[SYCL] Refactor address space handling in CodeGen library #2864

bader · 2020-12-04T16:07:00Z

This patch replaced SYCL specific patches aiming to implement address space handling with the approach proposed in the code review comments for https://reviews.llvm.org/D89909.
New approach: overload a couple of methods in SPIRTargetCodeGenInfo to set opencl_global address space for global variables and opencl_private for variables on the stack. The rest methods of the CodeGen library should already use these hooks and address space map to handle any inconsistencies in address spaces.

There are a lot of customizations in CodeGen library enabling generic address space as default for SYCL mode. Based on feedback from the community there is a better way to get similar results. CodeGen TargetInfo object is be used to set default address space for global variables and stack allocations.

This fixes clang/test/CodeGenSYCL/const-wg-init.cpp

bader · 2020-12-04T16:21:34Z

Check-clang results on my local machine:

Failed Tests (24):
Clang :: CodeGenSYCL/accessor_inheritance.cpp
Clang :: CodeGenSYCL/address-space-cond-op.cpp
Clang :: CodeGenSYCL/address-space-new.cpp
Clang :: CodeGenSYCL/address-space-of-returns.cpp
Clang :: CodeGenSYCL/address-space-parameter-conversions.cpp
Clang :: CodeGenSYCL/basic-kernel-wrapper.cpp
Clang :: CodeGenSYCL/device-variables.cpp
Clang :: CodeGenSYCL/inheritance.cpp
Clang :: CodeGenSYCL/inline_asm.cpp
Clang :: CodeGenSYCL/intel-fpga-ivdep-array.cpp
Clang :: CodeGenSYCL/intel-fpga-ivdep-embedded-loops.cpp
Clang :: CodeGenSYCL/intel-fpga-ivdep-global.cpp
Clang :: CodeGenSYCL/intel-fpga-local.cpp
Clang :: CodeGenSYCL/intel-fpga-loops.cpp
Clang :: CodeGenSYCL/intel-fpga-mem-builtin.cpp
Clang :: CodeGenSYCL/intel-fpga-reg.cpp
Clang :: CodeGenSYCL/kernel-param-acc-array.cpp
Clang :: CodeGenSYCL/kernel-param-member-acc-array.cpp
Clang :: CodeGenSYCL/kernel-param-pod-array.cpp
Clang :: CodeGenSYCL/sampler.cpp
Clang :: CodeGenSYCL/spir-calling-conv.cpp
Clang :: CodeGenSYCL/spir-enum.cpp
Clang :: CodeGenSYCL/sycl-device-alias.cpp
Clang :: CodeGenSYCL/unique-stable-name.cpp

Most of the tests fail due to invalidated FileCheck checks, but the are a few compiler crashes:

CodeGenSYCL/sycl-device-alias.cpp crashes inside EmitAliasDefinition due to invalid cast. @premanandrao, you seem to add this code, please, take a look at this crash.
CodeGenSYCL/intel-fpga-local.cpp crashes inside EmitAutoVarAlloca due to invalid cast. @premanandrao, could you investigate this as well, please?

In addition to check-sycl, I see compiler crash on check-sycl suite:

on-device/hier_par/hier_par_wgscope_O0.cpp crashes inside SYCLLowerWGScopeLegacyPass::runOnFunction due to UNREACHABLE executed at llvm/lib/SYCLLowerIR/LowerWGScope.cpp:211
This issue impacts also on-device/basic_tests/stream/stream.cpp, on-device/hier_par/hier_par_basic.cpp, and on-device/hier_par/hier_par_wgscope.cpp. @kbobrovs, @againull, please, take a look.

bader · 2020-12-04T19:18:35Z

/summary:run

premanandrao · 2020-12-07T19:14:20Z

CodeGenSYCL/sycl-device-alias.cpp crashes inside EmitAliasDefinition due to invalid cast. @premanandrao, you seem to add this code, please, take a look at this crash.

