---
- lib/SPIRV/SPIRVLowerOCLBlocks.cpp | 413 ++++++++----------------------
- test/global_block.ll | 71 ++---
- test/literal-struct.ll | 31 ++-
- test/transcoding/block_w_struct_return.ll | 47 ++--
- test/transcoding/enqueue_kernel.ll | 237 ++++++++++-------
- 5 files changed, 317 insertions(+), 482 deletions(-)
+ lib/SPIRV/SPIRVLowerOCLBlocks.cpp | 249 ++++--------------------------
+ test/global_block.ll | 71 ++++-----
+ test/literal-struct.ll | 31 ++--
+ test/transcoding/block_w_struct_return.ll | 47 +++---
+ test/transcoding/enqueue_kernel.ll | 237 ++++++++++++++++------------
+ 5 files changed, 235 insertions(+), 400 deletions(-)
diff --git a/lib/SPIRV/SPIRVLowerOCLBlocks.cpp b/lib/SPIRV/SPIRVLowerOCLBlocks.cpp
-index 50e1838..b42a4ec 100644
+index c80bf04..b42a4ec 100644
--- a/lib/SPIRV/SPIRVLowerOCLBlocks.cpp
+++ b/lib/SPIRV/SPIRVLowerOCLBlocks.cpp
-@@ -1,303 +1,110 @@
--//===- SPIRVLowerOCLBlocks.cpp - OCL Utilities ----------------------------===//
--//
--// The LLVM/SPIRV Translator
--//
--// This file is distributed under the University of Illinois Open Source
--// License. See LICENSE.TXT for details.
--//
--// Copyright (c) 2018 Intel Corporation. All rights reserved.
--//
--// Permission is hereby granted, free of charge, to any person obtaining a
--// copy of this software and associated documentation files (the "Software"),
--// to deal with the Software without restriction, including without limitation
--// the rights to use, copy, modify, merge, publish, distribute, sublicense,
--// and/or sell copies of the Software, and to permit persons to whom the
--// Software is furnished to do so, subject to the following conditions:
--//
--// Redistributions of source code must retain the above copyright notice,
--// this list of conditions and the following disclaimers.
--// Redistributions in binary form must reproduce the above copyright notice,
--// this list of conditions and the following disclaimers in the documentation
--// and/or other materials provided with the distribution.
--// Neither the names of Intel Corporation, nor the names of its
--// contributors may be used to endorse or promote products derived from this
--// Software without specific prior written permission.
--// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
--// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
--// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
--// CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
--// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
--// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS WITH
--// THE SOFTWARE.
--//
--//===----------------------------------------------------------------------===//
--//
--// SPIR-V specification doesn't allow function pointers, so SPIR-V translator
--// is designed to fail if a value with function type (except calls) is occured.
--// Currently there is only two cases, when function pointers are generating in
--// LLVM IR in OpenCL - block calls and device side enqueue built-in calls.
--//
--// In both cases values with function type used as intermediate representation
--// for block literal structure.
--//
--// This pass is designed to find such cases and simplify them to avoid any
--// function pointer types occurrences in LLVM IR in 4 steps.
--//
--// 1. Find all function pointer allocas, like
--// %block = alloca void () *
--//
--// Then find a single store to that alloca:
--// %blockLit = alloca <{ i32, i32, ...}>, align 4
--// %0 = bitcast <{ i32, i32, ... }>* %blockLit to void ()*
--// > store void ()* %0, void ()** %block, align 4
--//
--// And replace the alloca users by new instructions which used stored value
--// %blockLit itself instead of function pointer alloca %block.
--//
--// 2. Find consecutive casts from block literal type to i8 addrspace(4)*
--// used function pointers as an intermediate type:
--// %0 = bitcast <{ i32, i32 }> %block to void() *
--// %1 = addrspacecast void() * %0 to i8 addrspace(4)*
--// And simplify them:
--// %2 = addrspacecast <{ i32, i32 }> %block to i8 addrspace(4)*
--//
--// 3. Find all unused instructions with function pointer type occured after
--// pp.1-2 and remove them.
--//
--// 4. Find unused globals with function pointer type, like
--// @block = constant void ()*
--// bitcast ({ i32, i32 }* @__block_literal_global to void ()*
--//
--// And remove them.
