[clang] [llvm] [AMDGPU][LTO] Assume closed world after linking (PR #105845)

[Anshil Gandhi via cfe-commits]

https://github.com/gandhi56 updated 
https://github.com/llvm/llvm-project/pull/105845

>From d4b8e5b213b4ea9b5b615354d264b71ed76508d5 Mon Sep 17 00:00:00 2001
From: Anshil Gandhi 
Date: Thu, 22 Aug 2024 18:57:33 +
Subject: [PATCH] [AMDGPU][LTO] Assume closed world after linking

Change-Id: I7d8fa4251c80a6f815f55a0998677d18ade25b72
---
 clang/test/CodeGenCUDA/gpu-rdc-amdgpu-attrs.cu | 12 
 llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp|  4 
 llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp |  3 ++-
 3 files changed, 18 insertions(+), 1 deletion(-)
 create mode 100644 clang/test/CodeGenCUDA/gpu-rdc-amdgpu-attrs.cu

diff --git a/clang/test/CodeGenCUDA/gpu-rdc-amdgpu-attrs.cu 
b/clang/test/CodeGenCUDA/gpu-rdc-amdgpu-attrs.cu
new file mode 100644
index 00..614917aecc0d60
--- /dev/null
+++ b/clang/test/CodeGenCUDA/gpu-rdc-amdgpu-attrs.cu
@@ -0,0 +1,12 @@
+// RUN: clang -x hip -O3 -fgpu-rdc %s -mllvm -debug-only=amdgpu-attributor -o 
- | FileCheck %s
+
+// CHECK: Module {{.*}} is not assumed to be a closed world
+// CHECK: Module ld-temp.o is assumed to be a closed world
+
+__attribute__((device)) int foo() {
+return 1;
+}
+
+int main() {
+return 0;
+}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp 
b/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
index d65e0ae92308e6..53ee3e42eef4c8 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
@@ -1066,6 +1066,10 @@ static bool runImpl(Module &M, AnalysisGetter &AG, 
TargetMachine &TM,
 
   Attributor A(Functions, InfoCache, AC);
 
+  LLVM_DEBUG(dbgs() << "Module " << M.getName() << " is "
+<< (AC.IsClosedWorldModule ? "" : "not ")
+<< "assumed to be a closed world\n");
+
   for (Function &F : M) {
 if (F.isIntrinsic())
   continue;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp 
b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 7ac7b3315bb972..869afdcc62dbf6 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -761,7 +761,8 @@ void 
AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
 if (EnableLowerModuleLDS)
   PM.addPass(AMDGPULowerModuleLDSPass(*this));
 if (EnableAMDGPUAttributor && Level != OptimizationLevel::O0)
-  PM.addPass(AMDGPUAttributorPass(*this));
+  PM.addPass(
+  AMDGPUAttributorPass(*this, AMDGPUAttributorOptions{true}));
   });
 
   PB.registerRegClassFilterParsingCallback(

___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [AMDGPU][LTO] Assume closed world after linking (PR #105845)

[Anshil Gandhi via cfe-commits]

https://github.com/gandhi56 edited 
https://github.com/llvm/llvm-project/pull/105845
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [AMDGPU][LTO] Assume closed world after linking (PR #105845)

[Anshil Gandhi via cfe-commits]

https://github.com/gandhi56 created 
https://github.com/llvm/llvm-project/pull/105845

Change-Id: I7d8fa4251c80a6f815f55a0998677d18ade25b72


>From 8830b6f390039c9a952a86ea52e8fe9559900448 Mon Sep 17 00:00:00 2001
From: Anshil Gandhi 
Date: Thu, 22 Aug 2024 18:57:33 +
Subject: [PATCH] [AMDGPU][LTO] Assume closed world after linking

Change-Id: I7d8fa4251c80a6f815f55a0998677d18ade25b72
---
 clang/test/CodeGenCUDA/gpu-rdc-amdgpu-attrs.cu | 12 
 llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp|  3 +++
 llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp |  2 +-
 3 files changed, 16 insertions(+), 1 deletion(-)
 create mode 100644 clang/test/CodeGenCUDA/gpu-rdc-amdgpu-attrs.cu

diff --git a/clang/test/CodeGenCUDA/gpu-rdc-amdgpu-attrs.cu 
b/clang/test/CodeGenCUDA/gpu-rdc-amdgpu-attrs.cu
new file mode 100644
index 00..614917aecc0d60
--- /dev/null
+++ b/clang/test/CodeGenCUDA/gpu-rdc-amdgpu-attrs.cu
@@ -0,0 +1,12 @@
+// RUN: clang -x hip -O3 -fgpu-rdc %s -mllvm -debug-only=amdgpu-attributor -o 
- | FileCheck %s
+
+// CHECK: Module {{.*}} is not assumed to be a closed world
+// CHECK: Module ld-temp.o is assumed to be a closed world
+
+__attribute__((device)) int foo() {
+return 1;
+}
+
+int main() {
+return 0;
+}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp 
b/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
index d65e0ae92308e6..c78fc66e41ec58 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
@@ -1066,6 +1066,9 @@ static bool runImpl(Module &M, AnalysisGetter &AG, 
TargetMachine &TM,
 
   Attributor A(Functions, InfoCache, AC);
 
+  LLVM_DEBUG(dbgs() << "Module " << M.getName() << " is " << 
(AC.IsClosedWorldModule ? "" : "not ")
+<< "assumed to be a closed world\n");
+
   for (Function &F : M) {
 if (F.isIntrinsic())
   continue;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp 
b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 7ac7b3315bb972..a4898366a21ee1 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -761,7 +761,7 @@ void 
AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
 if (EnableLowerModuleLDS)
   PM.addPass(AMDGPULowerModuleLDSPass(*this));
 if (EnableAMDGPUAttributor && Level != OptimizationLevel::O0)
-  PM.addPass(AMDGPUAttributorPass(*this));
+  PM.addPass(AMDGPUAttributorPass(*this, AMDGPUAttributorOptions{true} 
));
   });
 
   PB.registerRegClassFilterParsingCallback(

___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] a955a31 - [AMDGPU] Replace target feature for global fadd32

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2023-03-28T15:58:30-06:00
New Revision: a955a31896370b67c6490251eca0095295d55f1f

URL: 
https://github.com/llvm/llvm-project/commit/a955a31896370b67c6490251eca0095295d55f1f
DIFF: 
https://github.com/llvm/llvm-project/commit/a955a31896370b67c6490251eca0095295d55f1f.diff

LOG: [AMDGPU] Replace target feature for global fadd32

Change target feature of __builtin_amdgcn_global_atomic_fadd_f32
to atomic-fadd-rtn-insts. Enable atomic-fadd-rtn-insts for gfx90a,
gfx940 and gfx1100 as they all support the return variant of
`global_atomic_add_f32`.

Fixes https://github.com/llvm/llvm-project/issues/61331.

Reviewed By: rampitec

Differential Revision: https://reviews.llvm.org/D146840

Added: 


Modified: 
clang/include/clang/Basic/BuiltinsAMDGPU.def
clang/lib/Basic/Targets/AMDGPU.cpp
clang/test/CodeGenOpenCL/amdgpu-features.cl
clang/test/CodeGenOpenCL/builtins-amdgcn-fp-atomics-gfx908-err.cl
clang/test/CodeGenOpenCL/builtins-amdgcn-gfx11.cl
clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx90a.cl
clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx940.cl

Removed: 




diff  --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 965bd97a97d79..0196100cccac5 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -214,7 +214,7 @@ TARGET_BUILTIN(__builtin_amdgcn_perm, "UiUiUiUi", "nc", 
"gfx8-insts")
 TARGET_BUILTIN(__builtin_amdgcn_fmed3h, "", "nc", "gfx9-insts")
 
 TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_f64, "dd*1d", "t", 
"gfx90a-insts")
-TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_f32, "ff*1f", "t", 
"gfx90a-insts")
+TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_f32, "ff*1f", "t", 
"atomic-fadd-rtn-insts")
 TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_v2f16, "V2hV2h*1V2h", "t", 
"atomic-buffer-global-pk-add-f16-insts")
 TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fmin_f64, "dd*1d", "t", 
"gfx90a-insts")
 TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fmax_f64, "dd*1d", "t", 
"gfx90a-insts")

diff  --git a/clang/lib/Basic/Targets/AMDGPU.cpp 
b/clang/lib/Basic/Targets/AMDGPU.cpp
index 72dfb07804dff..9b3a0b0f40edb 100644
--- a/clang/lib/Basic/Targets/AMDGPU.cpp
+++ b/clang/lib/Basic/Targets/AMDGPU.cpp
@@ -206,6 +206,7 @@ bool AMDGPUTargetInfo::initFeatureMap(
   Features["gfx10-insts"] = true;
   Features["gfx10-3-insts"] = true;
   Features["gfx11-insts"] = true;
+  Features["atomic-fadd-rtn-insts"] = true;
   break;
 case GK_GFX1036:
 case GK_GFX1035:
@@ -264,6 +265,7 @@ bool AMDGPUTargetInfo::initFeatureMap(
 case GK_GFX90A:
   Features["gfx90a-insts"] = true;
   Features["atomic-buffer-global-pk-add-f16-insts"] = true;
+  Features["atomic-fadd-rtn-insts"] = true;
   [[fallthrough]];
 case GK_GFX908:
   Features["dot3-insts"] = true;

diff  --git a/clang/test/CodeGenOpenCL/amdgpu-features.cl 
b/clang/test/CodeGenOpenCL/amdgpu-features.cl
index 4a4da6b270b9a..e000239cd03fe 100644
--- a/clang/test/CodeGenOpenCL/amdgpu-features.cl
+++ b/clang/test/CodeGenOpenCL/amdgpu-features.cl
@@ -72,9 +72,9 @@
 // GFX906: 
"target-features"="+16-bit-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx9-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64"
 // GFX908: 
"target-features"="+16-bit-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx9-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64"
 // GFX909: 
"target-features"="+16-bit-insts,+ci-insts,+dpp,+gfx8-insts,+gfx9-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64"
-// GFX90A: 
"target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64"
+// GFX90A: 
"target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64"
 // GFX90C: 
"target-features"="+16-bit-insts,+ci-insts,+dpp,+gfx8-insts,+gfx9-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64"
-// GFX940: 
"target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-ds-pk-add-16-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+fp8-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gfx940-insts,+mai-insts,+s-mem

[clang] df0560c - [HIP] Add atomic load, atomic store and atomic cmpxchng_weak builtin support in HIP-clang

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-11-29T12:07:13-07:00
New Revision: df0560ca00182364e0a786d35adb294c3c98dbd0

URL: 
https://github.com/llvm/llvm-project/commit/df0560ca00182364e0a786d35adb294c3c98dbd0
DIFF: 
https://github.com/llvm/llvm-project/commit/df0560ca00182364e0a786d35adb294c3c98dbd0.diff

LOG: [HIP] Add atomic load, atomic store and atomic cmpxchng_weak builtin 
support in HIP-clang

Introduce `__hip_atomic_load`, `__hip_atomic_store` and 
`__hip_atomic_compare_exchange_weak`
builtins in HIP.