@bader, the difference seems to be that before, we created:

(gdb) p Aliasee->dump()
@vint = addrspace(1) constant i32 7, align 4
$6 = void

Whereas, now we create:

(gdb) p Aliasee->dump()
i32* addrspacecast (i32 addrspace(1)* @vint to i32*)
$16 = void

This is after the call to:
Aliasee = GetOrCreateLLVMGlobal(AA->getAliasee(), DeclTy->getPointerTo(AS), /*D=*/nullptr);
Is this change expected?

bader · 2020-12-07T19:33:54Z

This looks like a bug. I think the address space for the type must be set like this:

llvm/clang/lib/CodeGen/CodeGenModule.cpp

Lines 4064 to 4065 in 20cb0da

    
           llvm::PointerType *PTy = 
        
             llvm::PointerType::get(Ty, getContext().getTargetAddressSpace(ASTTy));

bader · 2020-12-07T19:43:22Z

Actually the fix might require calling GetGlobalVarAddressSpace...

Anyway SYCL allows casts from "named" address spaces (i.e. global, local, private in AST or (1), (3), empty - no address space in LLVM IR for SPIR) to "Default" in AST or (4) in LLVM IR for SPIR.

So casting from (1) to empty (i.e. 0) address space in LLVM IR is not valid for SPIR.

premanandrao · 2020-12-18T20:48:52Z

CodeGenSYCL/sycl-device-alias.cpp crashes inside EmitAliasDefinition due to invalid cast. @premanandrao, you seem to add this code, please, take a look at this crash.

For this failure, this change in CodeGenModule.cpp fixes it:

@@ -5064,7 +5048,8 @@ void CodeGenModule::EmitAliasDefinition(GlobalDecl GD) {
     LT = getFunctionLinkage(GD);
     AS = Aliasee->getType()->getPointerAddressSpace();
   } else {
-    AS = ArgInfoAddressSpace(GetGlobalVarAddressSpace(/*D=*/nullptr));
+    const auto *VarD = cast<VarDecl>(GD.getDecl());
+    AS = ArgInfoAddressSpace(GetGlobalVarAddressSpace(VarD));
     Aliasee = GetOrCreateLLVMGlobal(AA->getAliasee(), DeclTy->getPointerTo(AS),
                                     /*D=*/nullptr);
     if (const auto *VD = dyn_cast<VarDecl>(GD.getDecl()))

CodeGenSYCL/intel-fpga-local.cpp crashes inside EmitAutoVarAlloca due to invalid cast. @premanandrao, could you investigate this as well, please?

I am not sure how to fix this failure. Before your change, this is what CGDecl.cpp:~1645 show after:
llvm::Value *V = address.getPointer();

(gdb) p V->dump()
  %numbanks = alloca i32, align 4

After your change, this is what is shows:

(gdb) p V->dump()
  %numbanks.ascast = addrspacecast i32* %numbanks to i32 addrspace(4)*

bader · 2020-12-29T17:46:17Z

CodeGenSYCL/sycl-device-alias.cpp crashes inside EmitAliasDefinition due to invalid cast. @premanandrao, you seem to add this code, please, take a look at this crash.

For this failure, this change in CodeGenModule.cpp fixes it:

@@ -5064,7 +5048,8 @@ void CodeGenModule::EmitAliasDefinition(GlobalDecl GD) {
     LT = getFunctionLinkage(GD);
     AS = Aliasee->getType()->getPointerAddressSpace();
   } else {
-    AS = ArgInfoAddressSpace(GetGlobalVarAddressSpace(/*D=*/nullptr));
+    const auto *VarD = cast<VarDecl>(GD.getDecl());
+    AS = ArgInfoAddressSpace(GetGlobalVarAddressSpace(VarD));
     Aliasee = GetOrCreateLLVMGlobal(AA->getAliasee(), DeclTy->getPointerTo(AS),
                                     /*D=*/nullptr);
     if (const auto *VD = dyn_cast<VarDecl>(GD.getDecl()))

Thanks. I applied this change.

CodeGenSYCL/intel-fpga-local.cpp crashes inside EmitAutoVarAlloca due to invalid cast. @premanandrao, could you investigate this as well, please?