--//
--//===----------------------------------------------------------------------===//
--#define DEBUG_TYPE "spv-lower-ocl-blocks"
--
--#include "OCLUtil.h"
--#include "SPIRVInternal.h"
--
--#include "llvm/ADT/SetVector.h"
--#include "llvm/Analysis/ValueTracking.h"
--#include "llvm/IR/GlobalVariable.h"
--#include "llvm/IR/InstIterator.h"
--#include "llvm/IR/Module.h"
--#include "llvm/Pass.h"
--#include "llvm/PassSupport.h"
--#include "llvm/Support/Casting.h"
--
--using namespace llvm;
--
--namespace {
--
--static void
--removeUnusedFunctionPtrInst(Instruction *I,
-- SmallSetVector<Instruction *, 16> &FuncPtrInsts) {
-- for (unsigned OpIdx = 0, Ops = I->getNumOperands(); OpIdx != Ops; ++OpIdx) {
-- Instruction *OpI = dyn_cast<Instruction>(I->getOperand(OpIdx));
-- I->setOperand(OpIdx, nullptr);
-- if (OpI && OpI != I && OpI->user_empty())
-- FuncPtrInsts.insert(OpI);
-- }
-- I->eraseFromParent();
--}
--
--static bool isFuncPtrAlloca(const AllocaInst *AI) {
-- auto *ET = dyn_cast<PointerType>(AI->getAllocatedType());
-- return ET && ET->getElementType()->isFunctionTy();
--}
--
--static bool hasFuncPtrType(const Value *V) {
-- auto *PT = dyn_cast<PointerType>(V->getType());
-- return PT && PT->getElementType()->isFunctionTy();
--}
--
--static bool isFuncPtrInst(const Instruction *I) {
-- if (auto *AI = dyn_cast<AllocaInst>(I))
-- return isFuncPtrAlloca(AI);
--
-- for (auto &Op : I->operands()) {
-- if (auto *AI = dyn_cast<AllocaInst>(Op))
-- return isFuncPtrAlloca(AI);
--
-- auto *OpI = dyn_cast<Instruction>(&Op);
-- if (OpI && OpI != I && hasFuncPtrType(OpI))
-- return true;
-- }
-- return false;
--}
--
--static StoreInst *findSingleStore(AllocaInst *AI) {
-- StoreInst *Store = nullptr;
-- for (auto *U : AI->users()) {
-- if (!isa<StoreInst>(U))
-- continue; // not a store
-- if (Store)
-- return nullptr; // there are more than one stores
-- Store = dyn_cast<StoreInst>(U);
-- }
-- return Store;
--}
--
--static void fixFunctionPtrAllocaUsers(AllocaInst *AI) {
-- // Find and remove a single store to alloca
-- auto *SingleStore = findSingleStore(AI);
-- assert(SingleStore && "More than one store to the function pointer alloca");
-- auto *StoredVal = SingleStore->getValueOperand();
-- SingleStore->eraseFromParent();
--
-- // Find loads from the alloca and replace thier users
-- for (auto *U : AI->users()) {
-- auto *LI = dyn_cast<LoadInst>(U);
-- if (!LI)
-- continue;
--
-- for (auto *U : LI->users()) {
-- auto *UInst = cast<Instruction>(U);
-- auto *Cast = CastInst::CreatePointerBitCastOrAddrSpaceCast(
-- StoredVal, UInst->getType(), "", UInst);
-- UInst->replaceAllUsesWith(Cast);
-- }
-- }
--}
--
--static int getBlockLiteralIdx(const Function &F) {
-- StringRef FName = F.getName();
-- if (isEnqueueKernelBI(FName))
-- return FName.contains("events") ? 7 : 4;
-- if (isKernelQueryBI(FName))
-- return FName.contains("for_ndrange") ? 2 : 1;
-- if (FName.startswith("__") && FName.contains("_block_invoke"))
-- return F.hasStructRetAttr() ? 1 : 0;
--
-- return -1; // No block literal argument
--}
--
--static bool hasBlockLiteralArg(const Function &F) {
-- return getBlockLiteralIdx(F) != -1;
--}
--
--static bool simplifyFunctionPtrCasts(Function &F) {
-- bool Changed = false;
-- int BlockLiteralIdx = getBlockLiteralIdx(F);
-- for (auto *U : F.users()) {
-- auto *Call = dyn_cast<CallInst>(U);
-- if (!Call)
-- continue;
-- if (Call->getFunction()->getName() == F.getName().str() + "_kernel")
-- continue; // Skip block invoke function calls inside block invoke kernels
--
-- const DataLayout &DL = F.getParent()->getDataLayout();
-- auto *BlockLiteral = Call->getOperand(BlockLiteralIdx);
-- auto *BlockLiteralVal = GetUnderlyingObject(BlockLiteral, DL);
-- if (isa<GlobalVariable>(BlockLiteralVal))
-- continue; // nothing to do with globals
--
-- auto *BlockLiteralAlloca = cast<AllocaInst>(BlockLiteralVal);
-- assert(!BlockLiteralAlloca->getAllocatedType()->isFunctionTy() &&
-- "Function type shouldn't be there");
--
-- auto *NewBlockLiteral = CastInst::CreatePointerBitCastOrAddrSpaceCast(
-- BlockLiteralAlloca, BlockLiteral->getType(), "", Call);
-- BlockLiteral->replaceAllUsesWith(NewBlockLiteral);
-- Changed |= true;
-- }
-- return Changed;
--}
--
--static void
--findFunctionPtrAllocas(Module &M,
-- SmallVectorImpl<AllocaInst *> &FuncPtrAllocas) {
-- for (auto &F : M) {
-- if (F.isDeclaration())
-- continue;
-- for (auto &I : instructions(F)) {
-- auto *AI = dyn_cast<AllocaInst>(&I);
-- if (!AI || !isFuncPtrAlloca(AI))
-- continue;
-- FuncPtrAllocas.push_back(AI);
-- }
-- }
--}
--
--static void
--findUnusedFunctionPtrInsts(Module &M,
-- SmallSetVector<Instruction *, 16> &FuncPtrInsts) {
-- for (auto &F : M) {
-- if (F.isDeclaration())
-- continue;
-- for (auto &I : instructions(F))
-- if (I.user_empty() && isFuncPtrInst(&I))
-- FuncPtrInsts.insert(&I);
-- }
--}
--
--static void
--findUnusedFunctionPtrGlbs(Module &M,
-- SmallVectorImpl<GlobalVariable *> &FuncPtrGlbs) {
-- for (auto &GV : M.globals()) {
-- if (!GV.user_empty())
-- continue;