Reviewed By: yaxunl

Differential Revision: https://reviews.llvm.org/D114553

Added: 
clang/test/SemaCUDA/atomic-ops.cu

Modified: 
clang/include/clang/AST/Expr.h
clang/include/clang/Basic/Builtins.def
clang/lib/AST/Expr.cpp
clang/lib/AST/StmtPrinter.cpp
clang/lib/CodeGen/CGAtomic.cpp
clang/lib/Sema/SemaChecking.cpp
clang/test/CodeGenCUDA/atomic-ops.cu

Removed: 




diff  --git a/clang/include/clang/AST/Expr.h b/clang/include/clang/AST/Expr.h
index 246585e1205fa..2c63406fba18d 100644
--- a/clang/include/clang/AST/Expr.h
+++ b/clang/include/clang/AST/Expr.h
@@ -6308,6 +6308,7 @@ class AtomicExpr : public Expr {
getOp() == AO__hip_atomic_compare_exchange_strong ||
getOp() == AO__opencl_atomic_compare_exchange_strong ||
getOp() == AO__opencl_atomic_compare_exchange_weak ||
+   getOp() == AO__hip_atomic_compare_exchange_weak ||
getOp() == AO__atomic_compare_exchange ||
getOp() == AO__atomic_compare_exchange_n;
   }
@@ -6342,10 +6343,9 @@ class AtomicExpr : public Expr {
 auto Kind =
 (Op >= AO__opencl_atomic_load && Op <= AO__opencl_atomic_fetch_max)
 ? AtomicScopeModelKind::OpenCL
-: (Op >= AO__hip_atomic_compare_exchange_strong &&
-   Op <= AO__hip_atomic_fetch_max)
-  ? AtomicScopeModelKind::HIP
-  : AtomicScopeModelKind::None;
+: (Op >= AO__hip_atomic_load && Op <= AO__hip_atomic_fetch_max)
+? AtomicScopeModelKind::HIP
+: AtomicScopeModelKind::None;
 return AtomicScopeModel::create(Kind);
   }
 

diff  --git a/clang/include/clang/Basic/Builtins.def 
b/clang/include/clang/Basic/Builtins.def
index 1f7680e0d923c..ad8b66aa490be 100644
--- a/clang/include/clang/Basic/Builtins.def
+++ b/clang/include/clang/Basic/Builtins.def
@@ -855,8 +855,9 @@ ATOMIC_BUILTIN(__atomic_fetch_min, "v.", "t")
 ATOMIC_BUILTIN(__atomic_fetch_max, "v.", "t")
 
 // HIP atomic builtins.
-// FIXME: Is `__hip_atomic_compare_exchange_n` or
-// `__hip_atomic_compare_exchange_weak` needed?
+ATOMIC_BUILTIN(__hip_atomic_load, "v.", "t")
+ATOMIC_BUILTIN(__hip_atomic_store, "v.", "t")
+ATOMIC_BUILTIN(__hip_atomic_compare_exchange_weak, "v.", "t")
 ATOMIC_BUILTIN(__hip_atomic_compare_exchange_strong, "v.", "t")
 ATOMIC_BUILTIN(__hip_atomic_exchange, "v.", "t")
 ATOMIC_BUILTIN(__hip_atomic_fetch_add, "v.", "t")

diff  --git a/clang/lib/AST/Expr.cpp b/clang/lib/AST/Expr.cpp
index ce6e30697f856..d3cb2ff3734cb 100644
--- a/clang/lib/AST/Expr.cpp
+++ b/clang/lib/AST/Expr.cpp
@@ -4681,6 +4681,7 @@ unsigned AtomicExpr::getNumSubExprs(AtomicOp Op) {
 return 2;
 
   case AO__opencl_atomic_load:
+  case AO__hip_atomic_load:
   case AO__c11_atomic_store:
   case AO__c11_atomic_exchange:
   case AO__atomic_load:
@@ -4721,6 +4722,7 @@ unsigned AtomicExpr::getNumSubExprs(AtomicOp Op) {
   case AO__hip_atomic_fetch_min:
   case AO__hip_atomic_fetch_max:
   case AO__opencl_atomic_store:
+  case AO__hip_atomic_store:
   case AO__opencl_atomic_exchange:
   case AO__opencl_atomic_fetch_add:
   case AO__opencl_atomic_fetch_sub:
@@ -4738,6 +4740,7 @@ unsigned AtomicExpr::getNumSubExprs(AtomicOp Op) {
   case AO__hip_atomic_compare_exchange_strong:
   case AO__opencl_atomic_compare_exchange_strong:
   case AO__opencl_atomic_compare_exchange_weak:
+  case AO__hip_atomic_compare_exchange_weak:
   case AO__atomic_compare_exchange:
   case AO__atomic_compare_exchange_n:
 return 6;

diff  --git a/clang/lib/AST/StmtPrinter.cpp b/clang/lib/AST/StmtPrinter.cpp
index fc267d7006a1b..b65a38d1e5665 100644
--- a/clang/lib/AST/StmtPrinter.cpp
+++ b/clang/lib/AST/StmtPrinter.cpp
@@ -1691,7 +1691,8 @@ void StmtPrinter::VisitAtomicExpr(AtomicExpr *Node) {
   PrintExpr(Node->getPtr());
   if (Node->getOp() != AtomicExpr::AO__c11_atomic_load &&
   Node->getOp() != AtomicExpr::AO__atomic_load_n &&
-  Node->getOp() != AtomicExpr::AO__opencl_atomic_load) {
+  Node->getOp() != AtomicExpr::AO__opencl_atomic_load &&
+  Node->getOp() != AtomicExpr::AO__hip_atomic_load) {
 OS << ", ";
 PrintExpr(Node->getVal1());
   }

diff  --git a/clang/lib/CodeGen/CGAtomic.cpp b/clang/lib/CodeGen/CGAtomic.cpp
index 9b507b87213a3..b68e6328acdfd 100644
--- a/clang/lib/CodeGen/CGAtomic.cpp
+++ b/clang/lib/CodeGen/CGAtomic.cpp
@@ -531,6 +531,7 @@ static void EmitAtomicOp(CodeGenFunction

[clang] 0567f03 - [HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined symbols

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-10-18T16:53:15-06:00
New Revision: 0567f0333176e476e15b7f32b463f58f7475ff22

URL: 
https://github.com/llvm/llvm-project/commit/0567f0333176e476e15b7f32b463f58f7475ff22
DIFF: 
https://github.com/llvm/llvm-project/commit/0567f0333176e476e15b7f32b463f58f7475ff22.diff

LOG: [HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined symbols

By default clang emits complete contructors as alias of base constructors if 
they are the same.
The backend is supposed to emit symbols for the alias, otherwise it causes 
undefined symbols.
@yaxunl observed that this issue is related to the llvm options 
`-amdgpu-early-inline-all=true`
and `-amdgpu-function-calls=false`. This issue is resolved by only inlining 
global values
with internal linkage. The `getCalleeFunction()` in AMDGPUResourceUsageAnalysis 
also had
to be extended to support aliases to functions. inline-calls.ll was corrected 
appropriately.

Reviewed By: yaxunl, #amdgpu

Differential Revision: https://reviews.llvm.org/D109707

Added: 
clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu

Modified: 
clang/lib/Driver/ToolChains/Clang.cpp
llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
llvm/test/CodeGen/AMDGPU/inline-calls.ll

Removed: 




diff  --git a/clang/lib/Driver/ToolChains/Clang.cpp 
b/clang/lib/Driver/ToolChains/Clang.cpp
index 83afbc3952d84..316c6026adf5c 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -5089,9 +5089,9 @@ void Clang::ConstructJob(Compilation &C, const JobAction 
&JA,
   }
 
   // Enable -mconstructor-aliases except on darwin, where we have to work 
around
-  // a linker bug (see ), and CUDA/AMDGPU device code,
-  // where aliases aren't supported.
-  if (!RawTriple.isOSDarwin() && !RawTriple.isNVPTX() && !RawTriple.isAMDGPU())
+  // a linker bug (see ), and CUDA device code, where
+  // aliases aren't supported.
+  if (!RawTriple.isOSDarwin() && !RawTriple.isNVPTX())
 CmdArgs.push_back("-mconstructor-aliases");
 
   // Darwin's kernel doesn't support guard variables; just die if we

diff  --git a/clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu 
b/clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu
new file mode 100644
index 0..ec7b7c3b7ff4c
--- /dev/null
+++ b/clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu
@@ -0,0 +1,17 @@
+// REQUIRES: amdgpu-registered-target, clang-driver
+
+// RUN: %clang -target x86_64-unknown-linux-gnu --offload-arch=gfx906 
--cuda-device-only -nogpulib -nogpuinc -x hip -emit-llvm -S -o - %s \
+// RUN:   -fgpu-rdc -O3 -mllvm -amdgpu-early-inline-all=true -mllvm 
-amdgpu-function-calls=false | \
+// RUN:   FileCheck %s
+
+#include "Inputs/cuda.h"
+
+// CHECK: %struct.B = type { i8 }
+struct B {
+
+  // CHECK: @_ZN1BC1Ei = hidden unnamed_addr alias void (%struct.B*, i32), 
void (%struct.B*, i32)* @_ZN1BC2Ei
+  __device__ B(int x);
+};
+
+__device__ B::B(int x) {
+}

diff  --git a/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp 
b/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
index 7ff24d1e9c62b..2e24e9f929d2a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
@@ -15,6 +15,7 @@
 #include "AMDGPU.h"
 #include "AMDGPUTargetMachine.h"
 #include "Utils/AMDGPUBaseInfo.h"
+#include "llvm/CodeGen/CommandFlags.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/CommandLine.h"
@@ -90,9 +91,13 @@ static bool alwaysInlineImpl(Module &M, bool GlobalOpt) {
 
   SmallPtrSet FuncsToAlwaysInline;
   SmallPtrSet FuncsToNoInline;
+  Triple TT(M.getTargetTriple());
 
   for (GlobalAlias &A : M.aliases()) {
 if (Function* F = dyn_cast(A.getAliasee())) {
+  if (TT.getArch() == Triple::amdgcn &&
+  A.getLinkage() != GlobalValue::InternalLinkage)
+continue;
   A.replaceAllUsesWith(F);
   AliasesToRemove.push_back(&A);
 }

diff  --git a/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp 
b/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
index e841e939ef34b..3c5cb6e190850 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
@@ -29,6 +29,8 @@
 #include "SIMachineFunctionInfo.h"
 #include "llvm/Analysis/CallGraph.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalValue.h"
 #include "llvm/Target/TargetMachine.h"
 
 using namespace llvm;
@@ -61,7 +63,8 @@ static const Function *getCalleeFunction(const MachineOperand 
&Op) {
 assert(Op.getImm() == 0);
 return nullptr;
   }
-
+  if (auto *GA = dyn_cast(Op.getGlobal()))
+return cast(GA->getOperand(0));
   return cast(Op.getGlobal());
 }
 

diff  --git a/llvm/test/CodeGen/AMDGPU/inline-calls.ll 
b/llvm/test/CodeGen/AMDGPU/inline-calls

[clang] 1830ec9 - Revert "[HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined symbols"

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-10-15T16:16:18-06:00
New Revision: 1830ec94ac022ae0b6d6876fc2251e6b91e5931e

URL: 
https://github.com/llvm/llvm-project/commit/1830ec94ac022ae0b6d6876fc2251e6b91e5931e
DIFF: 
https://github.com/llvm/llvm-project/commit/1830ec94ac022ae0b6d6876fc2251e6b91e5931e.diff

LOG: Revert "[HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined 
symbols"

This reverts commit 03375a3fb33b11e1249d9c88070b7f33cb97802a.