I am not sure how to fix this failure. Before your change, this is what CGDecl.cpp:~1645 show after:
llvm::Value *V = address.getPointer();
(gdb) p V->dump()
  %numbanks = alloca i32, align 4
After your change, this is what is shows:
(gdb) p V->dump()
  %numbanks.ascast = addrspacecast i32* %numbanks to i32 addrspace(4)*

Right. This is expected. With this change addrspacecast is inserted more aggressively compared to the current approach. I'm not sure what is the reason of the crash, but the difference is that now we have one more instruction - %numbanks.ascast = addrspacecast i32* %numbanks to i32 addrspace(4)* right after %numbanks = alloca i32, align 4.

kbobrovs

Please add the test case from #3005 (https://github.com/intel/llvm/blob/979f66c7c79dd1f56ecaa74188e5f156a75c8125/clang/test/CodeGenSYCL/local_var_big_init_as.cpp) to this PR

kbobrovs · 2021-01-12T00:44:52Z

@bader, could you also please run https://github.com/intel/llvm-test-suite/tree/intel/SYCL/ESIMD - vector BE is very sensitive to AS changes.

kbobrovs · 2021-01-12T00:53:07Z

This issue impacts also on-device/basic_tests/stream/stream.cpp, on-device/hier_par/hier_par_basic.cpp, and on-device/hier_par/hier_par_wgscope.cpp. @kbobrovs, @againull, please, take a look.

@againull - please let me know if you can take a look.

Clang creates a global variable to initialize arrays with 16+ elements. This change fixes address space for such global variables, which might be target dependent.

AaronBallman

I'm not super familiar with the CodeGen side of things, but this looks mostly reasonable to me (aside from some very minor nits).

clang/lib/Basic/Targets/SPIR.h

clang/lib/CodeGen/CGExprScalar.cpp

clang/lib/CodeGen/CodeGenModule.cpp

clang/lib/CodeGen/TargetInfo.cpp

AaronBallman · 2021-02-01T13:08:38Z

clang/lib/CodeGen/TargetInfo.cpp

+    return AddrSpace;
+
+  if (CGM.isTypeConstant(D->getType(), false)) {
+    if (auto ConstAS = CGM.getTarget().getConstantAddressSpace())


Can you spell out auto?

This is a copy-paste from AMDGPUTargetCodeGenInfo::getGlobalVarAddressSpace.

All the places where getConstantAddressSpace is called, authors use auto: CodeGenModule::getStringLiteralAddressSpace, castStringLiteralToDefaultAddressSpace and createReferenceTemporary.

I'm okay to apply this comment, but I think its should be applied to all instances. Right?
If so, I can create an NFC patch and merge it before - https://reviews.llvm.org/D89909. Does it sound okay to you?

Your call -- I called it out because the type wasn't obvious from the context and on line 9996 we're doing LangAS AddrSpace = D->getType().getAddressSpace(); with the type spelled out, so I figured we might as well spell it out. However, given how trivial the use is on the next line, it's not critical.

I'll ask John McCall (CodeGen code owner) in https://reviews.llvm.org/D89909 about this.

llvm/lib/SYCLLowerIR/LowerWGScope.cpp

bader

A little be of context for reviewers.

As noted in the description this PR reverts existing approach for handling AS mismatches in CodeGen, which brought a couple of whitespace changes breaking code formatting guidelines, but it's because llorg code violates them.

In addition to that, one of the patches was uploaded by @againull in a separate PR - #3120. I hope his patch will go first and all changes in llvm/* directory will go away from this patch. Alternative solution would be close #3120 and commit it in scope of this PR. @againull, which way do you prefer?

clang/lib/CodeGen/CGExprScalar.cpp

clang/lib/Basic/Targets/SPIR.h

clang/lib/CodeGen/CodeGenModule.cpp

clang/lib/CodeGen/TargetInfo.cpp

bader · 2021-02-01T14:05:55Z

clang/lib/CodeGen/TargetInfo.cpp

+    return AddrSpace;
+
+  if (CGM.isTypeConstant(D->getType(), false)) {
+    if (auto ConstAS = CGM.getTarget().getConstantAddressSpace())


This is a copy-paste from AMDGPUTargetCodeGenInfo::getGlobalVarAddressSpace.

All the places where getConstantAddressSpace is called, authors use auto: CodeGenModule::getStringLiteralAddressSpace, castStringLiteralToDefaultAddressSpace and createReferenceTemporary.

I'm okay to apply this comment, but I think its should be applied to all instances. Right?
If so, I can create an NFC patch and merge it before - https://reviews.llvm.org/D89909. Does it sound okay to you?

llvm/lib/SYCLLowerIR/LowerWGScope.cpp

againull · 2021-02-02T04:23:17Z

A little be of context for reviewers.