-- auto *GVType = dyn_cast<PointerType>(GV.getType()->getElementType());
-- if (GVType && GVType->getElementType()->isFunctionTy())
-- FuncPtrGlbs.push_back(&GV);
-- }
--}
--
--class SPIRVLowerOCLBlocks : public ModulePass {
--
--public:
-- SPIRVLowerOCLBlocks() : ModulePass(ID) {}
--
-- bool runOnModule(Module &M) {
-- bool Changed = false;
--
-- // 1. Find function pointer allocas and fix their users
-- SmallVector<AllocaInst *, 16> FuncPtrAllocas;
-- findFunctionPtrAllocas(M, FuncPtrAllocas);
--
-- Changed |= !FuncPtrAllocas.empty();
-- for (auto *AI : FuncPtrAllocas)
-- fixFunctionPtrAllocaUsers(AI);
--
-- // 2. Simplify consecutive casts which use function pointer types
-- for (auto &F : M)
-- if (hasBlockLiteralArg(F))
-- Changed |= simplifyFunctionPtrCasts(F);
--
-- // 3. Cleanup unused instructions with function pointer type
-- // which are occured after pp. 1-2
-- SmallSetVector<Instruction *, 16> FuncPtrInsts;
-- findUnusedFunctionPtrInsts(M, FuncPtrInsts);
--
-- Changed |= !FuncPtrInsts.empty();
-- while (!FuncPtrInsts.empty()) {
-- Instruction *I = FuncPtrInsts.pop_back_val();
-- removeUnusedFunctionPtrInst(I, FuncPtrInsts);
-- }
--
-- // 4. Find and remove unused global variables with function pointer type
-- SmallVector<GlobalVariable *, 16> FuncPtrGlbs;
-- findUnusedFunctionPtrGlbs(M, FuncPtrGlbs);
--
-- Changed |= !FuncPtrGlbs.empty();
-- for (auto *GV : FuncPtrGlbs)
-- GV->eraseFromParent();
--
-- return Changed;
-- }
--
-- static char ID;
--}; // class SPIRVLowerOCLBlocks
--
--char SPIRVLowerOCLBlocks::ID = 0;
--
--} // namespace
--
--INITIALIZE_PASS(
-- SPIRVLowerOCLBlocks, "spv-lower-ocl-blocks",
-- "Remove function pointers occured in case of using OpenCL blocks", false,
-- false)
--
--llvm::ModulePass *llvm::createSPIRVLowerOCLBlocks() {
-- return new SPIRVLowerOCLBlocks();
--}
-+//===- SPIRVLowerOCLBlocks.cpp - OCL Utilities ----------------------------===//
-+//
-+// The LLVM/SPIRV Translator
-+//
-+// This file is distributed under the University of Illinois Open Source
-+// License. See LICENSE.TXT for details.
-+//
-+// Copyright (c) 2018 Intel Corporation. All rights reserved.
-+//
-+// Permission is hereby granted, free of charge, to any person obtaining a
-+// copy of this software and associated documentation files (the "Software"),
-+// to deal with the Software without restriction, including without limitation
-+// the rights to use, copy, modify, merge, publish, distribute, sublicense,
-+// and/or sell copies of the Software, and to permit persons to whom the
-+// Software is furnished to do so, subject to the following conditions:
-+//
-+// Redistributions of source code must retain the above copyright notice,
-+// this list of conditions and the following disclaimers.
-+// Redistributions in binary form must reproduce the above copyright notice,
-+// this list of conditions and the following disclaimers in the documentation
-+// and/or other materials provided with the distribution.
-+// Neither the names of Intel Corporation, nor the names of its
-+// contributors may be used to endorse or promote products derived from this
-+// Software without specific prior written permission.
-+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-+// CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS WITH
-+// THE SOFTWARE.
-+//
-+//===----------------------------------------------------------------------===//
-+//
-+// SPIR-V specification doesn't allow function pointers, so SPIR-V translator
-+// is designed to fail if a value with function type (except calls) is occured.
-+// Currently there is only two cases, when function pointers are generating in
-+// LLVM IR in OpenCL - block calls and device side enqueue built-in calls.
-+//
-+// In both cases values with function type used as intermediate representation
-+// for block literal structure.
-+//
+@@ -40,207 +40,34 @@
+ // In both cases values with function type used as intermediate representation
+ // for block literal structure.
+ //
+-// This pass is designed to find such cases and simplify them to avoid any
+-// function pointer types occurrences in LLVM IR in 4 steps.
+-//
+-// 1. Find all function pointer allocas, like
+-// %block = alloca void () *
+-//
+-// Then find a single store to that alloca:
+-// %blockLit = alloca <{ i32, i32, ...}>, align 4
+-// %0 = bitcast <{ i32, i32, ... }>* %blockLit to void ()*
+-// > store void ()* %0, void ()** %block, align 4
+-//
+-// And replace the alloca users by new instructions which used stored value
+-// %blockLit itself instead of function pointer alloca %block.
+-//
+-// 2. Find consecutive casts from block literal type to i8 addrspace(4)*
+-// used function pointers as an intermediate type:
+-// %0 = bitcast <{ i32, i32 }> %block to void() *
+-// %1 = addrspacecast void() * %0 to i8 addrspace(4)*
+-// And simplify them:
+-// %2 = addrspacecast <{ i32, i32 }> %block to i8 addrspace(4)*
+-//
+-// 3. Find all unused instructions with function pointer type occured after
+-// pp.1-2 and remove them.