Added: 


Modified: 
clang/lib/Driver/ToolChains/Clang.cpp
llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
llvm/test/CodeGen/AMDGPU/inline-calls.ll

Removed: 
clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu



diff  --git a/clang/lib/Driver/ToolChains/Clang.cpp 
b/clang/lib/Driver/ToolChains/Clang.cpp
index 316c6026adf5c..83afbc3952d84 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -5089,9 +5089,9 @@ void Clang::ConstructJob(Compilation &C, const JobAction 
&JA,
   }
 
   // Enable -mconstructor-aliases except on darwin, where we have to work 
around
-  // a linker bug (see ), and CUDA device code, where
-  // aliases aren't supported.
-  if (!RawTriple.isOSDarwin() && !RawTriple.isNVPTX())
+  // a linker bug (see ), and CUDA/AMDGPU device code,
+  // where aliases aren't supported.
+  if (!RawTriple.isOSDarwin() && !RawTriple.isNVPTX() && !RawTriple.isAMDGPU())
 CmdArgs.push_back("-mconstructor-aliases");
 
   // Darwin's kernel doesn't support guard variables; just die if we

diff  --git a/clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu 
b/clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu
deleted file mode 100644
index f75088f8e1415..0
--- a/clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu
+++ /dev/null
@@ -1,17 +0,0 @@
-// REQUIRES: amdgpu-registered-target, clang-driver
-
-// RUN: %clang --offload-arch=gfx906 --cuda-device-only -nogpulib -nogpuinc -x 
hip -emit-llvm -S -o - %s \
-// RUN:   -fgpu-rdc -O3 -mllvm -amdgpu-early-inline-all=true -mllvm 
-amdgpu-function-calls=false | \
-// RUN:   FileCheck %s
-
-#include "Inputs/cuda.h"
-
-// CHECK: %struct.B = type { i8 }
-struct B {
-
-  // CHECK: @_ZN1BC1Ei = hidden unnamed_addr alias void (%struct.B*, i32), 
void (%struct.B*, i32)* @_ZN1BC2Ei
-  __device__ B(int x);
-};
-
-__device__ B::B(int x) {
-}

diff  --git a/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp 
b/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
index 2e24e9f929d2a..7ff24d1e9c62b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
@@ -15,7 +15,6 @@
 #include "AMDGPU.h"
 #include "AMDGPUTargetMachine.h"
 #include "Utils/AMDGPUBaseInfo.h"
-#include "llvm/CodeGen/CommandFlags.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/CommandLine.h"
@@ -91,13 +90,9 @@ static bool alwaysInlineImpl(Module &M, bool GlobalOpt) {
 
   SmallPtrSet FuncsToAlwaysInline;
   SmallPtrSet FuncsToNoInline;
-  Triple TT(M.getTargetTriple());
 
   for (GlobalAlias &A : M.aliases()) {
 if (Function* F = dyn_cast(A.getAliasee())) {
-  if (TT.getArch() == Triple::amdgcn &&
-  A.getLinkage() != GlobalValue::InternalLinkage)
-continue;
   A.replaceAllUsesWith(F);
   AliasesToRemove.push_back(&A);
 }

diff  --git a/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp 
b/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
index 3c5cb6e190850..e841e939ef34b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
@@ -29,8 +29,6 @@
 #include "SIMachineFunctionInfo.h"
 #include "llvm/Analysis/CallGraph.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
-#include "llvm/IR/GlobalAlias.h"
-#include "llvm/IR/GlobalValue.h"
 #include "llvm/Target/TargetMachine.h"
 
 using namespace llvm;
@@ -63,8 +61,7 @@ static const Function *getCalleeFunction(const MachineOperand 
&Op) {
 assert(Op.getImm() == 0);
 return nullptr;
   }
-  if (auto *GA = dyn_cast(Op.getGlobal()))
-return cast(GA->getOperand(0));
+
   return cast(Op.getGlobal());
 }
 

diff  --git a/llvm/test/CodeGen/AMDGPU/inline-calls.ll 
b/llvm/test/CodeGen/AMDGPU/inline-calls.ll
index 134cd301b9743..233485a202057 100644
--- a/llvm/test/CodeGen/AMDGPU/inline-calls.ll
+++ b/llvm/test/CodeGen/AMDGPU/inline-calls.ll
@@ -1,6 +1,6 @@
-; RUN: llc -mtriple amdgcn-unknown-linux-gnu -mcpu=tahiti 
-verify-machineinstrs < %s | FileCheck  %s
-; RUN: llc -mtriple amdgcn-unknown-linux-gnu -mcpu=tonga -verify-machineinstrs 
< %s | FileCheck  %s
-; RUN: llc -mtriple r600-unknown-linux-gnu -mcpu=redwood -verify-machineinstrs 
< %s | FileCheck %s --check-prefix=R600
+; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck  
%s
+; RUN: llc -march=amdgcn -mcpu=tonga -veri

[clang] f92db6d - [HIP] Relax conditions for address space cast in builtin args

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-10-15T15:35:52-06:00
New Revision: f92db6d3fff13bdacdf9b24660eb3f3158c58a17

URL: 
https://github.com/llvm/llvm-project/commit/f92db6d3fff13bdacdf9b24660eb3f3158c58a17
DIFF: 
https://github.com/llvm/llvm-project/commit/f92db6d3fff13bdacdf9b24660eb3f3158c58a17.diff

LOG: [HIP] Relax conditions for address space cast in builtin args

Allow (implicit) address space casting between LLVM-equivalent
target address spaces.

Reviewed By: yaxunl, tra

Differential Revision: https://reviews.llvm.org/D111734

Added: 
clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu
clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu

Modified: 
clang/lib/Sema/SemaExpr.cpp

Removed: 




diff  --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
index 94b44714b530d..472b15b9ea06b 100644
--- a/clang/lib/Sema/SemaExpr.cpp
+++ b/clang/lib/Sema/SemaExpr.cpp
@@ -6545,9 +6545,13 @@ ExprResult Sema::BuildCallExpr(Scope *Scope, Expr *Fn, 
SourceLocation LParenLoc,
 auto ArgPtTy = ArgTy->getPointeeType();
 auto ArgAS = ArgPtTy.getAddressSpace();
 
-// Only allow implicit casting from a non-default address space pointee
-// type to a default address space pointee type
-if (ArgAS != LangAS::Default || ParamAS == LangAS::Default)
+// Add address space cast if target address spaces are 
diff erent
+bool NeedImplicitASC = 
+  ParamAS != LangAS::Default &&   // Pointer params in generic AS 
don't need special handling.
+  ( ArgAS == LangAS::Default  ||  // We do allow implicit 
conversion from generic AS 
+  // or from specific AS which has 
target AS matching that of Param.
+  getASTContext().getTargetAddressSpace(ArgAS) == 
getASTContext().getTargetAddressSpace(ParamAS));
+if (!NeedImplicitASC)
   continue;
 
 // First, ensure that the Arg is an RValue.

diff  --git a/clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu 
b/clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu
new file mode 100644
index 0..d15953b3cacaa
--- /dev/null
+++ b/clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu
@@ -0,0 +1,20 @@
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -target-cpu gfx90a -x hip \
+// RUN:  -aux-triple x86_64-unknown-linux-gnu -fcuda-is-device -emit-llvm %s \
+// RUN:  -o - | FileCheck %s
+
+#define __device__ __attribute__((device))
+typedef __attribute__((address_space(3))) float *LP;
+
+// CHECK-LABEL: test_ds_atomic_add_f32
+// CHECK: %[[ADDR_ADDR:.*]] = alloca float*, align 8, addrspace(5)
+// CHECK: %[[ADDR_ADDR_ASCAST_PTR:.*]] = addrspacecast float* addrspace(5)* 
%[[ADDR_ADDR]] to float**
+// CHECK: store float* %addr, float** %[[ADDR_ADDR_ASCAST_PTR]], align 8
+// CHECK: %[[ADDR_ADDR_ASCAST:.*]] = load float*, float** 
%[[ADDR_ADDR_ASCAST_PTR]], align 8
+// CHECK: %[[AS_CAST:.*]] = addrspacecast float* %[[ADDR_ADDR_ASCAST]] to 
float addrspace(3)*
+// CHECK: %3 = call contract float @llvm.amdgcn.ds.fadd.f32(float 
addrspace(3)* %[[AS_CAST]]
+// CHECK: %4 = load float*, float** %rtn.ascast, align 8
+// CHECK: store float %3, float* %4, align 4
+__device__ void test_ds_atomic_add_f32(float *addr, float val) {
+  float *rtn;
+  *rtn = __builtin_amdgcn_ds_faddf((LP)addr, val, 0, 0, 0);
+}

diff  --git a/clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu 
b/clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu
new file mode 100644
index 0..6f1484c68ec71
--- /dev/null
+++ b/clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu
@@ -0,0 +1,12 @@
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -target-cpu gfx90a -x hip \
+// RUN:  -aux-triple x86_64-unknown-linux-gnu -fcuda-is-device %s \
+// RUN:  -fsyntax-only -verify
+// expected-no-diagnostics
+
+#define __device__ __attribute__((device))
+typedef __attribute__((address_space(3))) float *LP;
+
+__device__ void test_ds_atomic_add_f32(float *addr, float val) {
+  float *rtn;
+  *rtn = __builtin_amdgcn_ds_faddf((LP)addr, val, 0, 0, 0);
+}



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] 53fc510 - Revert "[HIP] Relax conditions for address space cast in builtin args"

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-10-15T14:42:28-06:00
New Revision: 53fc5100e07ac078782ffb4e8e2b627c3cc8d046

URL: 
https://github.com/llvm/llvm-project/commit/53fc5100e07ac078782ffb4e8e2b627c3cc8d046
DIFF: 
https://github.com/llvm/llvm-project/commit/53fc5100e07ac078782ffb4e8e2b627c3cc8d046.diff

LOG: Revert "[HIP] Relax conditions for address space cast in builtin args"

This reverts commit 3b48e1170dc623a95ff13a1e34c839cc094bf321.

Added: 


Modified: 
clang/lib/Sema/SemaExpr.cpp

Removed: 
clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu
clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu



diff  --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
index b3211db8df2dc..94b44714b530d 100644
--- a/clang/lib/Sema/SemaExpr.cpp
+++ b/clang/lib/Sema/SemaExpr.cpp
@@ -6545,11 +6545,9 @@ ExprResult Sema::BuildCallExpr(Scope *Scope, Expr *Fn, 
SourceLocation LParenLoc,
 auto ArgPtTy = ArgTy->getPointeeType();
 auto ArgAS = ArgPtTy.getAddressSpace();
 
-// Add address space cast if target address spaces are 
diff erent
-if ((ArgAS != LangAS::Default &&
- getASTContext().getTargetAddressSpace(ArgAS) !=
- getASTContext().getTargetAddressSpace(ParamAS)) ||
-ParamAS == LangAS::Default)
+// Only allow implicit casting from a non-default address space pointee
+// type to a default address space pointee type
+if (ArgAS != LangAS::Default || ParamAS == LangAS::Default)
   continue;
 
 // First, ensure that the Arg is an RValue.

diff  --git a/clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu 
b/clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu
deleted file mode 100644
index d15953b3cacaa..0
--- a/clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu
+++ /dev/null
@@ -1,20 +0,0 @@
-// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -target-cpu gfx90a -x hip \
-// RUN:  -aux-triple x86_64-unknown-linux-gnu -fcuda-is-device -emit-llvm %s \
-// RUN:  -o - | FileCheck %s
-
-#define __device__ __attribute__((device))
-typedef __attribute__((address_space(3))) float *LP;
-
-// CHECK-LABEL: test_ds_atomic_add_f32
-// CHECK: %[[ADDR_ADDR:.*]] = alloca float*, align 8, addrspace(5)
-// CHECK: %[[ADDR_ADDR_ASCAST_PTR:.*]] = addrspacecast float* addrspace(5)* 
%[[ADDR_ADDR]] to float**
-// CHECK: store float* %addr, float** %[[ADDR_ADDR_ASCAST_PTR]], align 8
-// CHECK: %[[ADDR_ADDR_ASCAST:.*]] = load float*, float** 
%[[ADDR_ADDR_ASCAST_PTR]], align 8
-// CHECK: %[[AS_CAST:.*]] = addrspacecast float* %[[ADDR_ADDR_ASCAST]] to 
float addrspace(3)*
-// CHECK: %3 = call contract float @llvm.amdgcn.ds.fadd.f32(float 
addrspace(3)* %[[AS_CAST]]
-// CHECK: %4 = load float*, float** %rtn.ascast, align 8
-// CHECK: store float %3, float* %4, align 4
-__device__ void test_ds_atomic_add_f32(float *addr, float val) {
-  float *rtn;
-  *rtn = __builtin_amdgcn_ds_faddf((LP)addr, val, 0, 0, 0);
-}

diff  --git a/clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu 
b/clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu
deleted file mode 100644
index 6f1484c68ec71..0
--- a/clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu
+++ /dev/null
@@ -1,12 +0,0 @@
-// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -target-cpu gfx90a -x hip \
-// RUN:  -aux-triple x86_64-unknown-linux-gnu -fcuda-is-device %s \
-// RUN:  -fsyntax-only -verify
-// expected-no-diagnostics
-
-#define __device__ __attribute__((device))
-typedef __attribute__((address_space(3))) float *LP;
-
-__device__ void test_ds_atomic_add_f32(float *addr, float val) {
-  float *rtn;
-  *rtn = __builtin_amdgcn_ds_faddf((LP)addr, val, 0, 0, 0);
-}



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] 3b48e11 - [HIP] Relax conditions for address space cast in builtin args

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-10-15T14:06:47-06:00
New Revision: 3b48e1170dc623a95ff13a1e34c839cc094bf321

URL: 
https://github.com/llvm/llvm-project/commit/3b48e1170dc623a95ff13a1e34c839cc094bf321
DIFF: 
https://github.com/llvm/llvm-project/commit/3b48e1170dc623a95ff13a1e34c839cc094bf321.diff

LOG: [HIP] Relax conditions for address space cast in builtin args

Allow (implicit) address space casting between LLVM-equivalent
target address spaces.