As noted in the description this PR reverts existing approach for handling AS mismatches in CodeGen, which brought a couple of whitespace changes breaking code formatting guidelines, but it's because llorg code violates them.

In addition to that, one of the patches was uploaded by @againull in a separate PR - #3120. I hope his patch will go first and all changes in llvm/* directory will go away from this patch. Alternative solution would be close #3120 and commit it in scope of this PR. @againull, which way do you prefer?

@bader , I have merged PR#3120.

This reverts commit 2b2ded9.

This reverts commit 581330c.

bader · 2021-02-03T08:12:17Z

/summary:run

bader · 2021-02-04T05:54:58Z

@elizabethandrews, @premanandrao, could you take a look at this patch, please?
Pre-commit test status:

clang-format-check fails because this patch reverts clang-format change we applied to llorg code, so please, ignore this failure.
Jenkins/Summary - I believe this failure is caused by a bug in another component and I'm going to report it separately.

premanandrao

Looks okay to me @bader.

premanandrao · 2021-02-04T16:41:15Z

@elizabethandrews, @premanandrao, could you take a look at this patch, please?

Also, @elizabethandrews is traveling/on vacation.

bader · 2021-02-04T16:42:13Z

Thanks!

erichkeane · 2021-02-11T16:15:10Z

So I see that this causes a crash of a lit test. It looks like the cast you added in CodeGenModule is invalid.

FAIL: Clang :: CodeGen/alias.c (3169 of 27447)
******************** TEST 'Clang :: CodeGen/alias.c' FAILED ********************
Script:
--
: 'RUN: at line 2';   /iusers/ekeane1/workspaces/sycl/build/bin/clang -cc1 -internal-isystem /export/iusers/ekeane1/workspaces/sycl/build/lib/clang/13.0.0/include -nostdsysteminc -triple i386-pc-linux-gnu -emit-llvm -o - /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c | /iusers/ekeane1/workspaces/sycl/build/bin/FileCheck --allow-unused-prefixes=false -check-prefix=CHECKBASIC /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c
: 'RUN: at line 3';   /iusers/ekeane1/workspaces/sycl/build/bin/clang -cc1 -internal-isystem /export/iusers/ekeane1/workspaces/sycl/build/lib/clang/13.0.0/include -nostdsysteminc -triple armv7a-eabi -mfloat-abi hard -emit-llvm -o - /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c | /iusers/ekeane1/workspaces/sycl/build/bin/FileCheck --allow-unused-prefixes=false -check-prefix=CHECKCC /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c
: 'RUN: at line 4';   /iusers/ekeane1/workspaces/sycl/build/bin/clang -cc1 -internal-isystem /export/iusers/ekeane1/workspaces/sycl/build/lib/clang/13.0.0/include -nostdsysteminc -triple armv7a-eabi -mfloat-abi hard -S -o - /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c | /iusers/ekeane1/workspaces/sycl/build/bin/FileCheck --allow-unused-prefixes=false -check-prefix=CHECKASM /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c
: 'RUN: at line 5';   /iusers/ekeane1/workspaces/sycl/build/bin/clang -cc1 -internal-isystem /export/iusers/ekeane1/workspaces/sycl/build/lib/clang/13.0.0/include -nostdsysteminc -triple aarch64-linux-gnu -emit-llvm -o - /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c | /iusers/ekeane1/workspaces/sycl/build/bin/FileCheck --allow-unused-prefixes=false -check-prefix=CHECKGLOBALS /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c
--
Exit Code: 2