+-//
+-// 4. Find unused globals with function pointer type, like
+-// @block = constant void ()*
+-// bitcast ({ i32, i32 }* @__block_literal_global to void ()*
+-//
+-// And remove them.
+// In LLVM IR produced by clang, blocks are represented with the following
+// structure:
+// %struct.__opencl_block_literal_generic = type { i32, i32, i8 addrspace(4)* }
@@ -376,32 +63,212 @@ index 50e1838..b42a4ec 100644
+// Clang though has to store function pointers to this structure. Purpose of
+// this pass is to replace store of function pointers(not allowed in SPIR-V)
+// with null pointers.
-+//
-+//===----------------------------------------------------------------------===//
-+#define DEBUG_TYPE "spv-lower-ocl-blocks"
-+
-+#include "SPIRVInternal.h"
-+
-+#include "llvm/IR/Module.h"
-+#include "llvm/Pass.h"
+ //
+ //===----------------------------------------------------------------------===//
+ #define DEBUG_TYPE "spv-lower-ocl-blocks"
+
+-#include "OCLUtil.h"
+ #include "SPIRVInternal.h"
+
+-#include "llvm/ADT/SetVector.h"
+-#include "llvm/Analysis/ValueTracking.h"
+-#include "llvm/IR/GlobalVariable.h"
+-#include "llvm/IR/InstIterator.h"
+ #include "llvm/IR/Module.h"
+ #include "llvm/Pass.h"
+-#include "llvm/PassSupport.h"
+-#include "llvm/Support/Casting.h"
+#include "llvm/Support/Regex.h"
-+
-+using namespace llvm;
-+
-+namespace {
-+
+
+ using namespace llvm;
+
+ namespace {
+
+-static void
+-removeUnusedFunctionPtrInst(Instruction *I,
+- SmallSetVector<Instruction *, 16> &FuncPtrInsts) {
+- for (unsigned OpIdx = 0, Ops = I->getNumOperands(); OpIdx != Ops; ++OpIdx) {
+- Instruction *OpI = dyn_cast<Instruction>(I->getOperand(OpIdx));
+- I->setOperand(OpIdx, nullptr);
+- if (OpI && OpI != I && OpI->user_empty())
+- FuncPtrInsts.insert(OpI);
+- }
+- I->eraseFromParent();
+-}
+-
+-static bool isFuncPtrAlloca(const AllocaInst *AI) {
+- auto *ET = dyn_cast<PointerType>(AI->getAllocatedType());
+- return ET && ET->getElementType()->isFunctionTy();
+-}
+-
+-static bool hasFuncPtrType(const Value *V) {
+- auto *PT = dyn_cast<PointerType>(V->getType());
+- return PT && PT->getElementType()->isFunctionTy();
+-}
+-
+-static bool isFuncPtrInst(const Instruction *I) {
+- if (auto *AI = dyn_cast<AllocaInst>(I))
+- return isFuncPtrAlloca(AI);
+-
+- for (auto &Op : I->operands()) {
+- if (auto *AI = dyn_cast<AllocaInst>(Op))
+- return isFuncPtrAlloca(AI);
+-
+- auto *OpI = dyn_cast<Instruction>(&Op);
+- if (OpI && OpI != I && hasFuncPtrType(OpI))
+- return true;
+- }
+- return false;
+-}
+-
+-static StoreInst *findSingleStore(AllocaInst *AI) {
+- StoreInst *Store = nullptr;
+- for (auto *U : AI->users()) {
+- if (!isa<StoreInst>(U))
+- continue; // not a store
+- if (Store)
+- return nullptr; // there are more than one stores
+- Store = dyn_cast<StoreInst>(U);
+- }
+- return Store;
+-}
+-
+-static void fixFunctionPtrAllocaUsers(AllocaInst *AI) {
+- // Find and remove a single store to alloca
+- auto *SingleStore = findSingleStore(AI);
+- assert(SingleStore && "More than one store to the function pointer alloca");
+- auto *StoredVal = SingleStore->getValueOperand();
+- SingleStore->eraseFromParent();
+-
+- // Find loads from the alloca and replace thier users
+- for (auto *U : AI->users()) {
+- auto *LI = dyn_cast<LoadInst>(U);
+- if (!LI)
+- continue;
+-
+- for (auto *U : LI->users()) {
+- auto *UInst = cast<Instruction>(U);
+- auto *Cast = CastInst::CreatePointerBitCastOrAddrSpaceCast(
+- StoredVal, UInst->getType(), "", UInst);
+- UInst->replaceAllUsesWith(Cast);
+- }
+- }
+-}
+-
+-static int getBlockLiteralIdx(const Function &F) {
+- StringRef FName = F.getName();
+- if (isEnqueueKernelBI(FName))
+- return FName.contains("events") ? 7 : 4;
+- if (isKernelQueryBI(FName))
+- return FName.contains("for_ndrange") ? 2 : 1;
+- if (FName.startswith("__") && FName.contains("_block_invoke"))
+- return F.hasStructRetAttr() ? 1 : 0;
+-
+- return -1; // No block literal argument
+-}
+-
+-static bool hasBlockLiteralArg(const Function &F) {