Reviewed By: yaxunl

Differential Revision: https://reviews.llvm.org/D111734

Added: 
clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu
clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu

Modified: 
clang/lib/Sema/SemaExpr.cpp

Removed: 




diff  --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
index 94b44714b530d..b3211db8df2dc 100644
--- a/clang/lib/Sema/SemaExpr.cpp
+++ b/clang/lib/Sema/SemaExpr.cpp
@@ -6545,9 +6545,11 @@ ExprResult Sema::BuildCallExpr(Scope *Scope, Expr *Fn, 
SourceLocation LParenLoc,
 auto ArgPtTy = ArgTy->getPointeeType();
 auto ArgAS = ArgPtTy.getAddressSpace();
 
-// Only allow implicit casting from a non-default address space pointee
-// type to a default address space pointee type
-if (ArgAS != LangAS::Default || ParamAS == LangAS::Default)
+// Add address space cast if target address spaces are 
diff erent
+if ((ArgAS != LangAS::Default &&
+ getASTContext().getTargetAddressSpace(ArgAS) !=
+ getASTContext().getTargetAddressSpace(ParamAS)) ||
+ParamAS == LangAS::Default)
   continue;
 
 // First, ensure that the Arg is an RValue.

diff  --git a/clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu 
b/clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu
new file mode 100644
index 0..d15953b3cacaa
--- /dev/null
+++ b/clang/test/CodeGenCUDA/builtins-unsafe-atomics-gfx90a.cu
@@ -0,0 +1,20 @@
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -target-cpu gfx90a -x hip \
+// RUN:  -aux-triple x86_64-unknown-linux-gnu -fcuda-is-device -emit-llvm %s \
+// RUN:  -o - | FileCheck %s
+
+#define __device__ __attribute__((device))
+typedef __attribute__((address_space(3))) float *LP;
+
+// CHECK-LABEL: test_ds_atomic_add_f32
+// CHECK: %[[ADDR_ADDR:.*]] = alloca float*, align 8, addrspace(5)
+// CHECK: %[[ADDR_ADDR_ASCAST_PTR:.*]] = addrspacecast float* addrspace(5)* 
%[[ADDR_ADDR]] to float**
+// CHECK: store float* %addr, float** %[[ADDR_ADDR_ASCAST_PTR]], align 8
+// CHECK: %[[ADDR_ADDR_ASCAST:.*]] = load float*, float** 
%[[ADDR_ADDR_ASCAST_PTR]], align 8
+// CHECK: %[[AS_CAST:.*]] = addrspacecast float* %[[ADDR_ADDR_ASCAST]] to 
float addrspace(3)*
+// CHECK: %3 = call contract float @llvm.amdgcn.ds.fadd.f32(float 
addrspace(3)* %[[AS_CAST]]
+// CHECK: %4 = load float*, float** %rtn.ascast, align 8
+// CHECK: store float %3, float* %4, align 4
+__device__ void test_ds_atomic_add_f32(float *addr, float val) {
+  float *rtn;
+  *rtn = __builtin_amdgcn_ds_faddf((LP)addr, val, 0, 0, 0);
+}

diff  --git a/clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu 
b/clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu
new file mode 100644
index 0..6f1484c68ec71
--- /dev/null
+++ b/clang/test/SemaCUDA/builtins-unsafe-atomics-gfx90a.cu
@@ -0,0 +1,12 @@
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -target-cpu gfx90a -x hip \
+// RUN:  -aux-triple x86_64-unknown-linux-gnu -fcuda-is-device %s \
+// RUN:  -fsyntax-only -verify
+// expected-no-diagnostics
+
+#define __device__ __attribute__((device))
+typedef __attribute__((address_space(3))) float *LP;
+
+__device__ void test_ds_atomic_add_f32(float *addr, float val) {
+  float *rtn;
+  *rtn = __builtin_amdgcn_ds_faddf((LP)addr, val, 0, 0, 0);
+}



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] 03375a3 - [HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined symbols

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-10-15T11:39:15-06:00
New Revision: 03375a3fb33b11e1249d9c88070b7f33cb97802a

URL: 
https://github.com/llvm/llvm-project/commit/03375a3fb33b11e1249d9c88070b7f33cb97802a
DIFF: 
https://github.com/llvm/llvm-project/commit/03375a3fb33b11e1249d9c88070b7f33cb97802a.diff

LOG: [HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined symbols

By default clang emits complete contructors as alias of base constructors if 
they are the same.
The backend is supposed to emit symbols for the alias, otherwise it causes 
undefined symbols.
@yaxunl observed that this issue is related to the llvm options 
`-amdgpu-early-inline-all=true`
and `-amdgpu-function-calls=false`. This issue is resolved by only inlining 
global values
with internal linkage. The `getCalleeFunction()` in AMDGPUResourceUsageAnalysis 
also had
to be extended to support aliases to functions. inline-calls.ll was corrected 
appropriately.

Reviewed By: yaxunl, #amdgpu

Differential Revision: https://reviews.llvm.org/D109707

Added: 
clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu

Modified: 
clang/lib/Driver/ToolChains/Clang.cpp
llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
llvm/test/CodeGen/AMDGPU/inline-calls.ll

Removed: 




diff  --git a/clang/lib/Driver/ToolChains/Clang.cpp 
b/clang/lib/Driver/ToolChains/Clang.cpp
index 83afbc3952d84..316c6026adf5c 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -5089,9 +5089,9 @@ void Clang::ConstructJob(Compilation &C, const JobAction 
&JA,
   }
 
   // Enable -mconstructor-aliases except on darwin, where we have to work 
around
-  // a linker bug (see ), and CUDA/AMDGPU device code,
-  // where aliases aren't supported.
-  if (!RawTriple.isOSDarwin() && !RawTriple.isNVPTX() && !RawTriple.isAMDGPU())
+  // a linker bug (see ), and CUDA device code, where
+  // aliases aren't supported.
+  if (!RawTriple.isOSDarwin() && !RawTriple.isNVPTX())
 CmdArgs.push_back("-mconstructor-aliases");
 
   // Darwin's kernel doesn't support guard variables; just die if we

diff  --git a/clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu 
b/clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu
new file mode 100644
index 0..f75088f8e1415
--- /dev/null
+++ b/clang/test/CodeGenCUDA/amdgpu-alias-undef-symbols.cu
@@ -0,0 +1,17 @@
+// REQUIRES: amdgpu-registered-target, clang-driver
+
+// RUN: %clang --offload-arch=gfx906 --cuda-device-only -nogpulib -nogpuinc -x 
hip -emit-llvm -S -o - %s \
+// RUN:   -fgpu-rdc -O3 -mllvm -amdgpu-early-inline-all=true -mllvm 
-amdgpu-function-calls=false | \
+// RUN:   FileCheck %s
+
+#include "Inputs/cuda.h"
+
+// CHECK: %struct.B = type { i8 }
+struct B {
+
+  // CHECK: @_ZN1BC1Ei = hidden unnamed_addr alias void (%struct.B*, i32), 
void (%struct.B*, i32)* @_ZN1BC2Ei
+  __device__ B(int x);
+};
+
+__device__ B::B(int x) {
+}

diff  --git a/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp 
b/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
index 7ff24d1e9c62b..2e24e9f929d2a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
@@ -15,6 +15,7 @@
 #include "AMDGPU.h"
 #include "AMDGPUTargetMachine.h"
 #include "Utils/AMDGPUBaseInfo.h"
+#include "llvm/CodeGen/CommandFlags.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/CommandLine.h"
@@ -90,9 +91,13 @@ static bool alwaysInlineImpl(Module &M, bool GlobalOpt) {
 
   SmallPtrSet FuncsToAlwaysInline;
   SmallPtrSet FuncsToNoInline;
+  Triple TT(M.getTargetTriple());
 
   for (GlobalAlias &A : M.aliases()) {
 if (Function* F = dyn_cast(A.getAliasee())) {
+  if (TT.getArch() == Triple::amdgcn &&
+  A.getLinkage() != GlobalValue::InternalLinkage)
+continue;
   A.replaceAllUsesWith(F);
   AliasesToRemove.push_back(&A);
 }

diff  --git a/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp 
b/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
index e841e939ef34b..3c5cb6e190850 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
@@ -29,6 +29,8 @@
 #include "SIMachineFunctionInfo.h"
 #include "llvm/Analysis/CallGraph.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalValue.h"
 #include "llvm/Target/TargetMachine.h"
 
 using namespace llvm;
@@ -61,7 +63,8 @@ static const Function *getCalleeFunction(const MachineOperand 
&Op) {
 assert(Op.getImm() == 0);
 return nullptr;
   }
-
+  if (auto *GA = dyn_cast(Op.getGlobal()))
+return cast(GA->getOperand(0));
   return cast(Op.getGlobal());
 }
 

diff  --git a/llvm/test/CodeGen/AMDGPU/inline-calls.ll 
b/llvm/test/CodeGen/AMDGPU/inline-calls.ll
index 233485a202057..134cd301

[clang] 7063ac1 - [HIP] Allow target addr space in target builtins

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-08-19T23:51:58-06:00
New Revision: 7063ac1afa656bdbb851c8ef120ff699c2e98483

URL: 
https://github.com/llvm/llvm-project/commit/7063ac1afa656bdbb851c8ef120ff699c2e98483
DIFF: 
https://github.com/llvm/llvm-project/commit/7063ac1afa656bdbb851c8ef120ff699c2e98483.diff

LOG: [HIP] Allow target addr space in target builtins

This patch allows target specific addr space in target builtins for HIP. It 
inserts implicit addr
space cast for non-generic pointer to generic pointer in general, and inserts 
implicit addr
space cast for generic to non-generic for target builtin arguments only.

It is NFC for non-HIP languages.