Command Output (stderr):
--
/export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c:117:24: warning: declaration of 'enum undeclared_type' will not be visible outside of this function [-Wvisibility]
void test13_alias(enum undeclared_type y) __attribute__((alias ("test13")));
                       ^
clang: /iusers/ekeane1/workspaces/sycl/llvm/include/llvm/Support/Casting.h:269: typename llvm::cast_retty<X, Y*>::ret_type llvm::cast(Y*) [with X = clang::VarDecl; Y = const clang::Decl; typename llvm::cast_retty<X, Y*>::ret_type = const clang::VarDecl*]: Assertion `isa<X>(Val) && "cast<Ty>() argument of incompatible type!"' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.      Program arguments: /iusers/ekeane1/workspaces/sycl/build/bin/clang -cc1 -internal-isystem /export/iusers/ekeane1/workspaces/sycl/build/lib/clang/13.0.0/include -nostdsysteminc -triple i386-pc-linux-gnu -emit-llvm -o - /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c
1.      <eof> parser at end of file
2.      /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c:117:6: LLVM IR generation of declaration 'test13_alias'
 #0 0x000000000527a969 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) /iusers/ekeane1/workspaces/sycl/llvm/lib/Support/Unix/Signals.inc:565:22
 #1 0x000000000527aa20 PrintStackTraceSignalHandler(void*) /iusers/ekeane1/workspaces/sycl/llvm/lib/Support/Unix/Signals.inc:632:1
 #2 0x00000000052789e7 llvm::sys::RunSignalHandlers() /iusers/ekeane1/workspaces/sycl/llvm/lib/Support/Signals.cpp:71:20
 #3 0x000000000527a3b1 SignalHandler(int) /iusers/ekeane1/workspaces/sycl/llvm/lib/Support/Unix/Signals.inc:407:1
 #4 0x00007f74aab315e0 __restore_rt (/lib64/libpthread.so.0+0xf5e0)
 #5 0x00007f74a96aa1f7 raise (/lib64/libc.so.6+0x351f7)
 #6 0x00007f74a96ab8e8 abort (/lib64/libc.so.6+0x368e8)
 #7 0x00007f74a96a3266 __assert_fail_base (/lib64/libc.so.6+0x2e266)
 #8 0x00007f74a96a3312 (/lib64/libc.so.6+0x2e312)
 #9 0x0000000005676fba llvm::cast_retty<clang::VarDecl, clang::Decl const*>::ret_type llvm::cast<clang::VarDecl, clang::Decl const>(clang::Decl const*) /iusers/ekeane1/workspaces/sycl/llvm/include/llvm/Support/Casting.h:269:3
#10 0x000000000579a6b4 clang::CodeGen::CodeGenModule::EmitAliasDefinition(clang::GlobalDecl) /iusers/ekeane1/workspaces/sycl/clang/lib/CodeGen/CodeGenModule.cpp:5103:37
#11 0x0000000005791093 clang::CodeGen::CodeGenModule::EmitGlobal(clang::GlobalDecl) /iusers/ekeane1/workspaces/sycl/clang/lib/CodeGen/CodeGenModule.cpp:2874:36
#12 0x000000000579e08a clang::CodeGen::CodeGenModule::EmitTopLevelDecl(clang::Decl*) (.localalias.1) /iusers/ekeane1/workspaces/sycl/clang/lib/CodeGen/CodeGenModule.cpp:5877:37
#13 0x00000000068090c6 (anonymous namespace)::CodeGeneratorImpl::HandleTopLevelDecl(clang::DeclGroupRef) /iusers/ekeane1/workspaces/sycl/clang/lib/CodeGen/ModuleBuilder.cpp:169:7
#14 0x00000000068027c6 clang::BackendConsumer::HandleTopLevelDecl(clang::DeclGroupRef) /iusers/ekeane1/workspaces/sycl/clang/lib/CodeGen/CodeGenAction.cpp:222:11
#15 0x000000000813f73c clang::ParseAST(clang::Sema&, bool, bool) /iusers/ekeane1/workspaces/sycl/clang/lib/Parse/ParseAST.cpp:162:20
#16 0x0000000005ee3331 clang::ASTFrontendAction::ExecuteAction() /iusers/ekeane1/workspaces/sycl/clang/lib/Frontend/FrontendAction.cpp:1056:11
#17 0x0000000006800151 clang::CodeGenAction::ExecuteAction() /iusers/ekeane1/workspaces/sycl/clang/lib/CodeGen/CodeGenAction.cpp:1096:5
#18 0x0000000005ee2c5d clang::FrontendAction::Execute() /iusers/ekeane1/workspaces/sycl/clang/lib/Frontend/FrontendAction.cpp:953:38
#19 0x0000000005e3dd97 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) /iusers/ekeane1/workspaces/sycl/clang/lib/Frontend/CompilerInstance.cpp:951:42
#20 0x000000000606073a clang::ExecuteCompilerInvocation(clang::CompilerInstance*) /iusers/ekeane1/workspaces/sycl/clang/lib/FrontendTool/ExecuteCompilerInvocation.cpp:278:38
#21 0x000000000254fd3d cc1_main(llvm::ArrayRef<char const*>, char const*, void*) /iusers/ekeane1/workspaces/sycl/clang/tools/driver/cc1_main.cpp:240:40
#22 0x0000000002544d49 ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&) /iusers/ekeane1/workspaces/sycl/clang/tools/driver/driver.cpp:330:20
#23 0x000000000254544a main /iusers/ekeane1/workspaces/sycl/clang/tools/driver/driver.cpp:407:26
#24 0x00007f74a9696c05 __libc_start_main (/lib64/libc.so.6+0x21c05)
#25 0x00000000025435f9 _start (/iusers/ekeane1/workspaces/sycl/build/bin/clang+0x25435f9)
FileCheck error: '<stdin>' is empty.
FileCheck command line:  /iusers/ekeane1/workspaces/sycl/build/bin/FileCheck --allow-unused-prefixes=false -check-prefix=CHECKBASIC /export/iusers/ekeane1/workspaces/sycl/clang/test/CodeGen/alias.c