+- return getBlockLiteralIdx(F) != -1;
+-}
+-
+-static bool simplifyFunctionPtrCasts(Function &F) {
+- bool Changed = false;
+- int BlockLiteralIdx = getBlockLiteralIdx(F);
+- for (auto *U : F.users()) {
+- auto *Call = dyn_cast<CallInst>(U);
+- if (!Call)
+- continue;
+- if (Call->getFunction()->getName() == F.getName().str() + "_kernel")
+- continue; // Skip block invoke function calls inside block invoke kernels
+-
+- const DataLayout &DL = F.getParent()->getDataLayout();
+- auto *BlockLiteral = Call->getOperand(BlockLiteralIdx);
+- auto *BlockLiteralVal = GetUnderlyingObject(BlockLiteral, DL);
+- if (isa<GlobalVariable>(BlockLiteralVal))
+- continue; // nothing to do with globals
+-
+- auto *BlockLiteralAlloca = cast<AllocaInst>(BlockLiteralVal);
+- assert(!BlockLiteralAlloca->getAllocatedType()->isFunctionTy() &&
+- "Function type shouldn't be there");
+-
+- auto *NewBlockLiteral = CastInst::CreatePointerBitCastOrAddrSpaceCast(
+- BlockLiteralAlloca, BlockLiteral->getType(), "", Call);
+- BlockLiteral->replaceAllUsesWith(NewBlockLiteral);
+- Changed |= true;
+- }
+- return Changed;
+-}
+-
+-static void
+-findFunctionPtrAllocas(Module &M,
+- SmallVectorImpl<AllocaInst *> &FuncPtrAllocas) {
+- for (auto &F : M) {
+- if (F.isDeclaration())
+- continue;
+- for (auto &I : instructions(F)) {
+- auto *AI = dyn_cast<AllocaInst>(&I);
+- if (!AI || !isFuncPtrAlloca(AI))
+- continue;
+- FuncPtrAllocas.push_back(AI);
+- }
+- }
+-}
+-
+-static void
+-findUnusedFunctionPtrInsts(Module &M,
+- SmallSetVector<Instruction *, 16> &FuncPtrInsts) {
+- for (auto &F : M) {
+- if (F.isDeclaration())
+- continue;
+- for (auto &I : instructions(F))
+- if (I.user_empty() && isFuncPtrInst(&I))
+- FuncPtrInsts.insert(&I);
+- }
+-}
+-
+-static void
+-findUnusedFunctionPtrGlbs(Module &M,
+- SmallVectorImpl<GlobalVariable *> &FuncPtrGlbs) {
+- for (auto &GV : M.globals()) {
+- if (!GV.user_empty())
+- continue;
+- auto *GVType = dyn_cast<PointerType>(GV.getType()->getElementType());
+- if (GVType && GVType->getElementType()->isFunctionTy())
+- FuncPtrGlbs.push_back(&GV);
+- }
+static bool isBlockInvoke(Function &F) {
+ static Regex BlockInvokeRegex("_block_invoke_?[0-9]*$");
+ return BlockInvokeRegex.match(F.getName());
-+}
-+
-+class SPIRVLowerOCLBlocks : public ModulePass {
-+
-+public:
-+ SPIRVLowerOCLBlocks() : ModulePass(ID) {}
-+
-+ bool runOnModule(Module &M) {
-+ bool Changed = false;
+ }
+
+ class SPIRVLowerOCLBlocks : public ModulePass {
+@@ -250,44 +77,24 @@ public:
+
+ bool runOnModule(Module &M) {
+ bool Changed = false;
+-
+- // 1. Find function pointer allocas and fix their users
+- SmallVector<AllocaInst *, 16> FuncPtrAllocas;
+- findFunctionPtrAllocas(M, FuncPtrAllocas);
+-
+- Changed |= !FuncPtrAllocas.empty();
+- for (auto *AI : FuncPtrAllocas)
+- fixFunctionPtrAllocaUsers(AI);
+-
+- // 2. Simplify consecutive casts which use function pointer types
+- for (auto &F : M)
+- if (hasBlockLiteralArg(F))
+- Changed |= simplifyFunctionPtrCasts(F);
+-
+- // 3. Cleanup unused instructions with function pointer type
+- // which are occured after pp. 1-2
+- SmallSetVector<Instruction *, 16> FuncPtrInsts;
+- findUnusedFunctionPtrInsts(M, FuncPtrInsts);
+-
+- Changed |= !FuncPtrInsts.empty();
+- while (!FuncPtrInsts.empty()) {
+- Instruction *I = FuncPtrInsts.pop_back_val();
+- removeUnusedFunctionPtrInst(I, FuncPtrInsts);
+ for (Function &F : M) {
+ if (!isBlockInvoke(F))
+ continue;
@@ -414,31 +281,31 @@ index 50e1838..b42a4ec 100644
+ Changed = true;
+ }
+ }
-+ }
-+ return Changed;
-+ }
-+
-+ static char ID;
+ }
+-
+- // 4. Find and remove unused global variables with function pointer type
+- SmallVector<GlobalVariable *, 16> FuncPtrGlbs;
+- findUnusedFunctionPtrGlbs(M, FuncPtrGlbs);
+-
+- Changed |= !FuncPtrGlbs.empty();
+- for (auto *GV : FuncPtrGlbs)
+- GV->eraseFromParent();
+-
+ return Changed;
+ }
+
+ static char ID;
+-}; // class SPIRVLowerOCLBlocks
+};
-+
-+char SPIRVLowerOCLBlocks::ID = 0;