Differential Revision: https://reviews.llvm.org/D102405

Added: 


Modified: 
clang/include/clang/AST/Type.h
clang/lib/Basic/Targets/AMDGPU.h
clang/lib/Sema/SemaExpr.cpp
clang/test/CodeGenCUDA/builtins-amdgcn.cu

Removed: 




diff  --git a/clang/include/clang/AST/Type.h b/clang/include/clang/AST/Type.h
index 09e9705bd86b8..fc83c895afa2e 100644
--- a/clang/include/clang/AST/Type.h
+++ b/clang/include/clang/AST/Type.h
@@ -495,7 +495,12 @@ class Qualifiers {
(A == LangAS::Default &&
 (B == LangAS::sycl_private || B == LangAS::sycl_local ||
  B == LangAS::sycl_global || B == LangAS::sycl_global_device ||
- B == LangAS::sycl_global_host));
+ B == LangAS::sycl_global_host)) ||
+   // In HIP device compilation, any cuda address space is allowed
+   // to implicitly cast into the default address space.
+   (A == LangAS::Default &&
+(B == LangAS::cuda_constant || B == LangAS::cuda_device ||
+ B == LangAS::cuda_shared));
   }
 
   /// Returns true if the address space in these qualifiers is equal to or

diff  --git a/clang/lib/Basic/Targets/AMDGPU.h 
b/clang/lib/Basic/Targets/AMDGPU.h
index 2e580ecf24259..77c2c5fd50145 100644
--- a/clang/lib/Basic/Targets/AMDGPU.h
+++ b/clang/lib/Basic/Targets/AMDGPU.h
@@ -352,7 +352,18 @@ class LLVM_LIBRARY_VISIBILITY AMDGPUTargetInfo final : 
public TargetInfo {
   }
 
   LangAS getCUDABuiltinAddressSpace(unsigned AS) const override {
-return LangAS::Default;
+switch (AS) {
+case 0:
+  return LangAS::Default;
+case 1:
+  return LangAS::cuda_device;
+case 3:
+  return LangAS::cuda_shared;
+case 4:
+  return LangAS::cuda_constant;
+default:
+  return getLangASFromTargetAS(AS);
+}
   }
 
   llvm::Optional getConstantAddressSpace() const override {

diff  --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
index 8ef4a9d96320b..5bde87d02877e 100644
--- a/clang/lib/Sema/SemaExpr.cpp
+++ b/clang/lib/Sema/SemaExpr.cpp
@@ -6572,6 +6572,53 @@ ExprResult Sema::BuildCallExpr(Scope *Scope, Expr *Fn, 
SourceLocation LParenLoc,
   return ExprError();
 
 checkDirectCallValidity(*this, Fn, FD, ArgExprs);
+
+// If this expression is a call to a builtin function in HIP device
+// compilation, allow a pointer-type argument to default address space to 
be
+// passed as a pointer-type parameter to a non-default address space.
+// If Arg is declared in the default address space and Param is declared
+// in a non-default address space, perform an implicit address space cast 
to
+// the parameter type.
+if (getLangOpts().HIP && getLangOpts().CUDAIsDevice && FD &&
+FD->getBuiltinID()) {
+  for (unsigned Idx = 0; Idx < FD->param_size(); ++Idx) {
+ParmVarDecl *Param = FD->getParamDecl(Idx);
+if (!ArgExprs[Idx] || !Param || !Param->getType()->isPointerType() ||
+!ArgExprs[Idx]->getType()->isPointerType())
+  continue;
+
+auto ParamAS = Param->getType()->getPointeeType().getAddressSpace();
+auto ArgTy = ArgExprs[Idx]->getType();
+auto ArgPtTy = ArgTy->getPointeeType();
+auto ArgAS = ArgPtTy.getAddressSpace();
+
+// Only allow implicit casting from a non-default address space pointee
+// type to a default address space pointee type
+if (ArgAS != LangAS::Default || ParamAS == LangAS::Default)
+  continue;
+
+// First, ensure that the Arg is an RValue.
+if (ArgExprs[Idx]->isGLValue()) {
+  ArgExprs[Idx] = ImplicitCastExpr::Create(
+  Context, ArgExprs[Idx]->getType(), CK_NoOp, ArgExprs[Idx],
+  nullptr, VK_PRValue, FPOptionsOverride());
+}
+
+// Construct a new arg type with address space of Param
+Qualifiers ArgPtQuals = ArgPtTy.getQualifiers();
+ArgPtQuals.setAddressSpace(ParamAS);
+auto NewArgPtTy =
+Context.getQualifiedType(ArgPtTy.getUnqualifiedType(), ArgPtQuals);
+auto NewArgTy =
+Context.getQualifiedType(Context.getPointerType(NewArgPtTy),
+ ArgTy.getQualifiers());
+
+// Finally perf

[clang] 508b066 - [Remarks] [AMDGPU] Emit optimization remarks for atomics generating hardware instructions

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-08-19T20:51:19-06:00
New Revision: 508b06699a396cc6f2f2602dab350860cb69f087

URL: 
https://github.com/llvm/llvm-project/commit/508b06699a396cc6f2f2602dab350860cb69f087
DIFF: 
https://github.com/llvm/llvm-project/commit/508b06699a396cc6f2f2602dab350860cb69f087.diff

LOG: [Remarks] [AMDGPU] Emit optimization remarks for atomics generating 
hardware instructions

Produce remarks when atomic instructions are expanded into hardware instructions
in SIISelLowering.cpp. Currently, these remarks are only emitted for atomic fadd
instructions.

Differential Revision: https://reviews.llvm.org/D108150

Added: 
clang/test/CodeGenOpenCL/atomics-cas-remarks-gfx90a.cl
clang/test/CodeGenOpenCL/atomics-unsafe-hw-remarks-gfx90a.cl
llvm/test/CodeGen/AMDGPU/atomics-cas-remarks-gfx90a.ll
llvm/test/CodeGen/AMDGPU/atomics-hw-remarks-gfx90a.ll

Modified: 
llvm/lib/CodeGen/AtomicExpandPass.cpp
llvm/lib/Target/AMDGPU/SIISelLowering.cpp

Removed: 
clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
llvm/test/CodeGen/AMDGPU/atomics-remarks-gfx90a.ll



diff  --git a/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl 
b/clang/test/CodeGenOpenCL/atomics-cas-remarks-gfx90a.cl
similarity index 100%
rename from clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
rename to clang/test/CodeGenOpenCL/atomics-cas-remarks-gfx90a.cl

diff  --git a/clang/test/CodeGenOpenCL/atomics-unsafe-hw-remarks-gfx90a.cl 
b/clang/test/CodeGenOpenCL/atomics-unsafe-hw-remarks-gfx90a.cl
new file mode 100644
index 0..ea3324126c209
--- /dev/null
+++ b/clang/test/CodeGenOpenCL/atomics-unsafe-hw-remarks-gfx90a.cl
@@ -0,0 +1,44 @@
+// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple=amdgcn-amd-amdhsa -target-cpu 
gfx90a \
+// RUN: -Rpass=si-lower -munsafe-fp-atomics %s -S -emit-llvm -o - 2>&1 | \
+// RUN: FileCheck %s --check-prefix=GFX90A-HW
+
+// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple=amdgcn-amd-amdhsa -target-cpu 
gfx90a \
+// RUN: -Rpass=si-lower -munsafe-fp-atomics %s -S -o - 2>&1 | \
+// RUN: FileCheck %s --check-prefix=GFX90A-HW-REMARK
+
+
+// REQUIRES: amdgpu-registered-target
+
+typedef enum memory_order {
+  memory_order_relaxed = __ATOMIC_RELAXED,
+  memory_order_acquire = __ATOMIC_ACQUIRE,
+  memory_order_release = __ATOMIC_RELEASE,
+  memory_order_acq_rel = __ATOMIC_ACQ_REL,
+  memory_order_seq_cst = __ATOMIC_SEQ_CST
+} memory_order;
+
+typedef enum memory_scope {
+  memory_scope_work_item = __OPENCL_MEMORY_SCOPE_WORK_ITEM,
+  memory_scope_work_group = __OPENCL_MEMORY_SCOPE_WORK_GROUP,
+  memory_scope_device = __OPENCL_MEMORY_SCOPE_DEVICE,
+  memory_scope_all_svm_devices = __OPENCL_MEMORY_SCOPE_ALL_SVM_DEVICES,
+#if defined(cl_intel_subgroups) || defined(cl_khr_subgroups)
+  memory_scope_sub_group = __OPENCL_MEMORY_SCOPE_SUB_GROUP
+#endif
+} memory_scope;
+
+// GFX90A-HW-REMARK: Hardware instruction generated for atomic fadd operation 
at memory scope workgroup-one-as due to an unsafe request. [-Rpass=si-lower]
+// GFX90A-HW-REMARK: Hardware instruction generated for atomic fadd operation 
at memory scope agent-one-as due to an unsafe request. [-Rpass=si-lower]
+// GFX90A-HW-REMARK: Hardware instruction generated for atomic fadd operation 
at memory scope wavefront-one-as due to an unsafe request. [-Rpass=si-lower]
+// GFX90A-HW-REMARK: global_atomic_add_f32 v0, v[0:1], v2, off glc
+// GFX90A-HW-REMARK: global_atomic_add_f32 v0, v[0:1], v2, off glc
+// GFX90A-HW-REMARK: global_atomic_add_f32 v0, v[0:1], v2, off glc
+// GFX90A-HW-LABEL: @atomic_unsafe_hw
+// GFX90A-HW:   atomicrmw fadd float addrspace(1)* %{{.*}}, float %{{.*}} 
syncscope("workgroup-one-as") monotonic, align 4
+// GFX90A-HW:   atomicrmw fadd float addrspace(1)* %{{.*}}, float %{{.*}} 
syncscope("agent-one-as") monotonic, align 4
+// GFX90A-HW:   atomicrmw fadd float addrspace(1)* %{{.*}}, float %{{.*}} 
syncscope("wavefront-one-as") monotonic, align 4
+void atomic_unsafe_hw(__global atomic_float *d, float a) {
+  float ret1 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_work_group);
+  float ret2 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_device);
+  float ret3 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_sub_group);
+}

diff  --git a/llvm/lib/CodeGen/AtomicExpandPass.cpp 
b/llvm/lib/CodeGen/AtomicExpandPass.cpp
index 47cdd222702f2..1297f99698d8b 100644
--- a/llvm/lib/CodeGen/AtomicExpandPass.cpp
+++ b/llvm/lib/CodeGen/AtomicExpandPass.cpp
@@ -610,7 +610,7 @@ bool AtomicExpand::tryExpandAtomicRMW(AtomicRMWInst *AI) {
   : SSNs[AI->getSyncScopeID()];
   OptimizationRemarkEmitter ORE(AI->getFunction());
   ORE.emit([&]() {
-return OptimizationRemark(DEBUG_TYPE, "Passed", AI->getFunction())
+return OptimizationRemark(DEBUG_TYPE, "Passed", AI)
<< "A compare and swap loop was generated for an a

[clang] f5d5f17 - Revert "[HIP] Allow target addr space in target builtins"

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-08-18T21:38:42-06:00
New Revision: f5d5f17d3ad455de2fbb9448acea66cbc09561c5

URL: 
https://github.com/llvm/llvm-project/commit/f5d5f17d3ad455de2fbb9448acea66cbc09561c5
DIFF: 
https://github.com/llvm/llvm-project/commit/f5d5f17d3ad455de2fbb9448acea66cbc09561c5.diff

LOG: Revert "[HIP] Allow target addr space in target builtins"

This reverts commit a35008955fa606487f79a050f5cc80fc7ee84dda.