bader · 2021-02-11T16:35:58Z

@erichkeane, could you provide more details on how to reproduce this issue, please? I don't see this failure locally and all CI checks pass.

erichkeane · 2021-02-11T16:37:45Z

@erichkeane, could you provide more details on how to reproduce this issue, please? I don't see this failure locally and all CI checks pass.

it is happening with a check-clang on a clean version of scyl HEAD using a debug compiler.

AaronBallman · 2021-02-11T17:01:54Z

@erichkeane, could you provide more details on how to reproduce this issue, please? I don't see this failure locally and all CI checks pass.

it is happening with a check-clang on a clean version of scyl HEAD using a debug compiler.

If all CI checks are passing... are we doing a release test rather than a release + asserts test?

bader · 2021-02-11T17:03:12Z

We are doing Release and Release + Asserts.
I trying to check Debug configuration locally.

bader · 2021-02-11T17:46:14Z

// REQUIRES: arm-registered-target

That's the reason it's not caught by CI (and my local testing).
I don't recall any problems with my testing for https://reviews.llvm.org/D89909, which I do for all the targets, but it might be caused by sycl branch patches, not upstreamed to llorg yet.

bader · 2021-02-11T18:24:44Z

Fix - #3206.

Checks for Clang::CodeGenSYCL/intel-fpga-reg.cpp test were autogenerated to speed up fixing of LIT tests failures after a patch "[SYCL] Refactor address space handling in CodeGen library (#2864)". Reorganize test code and tidy up checks to improve readability. Signed-off-by: Mikhail Lychkov <mikhail.lychkov@intel.com>

bader added 3 commits December 4, 2020 17:26

Update clang/test/CodeGenSYCL/union-kernel-param.cpp checks.

848a65b

Reapply patch for invariant intrinsics.

73e99fe

This fixes clang/test/CodeGenSYCL/const-wg-init.cpp

bader added the upstream This change is related to upstreaming SYCL support to llorg. label Dec 4, 2020

bader requested review from premanandrao and erichkeane December 4, 2020 16:07

bader added 2 commits December 4, 2020 19:57

Fix formatting.

7e7312f

Update checks in a few tests.

f7810f9

Update checks in more tests.

c9aaac6

bader added 5 commits December 27, 2020 15:42

Fix CodeGen crash exposed by CodeGenSYCL/sycl-device-alias.cpp

4b42842

Merge remote-tracking branch 'intel/sycl' into codegen-refactor

6637530

Update FileCheck checks.

23f506c

Merge remote-tracking branch 'intel/sycl' into codegen-refactor

e49b81f

Apply clang-format to intel-fpga-reg.cpp test.

6a85e7b

bader mentioned this pull request Jan 11, 2021

[SYCL] Put constant initializer list data in non-generic addr space. #3005

Merged

bader added 3 commits January 11, 2021 11:37

Allow arbitrary size for the file name parameter.

82f0dd9

Apply clang-format.

3c510c4

Merge remote-tracking branch 'intel/sycl' into codegen-refactor

151ca6d

kbobrovs reviewed Jan 12, 2021

View reviewed changes

bader added 2 commits January 12, 2021 14:19

Set proper address space for auto-generated globals.