-+
-+} // namespace
-+
-+INITIALIZE_PASS(
-+ SPIRVLowerOCLBlocks, "spv-lower-ocl-blocks",
-+ "Remove function pointers occured in case of using OpenCL blocks", false,
-+ false)
-+
-+llvm::ModulePass *llvm::createSPIRVLowerOCLBlocks() {
-+ return new SPIRVLowerOCLBlocks();
-+}
+
+ char SPIRVLowerOCLBlocks::ID = 0;
+
diff --git a/test/global_block.ll b/test/global_block.ll
-index a9267d8..efb4cf3 100644
+index 4fc453b..b558213 100644
--- a/test/global_block.ll
+++ b/test/global_block.ll
-@@ -16,7 +16,7 @@
- ; RUN: llvm-spirv %t.bc -o %t.spv
+@@ -17,7 +17,7 @@
+ ; RUN: spirv-val %t.spv
; RUN: llvm-spirv -r %t.spv -o - | llvm-dis | FileCheck %s --check-prefix=CHECK-LLVM
-target datalayout = "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
@@ -446,7 +313,7 @@ index a9267d8..efb4cf3 100644
target triple = "spir-unknown-unknown"
; CHECK-SPIRV: Name [[block_invoke:[0-9]+]] "_block_invoke"
-@@ -26,71 +26,56 @@ target triple = "spir-unknown-unknown"
+@@ -27,71 +27,56 @@ target triple = "spir-unknown-unknown"
; CHECK-SPIRV: TypePointer [[int8Ptr:[0-9]+]] 8 [[int8]]
; CHECK-SPIRV: TypeFunction [[block_invoke_type:[0-9]+]] [[int]] [[int8Ptr]] [[int]]
@@ -546,7 +413,7 @@ index a9267d8..efb4cf3 100644
+!5 = !{!"int*"}
+!6 = !{!""}
diff --git a/test/literal-struct.ll b/test/literal-struct.ll
-index c52170a..52a731a 100644
+index b88187f..dec957a 100644
--- a/test/literal-struct.ll
+++ b/test/literal-struct.ll
@@ -2,7 +2,7 @@
@@ -558,9 +425,9 @@ index c52170a..52a731a 100644
; literal-struct.cl:
; void foo()
-@@ -14,25 +14,28 @@
- ; RUN: llvm-as < %s | llvm-spirv -spirv-text -o %t
- ; RUN: FileCheck < %t %s
+@@ -17,25 +17,28 @@
+ ; RUN: llvm-spirv %t.bc -o %t.spv
+ ; RUN: spirv-val %t.spv
-; CHECK-DAG: TypeInt [[Int:[0-9]+]] 32 0
-; CHECK-DAG: TypeStruct [[StructType:[0-9]+]] [[Int]] [[Int]] {{$}}
@@ -595,7 +462,7 @@ index c52170a..52a731a 100644
ret void
}
-@@ -40,14 +43,14 @@ entry:
+@@ -43,14 +46,14 @@ entry:
define internal spir_func void @__foo_block_invoke(i8 addrspace(4)* %.block_descriptor) #0 {
entry:
%.block_descriptor.addr = alloca i8 addrspace(4)*, align 4
@@ -614,14 +481,14 @@ index c52170a..52a731a 100644
attributes #1 = { convergent }
!llvm.module.flags = !{!0}
-@@ -57,4 +60,4 @@ attributes #1 = { convergent }
+@@ -60,4 +63,4 @@ attributes #1 = { convergent }
!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{i32 2, i32 0}
-!2 = !{!"clang version 8.0.0 "}
+!2 = !{!"clang version 9.0.0 (https://llvm.org/git/clang 04fb8964a801a5c5d7baa5a22272243a7d183896) (https://llvm.org/git/llvm 384f64397f6ad95a361b72d62c07d7bac9f24163)"}
diff --git a/test/transcoding/block_w_struct_return.ll b/test/transcoding/block_w_struct_return.ll
-index 76e29f0..df89b13 100644
+index a68820f..ebd2c5f 100644
--- a/test/transcoding/block_w_struct_return.ll
+++ b/test/transcoding/block_w_struct_return.ll
@@ -16,6 +16,8 @@
@@ -633,7 +500,7 @@ index 76e29f0..df89b13 100644
; RUN: llvm-as %s -o %t.bc
; RUN: llvm-spirv %t.bc -spirv-text -o %t.spv.txt
; RUN: FileCheck < %t.spv.txt %s --check-prefix=CHECK-SPIRV
-@@ -27,12 +29,14 @@
+@@ -28,12 +30,14 @@
; CHECK-SPIRV: Name [[BlockInv:[0-9]+]] "__block_ret_struct_block_invoke"
; CHECK-SPIRV: 4 TypeInt [[IntTy:[0-9]+]] 32
@@ -649,7 +516,7 @@ index 76e29f0..df89b13 100644
; CHECK-SPIRV: 7 FunctionCall {{[0-9]+}} {{[0-9]+}} [[BlockInv]] [[StructRet]] [[BlockLit]] [[StructArg]]
; CHECK-LLVM: %[[StructA:.*]] = type { i32 }
-@@ -41,20 +45,21 @@
+@@ -42,20 +46,21 @@
target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
target triple = "spir64-unknown-unknown"
@@ -675,7 +542,7 @@ index 76e29f0..df89b13 100644
%call = call spir_func i64 @_Z13get_global_idj(i32 0) #4
store i64 %call, i64* %tid, align 8
%0 = load i32 addrspace(1)*, i32 addrspace(1)** %res.addr, align 8
-@@ -63,7 +68,7 @@ entry:
+@@ -64,7 +69,7 @@ entry:
store i32 -1, i32 addrspace(1)* %arrayidx, align 4
%a = getelementptr inbounds %struct.A, %struct.A* %aa, i32 0, i32 0
store i32 5, i32* %a, align 4