Added: 


Modified: 
clang/include/clang/AST/Type.h
clang/lib/Basic/Targets/AMDGPU.h
clang/lib/Sema/SemaExpr.cpp
clang/test/CodeGenCUDA/builtins-amdgcn.cu

Removed: 




diff  --git a/clang/include/clang/AST/Type.h b/clang/include/clang/AST/Type.h
index fc83c895afa2e..09e9705bd86b8 100644
--- a/clang/include/clang/AST/Type.h
+++ b/clang/include/clang/AST/Type.h
@@ -495,12 +495,7 @@ class Qualifiers {
(A == LangAS::Default &&
 (B == LangAS::sycl_private || B == LangAS::sycl_local ||
  B == LangAS::sycl_global || B == LangAS::sycl_global_device ||
- B == LangAS::sycl_global_host)) ||
-   // In HIP device compilation, any cuda address space is allowed
-   // to implicitly cast into the default address space.
-   (A == LangAS::Default &&
-(B == LangAS::cuda_constant || B == LangAS::cuda_device ||
- B == LangAS::cuda_shared));
+ B == LangAS::sycl_global_host));
   }
 
   /// Returns true if the address space in these qualifiers is equal to or

diff  --git a/clang/lib/Basic/Targets/AMDGPU.h 
b/clang/lib/Basic/Targets/AMDGPU.h
index f8772cbe244f0..2e580ecf24259 100644
--- a/clang/lib/Basic/Targets/AMDGPU.h
+++ b/clang/lib/Basic/Targets/AMDGPU.h
@@ -352,16 +352,7 @@ class LLVM_LIBRARY_VISIBILITY AMDGPUTargetInfo final : 
public TargetInfo {
   }
 
   LangAS getCUDABuiltinAddressSpace(unsigned AS) const override {
-switch (AS) {
-case 1:
-  return LangAS::cuda_device;
-case 3:
-  return LangAS::cuda_shared;
-case 4:
-  return LangAS::cuda_constant;
-default:
-  return getLangASFromTargetAS(AS);
-}
+return LangAS::Default;
   }
 
   llvm::Optional getConstantAddressSpace() const override {

diff  --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
index 5bde87d02877e..8ef4a9d96320b 100644
--- a/clang/lib/Sema/SemaExpr.cpp
+++ b/clang/lib/Sema/SemaExpr.cpp
@@ -6572,53 +6572,6 @@ ExprResult Sema::BuildCallExpr(Scope *Scope, Expr *Fn, 
SourceLocation LParenLoc,
   return ExprError();
 
 checkDirectCallValidity(*this, Fn, FD, ArgExprs);
-
-// If this expression is a call to a builtin function in HIP device
-// compilation, allow a pointer-type argument to default address space to 
be
-// passed as a pointer-type parameter to a non-default address space.
-// If Arg is declared in the default address space and Param is declared
-// in a non-default address space, perform an implicit address space cast 
to
-// the parameter type.
-if (getLangOpts().HIP && getLangOpts().CUDAIsDevice && FD &&
-FD->getBuiltinID()) {
-  for (unsigned Idx = 0; Idx < FD->param_size(); ++Idx) {
-ParmVarDecl *Param = FD->getParamDecl(Idx);
-if (!ArgExprs[Idx] || !Param || !Param->getType()->isPointerType() ||
-!ArgExprs[Idx]->getType()->isPointerType())
-  continue;
-
-auto ParamAS = Param->getType()->getPointeeType().getAddressSpace();
-auto ArgTy = ArgExprs[Idx]->getType();
-auto ArgPtTy = ArgTy->getPointeeType();
-auto ArgAS = ArgPtTy.getAddressSpace();
-
-// Only allow implicit casting from a non-default address space pointee
-// type to a default address space pointee type
-if (ArgAS != LangAS::Default || ParamAS == LangAS::Default)
-  continue;
-
-// First, ensure that the Arg is an RValue.
-if (ArgExprs[Idx]->isGLValue()) {
-  ArgExprs[Idx] = ImplicitCastExpr::Create(
-  Context, ArgExprs[Idx]->getType(), CK_NoOp, ArgExprs[Idx],
-  nullptr, VK_PRValue, FPOptionsOverride());
-}
-
-// Construct a new arg type with address space of Param
-Qualifiers ArgPtQuals = ArgPtTy.getQualifiers();
-ArgPtQuals.setAddressSpace(ParamAS);
-auto NewArgPtTy =
-Context.getQualifiedType(ArgPtTy.getUnqualifiedType(), ArgPtQuals);
-auto NewArgTy =
-Context.getQualifiedType(Context.getPointerType(NewArgPtTy),
- ArgTy.getQualifiers());
-
-// Finally perform an implicit address space cast
-ArgExprs[Idx] = ImpCastExprToType(ArgExprs[Idx], NewArgTy,
-  CK_AddressSpaceConversion)
-.get();
-  }
-}
   }
 
   if (Context.isDependenceAllowed() &&

diff  --git a/clang/test/CodeGenCUDA/builtins-amdgcn.cu

[clang] f22ba51 - [Remarks] Emit optimization remarks for atomics generating CAS loop

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-08-16T14:56:01-06:00
New Revision: f22ba51873509b93732015176b778465f40c6db5

URL: 
https://github.com/llvm/llvm-project/commit/f22ba51873509b93732015176b778465f40c6db5
DIFF: 
https://github.com/llvm/llvm-project/commit/f22ba51873509b93732015176b778465f40c6db5.diff

LOG: [Remarks] Emit optimization remarks for atomics generating CAS loop

Implements ORE in AtomicExpand pass to report atomics generating a
compare and swap loop.

Differential Revision: https://reviews.llvm.org/D106891

Added: 
clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
llvm/test/CodeGen/AMDGPU/atomics-remarks-gfx90a.ll

Modified: 
llvm/lib/CodeGen/AtomicExpandPass.cpp

Removed: 




diff  --git a/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu 
b/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
new file mode 100644
index 0..96892286fd75e
--- /dev/null
+++ b/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
@@ -0,0 +1,16 @@
+// RUN: %clang_cc1 %s -triple=amdgcn-amd-amdhsa -fcuda-is-device \
+// RUN:   -target-cpu gfx90a -Rpass=atomic-expand -S -o - 2>&1 | \
+// RUN:   FileCheck %s --check-prefix=GFX90A-CAS
+
+// REQUIRES: amdgpu-registered-target
+
+#include "Inputs/cuda.h"
+#include 
+
+// GFX90A-CAS: A compare and swap loop was generated for an atomic fadd 
operation at system memory scope
+// GFX90A-CAS-LABEL: _Z14atomic_add_casPf
+// GFX90A-CAS:  flat_atomic_cmpswap v0, v[2:3], v[4:5] glc
+// GFX90A-CAS:  s_cbranch_execnz
+__device__ float atomic_add_cas(float *p) {
+  return __atomic_fetch_add(p, 1.0f, memory_order_relaxed);
+}

diff  --git a/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl 
b/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
new file mode 100644
index 0..127866e84e051
--- /dev/null
+++ b/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
@@ -0,0 +1,43 @@
+// RUN: %clang_cc1 %s -cl-std=CL2.0 -O0 -triple=amdgcn-amd-amdhsa -target-cpu 
gfx90a \
+// RUN: -Rpass=atomic-expand -S -o - 2>&1 | \
+// RUN: FileCheck %s --check-prefix=REMARK
+
+// RUN: %clang_cc1 %s -cl-std=CL2.0 -O0 -triple=amdgcn-amd-amdhsa -target-cpu 
gfx90a \
+// RUN: -Rpass=atomic-expand -S -emit-llvm -o - 2>&1 | \
+// RUN: FileCheck %s --check-prefix=GFX90A-CAS
+
+// REQUIRES: amdgpu-registered-target
+
+typedef enum memory_order {
+  memory_order_relaxed = __ATOMIC_RELAXED,
+  memory_order_acquire = __ATOMIC_ACQUIRE,
+  memory_order_release = __ATOMIC_RELEASE,
+  memory_order_acq_rel = __ATOMIC_ACQ_REL,
+  memory_order_seq_cst = __ATOMIC_SEQ_CST
+} memory_order;
+
+typedef enum memory_scope {
+  memory_scope_work_item = __OPENCL_MEMORY_SCOPE_WORK_ITEM,
+  memory_scope_work_group = __OPENCL_MEMORY_SCOPE_WORK_GROUP,
+  memory_scope_device = __OPENCL_MEMORY_SCOPE_DEVICE,
+  memory_scope_all_svm_devices = __OPENCL_MEMORY_SCOPE_ALL_SVM_DEVICES,
+#if defined(cl_intel_subgroups) || defined(cl_khr_subgroups)
+  memory_scope_sub_group = __OPENCL_MEMORY_SCOPE_SUB_GROUP
+#endif
+} memory_scope;
+
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at workgroup-one-as memory scope [-Rpass=atomic-expand]
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at agent-one-as memory scope [-Rpass=atomic-expand]
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at one-as memory scope [-Rpass=atomic-expand]
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at wavefront-one-as memory scope [-Rpass=atomic-expand]
+// GFX90A-CAS-LABEL: @atomic_cas
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("workgroup-one-as") monotonic
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("agent-one-as") monotonic
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} syncscope("one-as") 
monotonic
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("wavefront-one-as") monotonic
+float atomic_cas(__global atomic_float *d, float a) {
+  float ret1 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_work_group);
+  float ret2 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_device);
+  float ret3 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_all_svm_devices);
+  float ret4 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_sub_group);
+}

diff  --git a/llvm/lib/CodeGen/AtomicExpandPass.cpp 
b/llvm/lib/CodeGen/AtomicExpandPass.cpp
index 125a3be585cb5..a27d43e43a855 100644
--- a/llvm/lib/CodeGen/AtomicExpandPass.cpp
+++ b/llvm/lib/CodeGen/AtomicExpandPass.cpp
@@ -17,6 +17,7 @@
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallVector.h"
+#include "llvm/Analysis/OptimizationRemarkEmitter.h"
 #include "llvm/CodeGen/AtomicExpandUtils.h"
 #include "ll

[clang] 4357852 - [Remarks] Emit optimization remarks for atomics generating CAS loop

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-08-14T23:37:23-06:00
New Revision: 435785214f73ff0c92e97f2ade6356e3ba3bf661

URL: 
https://github.com/llvm/llvm-project/commit/435785214f73ff0c92e97f2ade6356e3ba3bf661
DIFF: 
https://github.com/llvm/llvm-project/commit/435785214f73ff0c92e97f2ade6356e3ba3bf661.diff

LOG: [Remarks] Emit optimization remarks for atomics generating CAS loop

Implements ORE in AtomicExpand pass to report atomics generating
a compare and swap loop.