3a31ee6

Clang creates a global variable to initialize arrays with 16+ elements. This change fixes address space for such global variables, which might be target dependent.

Minor formatting.

0c8f798

bader requested review from DenisBakhvalov and elizabethandrews as code owners January 30, 2021 13:48

bader requested a review from AaronBallman January 30, 2021 13:49

Apply comments from Aaron.

dd8d914

AaronBallman reviewed Feb 1, 2021

View reviewed changes

bader commented Feb 1, 2021

View reviewed changes

bader added 4 commits February 2, 2021 13:20

Revert "Update pfwg_and_pfwi.ll test."

a32be82

This reverts commit 2b2ded9.

Revert committed code to avoid merge conflicts.

ba8d836

Revert "[SYCL] Handle address space casts in the LowerWGScope"

226d8a6

This reverts commit 581330c.

Merge remote-tracking branch 'intel/sycl' into codegen-refactor

007c602

premanandrao approved these changes Feb 4, 2021

View reviewed changes

bader merged commit 1113d0a into intel:sycl Feb 4, 2021

bader deleted the codegen-refactor branch February 4, 2021 16:43

bader mentioned this pull request Mar 17, 2021

[SYCL] Cast to correct address space when a function object is passed #3303

Merged

mlychkov mentioned this pull request May 25, 2021

[SYCL][FPGA][NFC] Tidy up intel_fpga_reg codegen test #3810

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[SYCL] Refactor address space handling in CodeGen library #2864

[SYCL] Refactor address space handling in CodeGen library #2864

bader commented Dec 4, 2020

bader commented Dec 4, 2020

bader commented Dec 4, 2020

premanandrao commented Dec 7, 2020

bader commented Dec 7, 2020

bader commented Dec 7, 2020

premanandrao commented Dec 18, 2020

bader commented Dec 29, 2020

kbobrovs left a comment

kbobrovs commented Jan 12, 2021

kbobrovs commented Jan 12, 2021

AaronBallman left a comment

AaronBallman Feb 1, 2021

bader Feb 1, 2021

AaronBallman Feb 1, 2021

bader Feb 1, 2021

bader left a comment

bader Feb 1, 2021

againull commented Feb 2, 2021

bader commented Feb 3, 2021

bader commented Feb 4, 2021

premanandrao left a comment

premanandrao commented Feb 4, 2021

bader commented Feb 4, 2021

erichkeane commented Feb 11, 2021 •

edited by bader

Loading

bader commented Feb 11, 2021

erichkeane commented Feb 11, 2021

AaronBallman commented Feb 11, 2021

bader commented Feb 11, 2021

bader commented Feb 11, 2021

bader commented Feb 11, 2021

[SYCL] Refactor address space handling in CodeGen library #2864

[SYCL] Refactor address space handling in CodeGen library #2864

Conversation

bader commented Dec 4, 2020

bader commented Dec 4, 2020

bader commented Dec 4, 2020

premanandrao commented Dec 7, 2020

bader commented Dec 7, 2020

bader commented Dec 7, 2020

premanandrao commented Dec 18, 2020

bader commented Dec 29, 2020

kbobrovs left a comment

Choose a reason for hiding this comment

kbobrovs commented Jan 12, 2021

kbobrovs commented Jan 12, 2021

AaronBallman left a comment

Choose a reason for hiding this comment

AaronBallman Feb 1, 2021

Choose a reason for hiding this comment

bader Feb 1, 2021

Choose a reason for hiding this comment

AaronBallman Feb 1, 2021

Choose a reason for hiding this comment

bader Feb 1, 2021

Choose a reason for hiding this comment

bader left a comment

Choose a reason for hiding this comment

bader Feb 1, 2021

Choose a reason for hiding this comment

againull commented Feb 2, 2021

bader commented Feb 3, 2021

bader commented Feb 4, 2021

premanandrao left a comment

Choose a reason for hiding this comment

premanandrao commented Feb 4, 2021

bader commented Feb 4, 2021

erichkeane commented Feb 11, 2021 • edited by bader Loading

bader commented Feb 11, 2021

erichkeane commented Feb 11, 2021

AaronBallman commented Feb 11, 2021

bader commented Feb 11, 2021

bader commented Feb 11, 2021

bader commented Feb 11, 2021

erichkeane commented Feb 11, 2021 •

edited by bader

Loading