@@ -684,7 +551,7 @@ index 76e29f0..df89b13 100644
%a1 = getelementptr inbounds %struct.A, %struct.A* %tmp, i32 0, i32 0
%2 = load i32, i32* %a1, align 4
%sub = sub nsw i32 %2, 6
-@@ -78,10 +83,10 @@ entry:
+@@ -79,10 +84,10 @@ entry:
define internal spir_func void @__block_ret_struct_block_invoke(%struct.A* noalias sret %agg.result, i8 addrspace(4)* %.block_descriptor, %struct.A* byval align 4 %a) #1 {
entry:
%.block_descriptor.addr = alloca i8 addrspace(4)*, align 8
@@ -698,7 +565,7 @@ index 76e29f0..df89b13 100644
%a1 = getelementptr inbounds %struct.A, %struct.A* %a, i32 0, i32 0
store i32 6, i32* %a1, align 4
%0 = bitcast %struct.A* %agg.result to i8*
-@@ -96,30 +101,22 @@ declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture r
+@@ -97,30 +102,22 @@ declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture r
; Function Attrs: convergent nounwind readnone
declare spir_func i64 @_Z13get_global_idj(i32) #3
@@ -738,7 +605,7 @@ index 76e29f0..df89b13 100644
+!5 = !{!"int*"}
+!6 = !{!""}
diff --git a/test/transcoding/enqueue_kernel.ll b/test/transcoding/enqueue_kernel.ll
-index 0d29c71..435871d 100644
+index 1f0b360..761043e 100644
--- a/test/transcoding/enqueue_kernel.ll
+++ b/test/transcoding/enqueue_kernel.ll
@@ -51,11 +51,12 @@
@@ -781,14 +648,14 @@ index 0d29c71..435871d 100644
; CHECK-SPIRV: TypeFunction [[BlockTy3:[0-9]+]] [[VoidTy]] [[Int8PtrGenTy]]
; CHECK-SPIRV: ConstantNull [[EventPtrTy]] [[EventNull:[0-9]+]]
--; CHECK-LLVM: [[BlockTy1:%[0-9]+]] = type { i32, i32 }
--; CHECK-LLVM: [[BlockTy2:%[0-9]+]] = type <{ i32, i32, i32 addrspace(1)*, i32, i8 }>
--; CHECK-LLVM: [[BlockTy3:%[0-9]+]] = type <{ i32, i32, i32 addrspace(1)*, i32, i32 addrspace(1)* }>
--; CHECK-LLVM: [[BlockTy4:%[0-9]+]] = type <{ i32, i32 }>
-+; CHECK-LLVM: [[BlockTy1:%[0-9]+]] = type { i32, i32, i8 addrspace(4)* }
-+; CHECK-LLVM: [[BlockTy2:%[0-9]+]] = type <{ i32, i32, i8 addrspace(4)*, i32 addrspace(1)*, i32, i8 }>
-+; CHECK-LLVM: [[BlockTy3:%[0-9]+]] = type <{ i32, i32, i8 addrspace(4)*, i32 addrspace(1)*, i32, i32 addrspace(1)* }>
-+; CHECK-LLVM: [[BlockTy4:%[0-9]+]] = type <{ i32, i32, i8 addrspace(4)* }>
+-; CHECK-LLVM: [[BlockTy1:%[0-9a-z\.]+]] = type { i32, i32 }
+-; CHECK-LLVM: [[BlockTy2:%[0-9a-z\.]+]] = type <{ i32, i32, i32 addrspace(1)*, i32, i8 }>
+-; CHECK-LLVM: [[BlockTy3:%[0-9a-z\.]+]] = type <{ i32, i32, i32 addrspace(1)*, i32, i32 addrspace(1)* }>
+-; CHECK-LLVM: [[BlockTy4:%[0-9a-z\.]+]] = type <{ i32, i32 }>
++; CHECK-LLVM: [[BlockTy1:%[0-9a-z\.]+]] = type { i32, i32, i8 addrspace(4)* }
++; CHECK-LLVM: [[BlockTy2:%[0-9a-z\.]+]] = type <{ i32, i32, i8 addrspace(4)*, i32 addrspace(1)*, i32, i8 }>
++; CHECK-LLVM: [[BlockTy3:%[0-9a-z\.]+]] = type <{ i32, i32, i8 addrspace(4)*, i32 addrspace(1)*, i32, i32 addrspace(1)* }>
++; CHECK-LLVM: [[BlockTy4:%[0-9a-z\.]+]] = type <{ i32, i32, i8 addrspace(4)* }>
-; CHECK-LLVM: @__block_literal_global = internal addrspace(1) constant [[BlockTy1]] { i32 8, i32 4 }, align 4
-; CHECK-LLVM: @__block_literal_global.1 = internal addrspace(1) constant [[BlockTy1]] { i32 8, i32 4 }, align 4
@@ -1115,5 +982,5 @@ index 0d29c71..435871d 100644
+!5 = !{!"int*", !"int*", !"int", !"char"}
+!6 = !{!"", !"", !"", !""}
--
-1.8.3.1
+2.7.4
diff --git a/dynamic-layers/clang-layer/recipes-devtools/clang/files/0001-skip-building-tests.patch b/dynamic-layers/clang-layer/recipes-devtools/clang/files/0001-skip-building-tests.patch
new file mode 100644
index 0000000..c58e7d9
--- /dev/null
+++ b/dynamic-layers/clang-layer/recipes-devtools/clang/files/0001-skip-building-tests.patch
@@ -0,0 +1,50 @@
+From 91db4c3cf7f290a3cab5caa316fc25a60dd409f1 Mon Sep 17 00:00:00 2001
+From: Anuj Mittal <anuj.mittal at intel.com>
+Date: Fri, 16 Aug 2019 20:25:16 +0800
+Subject: [PATCH] llvm-spirv: skip including tests
+
+Some of these need clang to be built and since we're building this in-tree,
+that leads to problems when compiling libcxx, compiler-rt which aren't built
+in-tree.