Differential Revision: https://reviews.llvm.org/D106891

Added: 
clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
llvm/test/CodeGen/AMDGPU/atomics-remarks-gfx90a.ll

Modified: 
llvm/lib/CodeGen/AtomicExpandPass.cpp
llvm/test/CodeGen/AArch64/O0-pipeline.ll
llvm/test/CodeGen/AArch64/O3-pipeline.ll
llvm/test/CodeGen/AMDGPU/llc-pipeline.ll
llvm/test/CodeGen/ARM/O3-pipeline.ll
llvm/test/CodeGen/PowerPC/O3-pipeline.ll
llvm/test/CodeGen/X86/O0-pipeline.ll
llvm/test/CodeGen/X86/opt-pipeline.ll

Removed: 




diff  --git a/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu 
b/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
new file mode 100644
index 0..96892286fd75e
--- /dev/null
+++ b/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
@@ -0,0 +1,16 @@
+// RUN: %clang_cc1 %s -triple=amdgcn-amd-amdhsa -fcuda-is-device \
+// RUN:   -target-cpu gfx90a -Rpass=atomic-expand -S -o - 2>&1 | \
+// RUN:   FileCheck %s --check-prefix=GFX90A-CAS
+
+// REQUIRES: amdgpu-registered-target
+
+#include "Inputs/cuda.h"
+#include 
+
+// GFX90A-CAS: A compare and swap loop was generated for an atomic fadd 
operation at system memory scope
+// GFX90A-CAS-LABEL: _Z14atomic_add_casPf
+// GFX90A-CAS:  flat_atomic_cmpswap v0, v[2:3], v[4:5] glc
+// GFX90A-CAS:  s_cbranch_execnz
+__device__ float atomic_add_cas(float *p) {
+  return __atomic_fetch_add(p, 1.0f, memory_order_relaxed);
+}

diff  --git a/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl 
b/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
new file mode 100644
index 0..2d8b68f83b9d6
--- /dev/null
+++ b/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
@@ -0,0 +1,46 @@
+// RUN: %clang_cc1 %s -cl-std=CL2.0 -O0 -triple=amdgcn-amd-amdhsa -target-cpu 
gfx90a \
+// RUN: -Rpass=atomic-expand -S -o - 2>&1 | \
+// RUN: FileCheck %s --check-prefix=REMARK
+
+// RUN: %clang_cc1 %s -cl-std=CL2.0 -O0 -triple=amdgcn-amd-amdhsa -target-cpu 
gfx90a \
+// RUN: -Rpass=atomic-expand -S -emit-llvm -o - 2>&1 | \
+// RUN: FileCheck %s --check-prefix=GFX90A-CAS
+
+// REQUIRES: amdgpu-registered-target
+
+typedef enum memory_order {
+  memory_order_relaxed = __ATOMIC_RELAXED,
+  memory_order_acquire = __ATOMIC_ACQUIRE,
+  memory_order_release = __ATOMIC_RELEASE,
+  memory_order_acq_rel = __ATOMIC_ACQ_REL,
+  memory_order_seq_cst = __ATOMIC_SEQ_CST
+} memory_order;
+
+typedef enum memory_scope {
+  memory_scope_work_item = __OPENCL_MEMORY_SCOPE_WORK_ITEM,
+  memory_scope_work_group = __OPENCL_MEMORY_SCOPE_WORK_GROUP,
+  memory_scope_device = __OPENCL_MEMORY_SCOPE_DEVICE,
+  memory_scope_all_svm_devices = __OPENCL_MEMORY_SCOPE_ALL_SVM_DEVICES,
+#if defined(cl_intel_subgroups) || defined(cl_khr_subgroups)
+  memory_scope_sub_group = __OPENCL_MEMORY_SCOPE_SUB_GROUP
+#endif
+} memory_scope;
+
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at workgroup-one-as memory scope [-Rpass=atomic-expand]
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at agent-one-as memory scope [-Rpass=atomic-expand]
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at one-as memory scope [-Rpass=atomic-expand]
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at wavefront-one-as memory scope [-Rpass=atomic-expand]
+// GFX90A-CAS-LABEL: @atomic_cas
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("workgroup-one-as") monotonic
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("agent-one-as") monotonic
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} syncscope("one-as") 
monotonic
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("wavefront-one-as") monotonic
+float atomic_cas(__global atomic_float *d, float a) {
+  float ret1 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_work_group);
+  float ret2 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_device);
+  float ret3 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_all_svm_devices);
+  float ret4 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_sub_group);
+}
+
+
+

diff  --git a/llvm/lib/CodeGen/AtomicExpandPass.cpp 
b/llvm/lib/CodeGen/AtomicExpandPass.cpp
index 125a3be585cb5..5b5458e1058e8 100644
--- a/ll

[clang] 29e11a1 - Revert "[Remarks] Emit optimization remarks for atomics generating CAS loop"

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-08-13T23:58:04-06:00
New Revision: 29e11a1aa303cf81b81fdbab74fad4f31e5018d3

URL: 
https://github.com/llvm/llvm-project/commit/29e11a1aa303cf81b81fdbab74fad4f31e5018d3
DIFF: 
https://github.com/llvm/llvm-project/commit/29e11a1aa303cf81b81fdbab74fad4f31e5018d3.diff

LOG: Revert "[Remarks] Emit optimization remarks for atomics generating CAS 
loop"

This reverts commit c4e5425aa579d21530ef1766d7144b38a347f247.

Added: 


Modified: 
llvm/lib/CodeGen/AtomicExpandPass.cpp
llvm/test/CodeGen/AMDGPU/llc-pipeline.ll
llvm/test/CodeGen/X86/O0-pipeline.ll
llvm/test/CodeGen/X86/opt-pipeline.ll

Removed: 
clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
llvm/test/CodeGen/AMDGPU/atomics-remarks-gfx90a.ll



diff  --git a/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu 
b/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
deleted file mode 100644
index 96892286fd75e..0
--- a/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
+++ /dev/null
@@ -1,16 +0,0 @@
-// RUN: %clang_cc1 %s -triple=amdgcn-amd-amdhsa -fcuda-is-device \
-// RUN:   -target-cpu gfx90a -Rpass=atomic-expand -S -o - 2>&1 | \
-// RUN:   FileCheck %s --check-prefix=GFX90A-CAS
-
-// REQUIRES: amdgpu-registered-target
-
-#include "Inputs/cuda.h"
-#include 
-
-// GFX90A-CAS: A compare and swap loop was generated for an atomic fadd 
operation at system memory scope
-// GFX90A-CAS-LABEL: _Z14atomic_add_casPf
-// GFX90A-CAS:  flat_atomic_cmpswap v0, v[2:3], v[4:5] glc
-// GFX90A-CAS:  s_cbranch_execnz
-__device__ float atomic_add_cas(float *p) {
-  return __atomic_fetch_add(p, 1.0f, memory_order_relaxed);
-}

diff  --git a/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl 
b/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
deleted file mode 100644
index 2d8b68f83b9d6..0
--- a/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
+++ /dev/null
@@ -1,46 +0,0 @@
-// RUN: %clang_cc1 %s -cl-std=CL2.0 -O0 -triple=amdgcn-amd-amdhsa -target-cpu 
gfx90a \
-// RUN: -Rpass=atomic-expand -S -o - 2>&1 | \
-// RUN: FileCheck %s --check-prefix=REMARK
-
-// RUN: %clang_cc1 %s -cl-std=CL2.0 -O0 -triple=amdgcn-amd-amdhsa -target-cpu 
gfx90a \
-// RUN: -Rpass=atomic-expand -S -emit-llvm -o - 2>&1 | \
-// RUN: FileCheck %s --check-prefix=GFX90A-CAS
-
-// REQUIRES: amdgpu-registered-target
-
-typedef enum memory_order {
-  memory_order_relaxed = __ATOMIC_RELAXED,
-  memory_order_acquire = __ATOMIC_ACQUIRE,
-  memory_order_release = __ATOMIC_RELEASE,
-  memory_order_acq_rel = __ATOMIC_ACQ_REL,
-  memory_order_seq_cst = __ATOMIC_SEQ_CST
-} memory_order;
-
-typedef enum memory_scope {
-  memory_scope_work_item = __OPENCL_MEMORY_SCOPE_WORK_ITEM,
-  memory_scope_work_group = __OPENCL_MEMORY_SCOPE_WORK_GROUP,
-  memory_scope_device = __OPENCL_MEMORY_SCOPE_DEVICE,
-  memory_scope_all_svm_devices = __OPENCL_MEMORY_SCOPE_ALL_SVM_DEVICES,
-#if defined(cl_intel_subgroups) || defined(cl_khr_subgroups)
-  memory_scope_sub_group = __OPENCL_MEMORY_SCOPE_SUB_GROUP
-#endif
-} memory_scope;
-
-// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at workgroup-one-as memory scope [-Rpass=atomic-expand]
-// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at agent-one-as memory scope [-Rpass=atomic-expand]
-// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at one-as memory scope [-Rpass=atomic-expand]
-// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at wavefront-one-as memory scope [-Rpass=atomic-expand]
-// GFX90A-CAS-LABEL: @atomic_cas
-// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("workgroup-one-as") monotonic
-// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("agent-one-as") monotonic
-// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} syncscope("one-as") 
monotonic
-// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("wavefront-one-as") monotonic
-float atomic_cas(__global atomic_float *d, float a) {
-  float ret1 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_work_group);
-  float ret2 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_device);
-  float ret3 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_all_svm_devices);
-  float ret4 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_sub_group);
-}
-
-
-

diff  --git a/llvm/lib/CodeGen/AtomicExpandPass.cpp 
b/llvm/lib/CodeGen/AtomicExpandPass.cpp
index 5b5458e1058e8..125a3be585cb5 100644
--- a/llvm/lib/CodeGen/AtomicExpandPass.cpp
+++ b/llvm/lib/CodeGen/AtomicExpandPass.cpp
@@ -17,7 +17,6 @@
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallVector.h"
-#include "llvm/Analysis/OptimizationRemarkEm

[clang] c4e5425 - [Remarks] Emit optimization remarks for atomics generating CAS loop

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-08-13T22:44:08-06:00
New Revision: c4e5425aa579d21530ef1766d7144b38a347f247

URL: 
https://github.com/llvm/llvm-project/commit/c4e5425aa579d21530ef1766d7144b38a347f247
DIFF: 
https://github.com/llvm/llvm-project/commit/c4e5425aa579d21530ef1766d7144b38a347f247.diff

LOG: [Remarks] Emit optimization remarks for atomics generating CAS loop

Implements ORE in AtomicExpandPass to report atomics generating a compare
and swap loop.

Differential Revision: https://reviews.llvm.org/D106891

Added: 
clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
llvm/test/CodeGen/AMDGPU/atomics-remarks-gfx90a.ll

Modified: 
llvm/lib/CodeGen/AtomicExpandPass.cpp
llvm/test/CodeGen/AMDGPU/llc-pipeline.ll
llvm/test/CodeGen/X86/O0-pipeline.ll
llvm/test/CodeGen/X86/opt-pipeline.ll

Removed: 




diff  --git a/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu 
b/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
new file mode 100644
index 0..96892286fd75e
--- /dev/null
+++ b/clang/test/CodeGenCUDA/atomics-remarks-gfx90a.cu
@@ -0,0 +1,16 @@
+// RUN: %clang_cc1 %s -triple=amdgcn-amd-amdhsa -fcuda-is-device \
+// RUN:   -target-cpu gfx90a -Rpass=atomic-expand -S -o - 2>&1 | \
+// RUN:   FileCheck %s --check-prefix=GFX90A-CAS
+
+// REQUIRES: amdgpu-registered-target
+
+#include "Inputs/cuda.h"
+#include 
+
+// GFX90A-CAS: A compare and swap loop was generated for an atomic fadd 
operation at system memory scope
+// GFX90A-CAS-LABEL: _Z14atomic_add_casPf
+// GFX90A-CAS:  flat_atomic_cmpswap v0, v[2:3], v[4:5] glc
+// GFX90A-CAS:  s_cbranch_execnz
+__device__ float atomic_add_cas(float *p) {
+  return __atomic_fetch_add(p, 1.0f, memory_order_relaxed);
+}