+
+Instead of using SPIRV_SKIP_CLANG_BUILD to skip clang build and adding this to
+all components, disable the building of tests altogether.
+
+Upstream-Status: Inappropriate
+
+Signed-off-by: Anuj Mittal <anuj.mittal at intel.com>
+---
+ CMakeLists.txt | 10 ----------
+ 1 file changed, 10 deletions(-)
+
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index d632a50..81ddf62 100644
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -15,13 +15,6 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
+ set(CMAKE_CXX_STANDARD 11)
+ set(CMAKE_CXX_STANDARD_REQUIRED ON)
+
+- if(LLVM_INCLUDE_TESTS)
+- set(LLVM_TEST_COMPONENTS
+- llvm-as
+- llvm-dis
+- )
+- endif(LLVM_INCLUDE_TESTS)
+-
+ find_package(LLVM 8.0.0 REQUIRED
+ COMPONENTS
+ Analysis
+@@ -56,9 +49,6 @@ set(LLVM_SPIRV_INCLUDE_DIRS ${CMAKE_CURRENT_SOURCE_DIR}/include)
+
+ add_subdirectory(lib/SPIRV)
+ add_subdirectory(tools/llvm-spirv)
+-if(LLVM_INCLUDE_TESTS)
+- add_subdirectory(test)
+-endif(LLVM_INCLUDE_TESTS)
+
+ install(
+ FILES
+--
+2.7.4
+
diff --git a/dynamic-layers/clang-layer/recipes-devtools/clang/files/0002-Remove-extra-semicolon.patch b/dynamic-layers/clang-layer/recipes-devtools/clang/files/0002-Remove-extra-semicolon.patch
new file mode 100644
index 0000000..5f5a957
--- /dev/null
+++ b/dynamic-layers/clang-layer/recipes-devtools/clang/files/0002-Remove-extra-semicolon.patch
@@ -0,0 +1,27 @@
+From a2c093c8db7b4e3a5612d0fcce9e4fd1756d6e4b Mon Sep 17 00:00:00 2001
+From: Alexey Sotkin <alexey.sotkin at intel.com>
+Date: Mon, 5 Aug 2019 18:18:01 +0300
+Subject: [PATCH] Remove extra semicolon
+
+Upstream-Status: Backport
+Signed-off-by: Anuj Mittal <anuj.mittal at intel.com>
+---
+ lib/SPIRV/libSPIRV/SPIRVEnum.h | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/lib/SPIRV/libSPIRV/SPIRVEnum.h b/lib/SPIRV/libSPIRV/SPIRVEnum.h
+index c93a484..3a071e7 100644
+--- a/lib/SPIRV/libSPIRV/SPIRVEnum.h
++++ b/lib/SPIRV/libSPIRV/SPIRVEnum.h
+@@ -124,7 +124,7 @@ template <> inline void SPIRVMap<SPIRVExtensionKind, std::string>::init() {
+ add(SPV_INTEL_device_side_avc_motion_estimation,
+ "SPV_INTEL_device_side_avc_motion_estimation");
+ add(SPV_KHR_no_integer_wrap_decoration, "SPV_KHR_no_integer_wrap_decoration");
+-};
++}
+
+ template <> inline void SPIRVMap<SPIRVExtInstSetKind, std::string>::init() {
+ add(SPIRVEIS_OpenCL, "OpenCL.std");
+--
+2.7.4
+
diff --git a/dynamic-layers/clang-layer/recipes-devtools/clang/llvm-project-source.bbappend b/dynamic-layers/clang-layer/recipes-devtools/clang/llvm-project-source.bbappend
index 7631c02..50ab25e 100644
--- a/dynamic-layers/clang-layer/recipes-devtools/clang/llvm-project-source.bbappend
+++ b/dynamic-layers/clang-layer/recipes-devtools/clang/llvm-project-source.bbappend
@@ -7,6 +7,8 @@ SRC_URI_append_intel-x86-common = " \
file://0001-dont-export-targets-for-binaries.patch \
git://github.com/KhronosGroup/SPIRV-LLVM-Translator.git;protocol=https;branch=llvm_release_80;destsuffix=git/llvm/projects/llvm-spirv;name=spirv \
file://0001-Update-LowerOpenCL-pass-to-handle-new-blocks-represn.patch;patchdir=llvm/projects/llvm-spirv \
+ file://0002-Remove-extra-semicolon.patch;patchdir=llvm/projects/llvm-spirv \
+ file://0001-skip-building-tests.patch;patchdir=llvm/projects/llvm-spirv \
"
-SRCREV_spirv = "bd0f28fb92061d49c0f120b4dac3fd8956006745"
+SRCREV_spirv = "1d48cd84d04a2f60b43ea3f66eb7c86f4e5973a9"
--
2.7.4
More information about the meta-intel
mailing list