diff  --git a/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl 
b/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
new file mode 100644
index 0..2d8b68f83b9d6
--- /dev/null
+++ b/clang/test/CodeGenOpenCL/atomics-remarks-gfx90a.cl
@@ -0,0 +1,46 @@
+// RUN: %clang_cc1 %s -cl-std=CL2.0 -O0 -triple=amdgcn-amd-amdhsa -target-cpu 
gfx90a \
+// RUN: -Rpass=atomic-expand -S -o - 2>&1 | \
+// RUN: FileCheck %s --check-prefix=REMARK
+
+// RUN: %clang_cc1 %s -cl-std=CL2.0 -O0 -triple=amdgcn-amd-amdhsa -target-cpu 
gfx90a \
+// RUN: -Rpass=atomic-expand -S -emit-llvm -o - 2>&1 | \
+// RUN: FileCheck %s --check-prefix=GFX90A-CAS
+
+// REQUIRES: amdgpu-registered-target
+
+typedef enum memory_order {
+  memory_order_relaxed = __ATOMIC_RELAXED,
+  memory_order_acquire = __ATOMIC_ACQUIRE,
+  memory_order_release = __ATOMIC_RELEASE,
+  memory_order_acq_rel = __ATOMIC_ACQ_REL,
+  memory_order_seq_cst = __ATOMIC_SEQ_CST
+} memory_order;
+
+typedef enum memory_scope {
+  memory_scope_work_item = __OPENCL_MEMORY_SCOPE_WORK_ITEM,
+  memory_scope_work_group = __OPENCL_MEMORY_SCOPE_WORK_GROUP,
+  memory_scope_device = __OPENCL_MEMORY_SCOPE_DEVICE,
+  memory_scope_all_svm_devices = __OPENCL_MEMORY_SCOPE_ALL_SVM_DEVICES,
+#if defined(cl_intel_subgroups) || defined(cl_khr_subgroups)
+  memory_scope_sub_group = __OPENCL_MEMORY_SCOPE_SUB_GROUP
+#endif
+} memory_scope;
+
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at workgroup-one-as memory scope [-Rpass=atomic-expand]
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at agent-one-as memory scope [-Rpass=atomic-expand]
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at one-as memory scope [-Rpass=atomic-expand]
+// REMARK: remark: A compare and swap loop was generated for an atomic fadd 
operation at wavefront-one-as memory scope [-Rpass=atomic-expand]
+// GFX90A-CAS-LABEL: @atomic_cas
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("workgroup-one-as") monotonic
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("agent-one-as") monotonic
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} syncscope("one-as") 
monotonic
+// GFX90A-CAS: atomicrmw fadd float addrspace(1)* {{.*}} 
syncscope("wavefront-one-as") monotonic
+float atomic_cas(__global atomic_float *d, float a) {
+  float ret1 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_work_group);
+  float ret2 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_device);
+  float ret3 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_all_svm_devices);
+  float ret4 = __opencl_atomic_fetch_add(d, a, memory_order_relaxed, 
memory_scope_sub_group);
+}
+
+
+

diff  --git a/llvm/lib/CodeGen/AtomicExpandPass.cpp 
b/llvm/lib/CodeGen/AtomicExpandPass.cpp
index 125a3be585cb5..5b5458e1058e8 100644
--- a/llvm/lib/CodeGen/AtomicExpandPass.cpp
+++ b/llvm/lib/CodeGen/AtomicExpandPass.cpp
@@ -17,6 +17,7 @@
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llv

[clang] a350089 - [HIP] Allow target addr space in target builtins

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-08-09T16:38:04-06:00
New Revision: a35008955fa606487f79a050f5cc80fc7ee84dda

URL: 
https://github.com/llvm/llvm-project/commit/a35008955fa606487f79a050f5cc80fc7ee84dda
DIFF: 
https://github.com/llvm/llvm-project/commit/a35008955fa606487f79a050f5cc80fc7ee84dda.diff

LOG: [HIP] Allow target addr space in target builtins

This patch allows target specific addr space in target builtins for HIP. It 
inserts implicit addr
space cast for non-generic pointer to generic pointer in general, and inserts 
implicit addr
space cast for generic to non-generic for target builtin arguments only.

It is NFC for non-HIP languages.

Differential Revision: https://reviews.llvm.org/D102405

Added: 


Modified: 
clang/include/clang/AST/Type.h
clang/lib/Basic/Targets/AMDGPU.h
clang/lib/Sema/SemaExpr.cpp
clang/test/CodeGenCUDA/builtins-amdgcn.cu

Removed: 




diff  --git a/clang/include/clang/AST/Type.h b/clang/include/clang/AST/Type.h
index 9f46d53378976..4238667b8b076 100644
--- a/clang/include/clang/AST/Type.h
+++ b/clang/include/clang/AST/Type.h
@@ -495,7 +495,12 @@ class Qualifiers {
(A == LangAS::Default &&
 (B == LangAS::sycl_private || B == LangAS::sycl_local ||
  B == LangAS::sycl_global || B == LangAS::sycl_global_device ||
- B == LangAS::sycl_global_host));
+ B == LangAS::sycl_global_host)) ||
+   // In HIP device compilation, any cuda address space is allowed
+   // to implicitly cast into the default address space.
+   (A == LangAS::Default &&
+(B == LangAS::cuda_constant || B == LangAS::cuda_device ||
+ B == LangAS::cuda_shared));
   }
 
   /// Returns true if the address space in these qualifiers is equal to or

diff  --git a/clang/lib/Basic/Targets/AMDGPU.h 
b/clang/lib/Basic/Targets/AMDGPU.h
index 2e580ecf24259..f8772cbe244f0 100644
--- a/clang/lib/Basic/Targets/AMDGPU.h
+++ b/clang/lib/Basic/Targets/AMDGPU.h
@@ -352,7 +352,16 @@ class LLVM_LIBRARY_VISIBILITY AMDGPUTargetInfo final : 
public TargetInfo {
   }
 
   LangAS getCUDABuiltinAddressSpace(unsigned AS) const override {
-return LangAS::Default;
+switch (AS) {
+case 1:
+  return LangAS::cuda_device;
+case 3:
+  return LangAS::cuda_shared;
+case 4:
+  return LangAS::cuda_constant;
+default:
+  return getLangASFromTargetAS(AS);
+}
   }
 
   llvm::Optional getConstantAddressSpace() const override {

diff  --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
index 8ef4a9d96320b..5bde87d02877e 100644
--- a/clang/lib/Sema/SemaExpr.cpp
+++ b/clang/lib/Sema/SemaExpr.cpp
@@ -6572,6 +6572,53 @@ ExprResult Sema::BuildCallExpr(Scope *Scope, Expr *Fn, 
SourceLocation LParenLoc,
   return ExprError();
 
 checkDirectCallValidity(*this, Fn, FD, ArgExprs);
+
+// If this expression is a call to a builtin function in HIP device
+// compilation, allow a pointer-type argument to default address space to 
be
+// passed as a pointer-type parameter to a non-default address space.
+// If Arg is declared in the default address space and Param is declared
+// in a non-default address space, perform an implicit address space cast 
to
+// the parameter type.
+if (getLangOpts().HIP && getLangOpts().CUDAIsDevice && FD &&
+FD->getBuiltinID()) {
+  for (unsigned Idx = 0; Idx < FD->param_size(); ++Idx) {
+ParmVarDecl *Param = FD->getParamDecl(Idx);
+if (!ArgExprs[Idx] || !Param || !Param->getType()->isPointerType() ||
+!ArgExprs[Idx]->getType()->isPointerType())
+  continue;
+
+auto ParamAS = Param->getType()->getPointeeType().getAddressSpace();
+auto ArgTy = ArgExprs[Idx]->getType();
+auto ArgPtTy = ArgTy->getPointeeType();
+auto ArgAS = ArgPtTy.getAddressSpace();
+
+// Only allow implicit casting from a non-default address space pointee
+// type to a default address space pointee type
+if (ArgAS != LangAS::Default || ParamAS == LangAS::Default)
+  continue;
+
+// First, ensure that the Arg is an RValue.
+if (ArgExprs[Idx]->isGLValue()) {
+  ArgExprs[Idx] = ImplicitCastExpr::Create(
+  Context, ArgExprs[Idx]->getType(), CK_NoOp, ArgExprs[Idx],
+  nullptr, VK_PRValue, FPOptionsOverride());
+}
+
+// Construct a new arg type with address space of Param
+Qualifiers ArgPtQuals = ArgPtTy.getQualifiers();
+ArgPtQuals.setAddressSpace(ParamAS);
+auto NewArgPtTy =
+Context.getQualifiedType(ArgPtTy.getUnqualifiedType(), ArgPtQuals);
+auto NewArgTy =
+Context.getQualifiedType(Context.getPointerType(NewArgPtTy),
+ ArgTy.getQualifiers());
+
+// Finally perform an implicit address space cast
+

[clang] 39dac1f - [clang] Add clang builtins support for gfx90a

[Anshil Gandhi via cfe-commits]

Author: Anshil Gandhi
Date: 2021-08-05T02:08:06-06:00
New Revision: 39dac1f7f65691487dbdc969e343108db5b0f765

URL: 
https://github.com/llvm/llvm-project/commit/39dac1f7f65691487dbdc969e343108db5b0f765
DIFF: 
https://github.com/llvm/llvm-project/commit/39dac1f7f65691487dbdc969e343108db5b0f765.diff

LOG: [clang] Add clang builtins support for gfx90a

Implement target builtins for gfx90a including fadd64, fadd32, add2h,
max and min on various global, flat and ds address spaces for which
intrinsics are implemented.

Differential Revision: https://reviews.llvm.org/D106909

Added: 
clang/test/CodeGenOpenCL/builtins-amdgcn-fp-atomics-gfx7-err.cl
clang/test/CodeGenOpenCL/builtins-amdgcn-fp-atomics-gfx908-err.cl
clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx1030.cl
clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx8.cl
clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx90a.cl

Modified: 
clang/include/clang/Basic/BuiltinsAMDGPU.def
clang/lib/CodeGen/CGBuiltin.cpp

Removed: 




diff  --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def 
b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 3570431d952cb..2e1d3c7ccbff9 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -196,6 +196,19 @@ TARGET_BUILTIN(__builtin_amdgcn_perm, "UiUiUiUi", "nc", 
"gfx8-insts")
 
 TARGET_BUILTIN(__builtin_amdgcn_fmed3h, "", "nc", "gfx9-insts")
 
+TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_f64, "dd*1d", "t", 
"gfx90a-insts")
+TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_f32, "ff*1f", "t", 
"gfx90a-insts")
+TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_v2f16, "V2hV2h*1V2h", "t", 
"gfx90a-insts")
+TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fmin_f64, "dd*1d", "t", 
"gfx90a-insts")
+TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fmax_f64, "dd*1d", "t", 
"gfx90a-insts")
+
+TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fadd_f64, "dd*0d", "t", 
"gfx90a-insts")
+TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fmin_f64, "dd*0d", "t", 
"gfx90a-insts")
+TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fmax_f64, "dd*0d", "t", 
"gfx90a-insts")
+
+TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_f64, "dd*3d", "t", 
"gfx90a-insts")
+TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_f32, "ff*3f", "t", "gfx8-insts")
+
 
//===--===//
 // Deep learning builtins.
 
//===--===//

diff  --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index b316a865f2fc7..606689385199a 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -16197,6 +16197,74 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned 
BuiltinID,
 Src0 = Builder.CreatePointerBitCastOrAddrSpaceCast(Src0, PTy);
 return Builder.CreateCall(F, { Src0, Src1, Src2, Src3, Src4 });
   }
+  case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f64:
+  case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f32:
+  case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2f16:
+  case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
+  case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
+  case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f64:
+  case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
+  case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64: {
+Intrinsic::ID IID;
+llvm::Type *ArgTy = llvm::Type::getDoubleTy(getLLVMContext());
+switch (BuiltinID) {
+case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f32:
+  ArgTy = llvm::Type::getFloatTy(getLLVMContext());
+  IID = Intrinsic::amdgcn_global_atomic_fadd;
+  break;
+case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2f16:
+  ArgTy = llvm::FixedVectorType::get(
+  llvm::Type::getHalfTy(getLLVMContext()), 2);
+  IID = Intrinsic::amdgcn_global_atomic_fadd;
+  break;
+case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f64:
+  IID = Intrinsic::amdgcn_global_atomic_fadd;
+  break;
+case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
+  IID = Intrinsic::amdgcn_global_atomic_fmin;
+  break;
+case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
+  IID = Intrinsic::amdgcn_global_atomic_fmax;
+  break;
+case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f64:
+  IID = Intrinsic::amdgcn_flat_atomic_fadd;
+  break;
+case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
+  IID = Intrinsic::amdgcn_flat_atomic_fmin;
+  break;
+case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64:
+  IID = Intrinsic::amdgcn_flat_atomic_fmax;
+  break;
+}
+llvm::Value *Addr = EmitScalarExpr(E->getArg(0));
+llvm::Value *Val = EmitScalarExpr(E->getArg(1));
+llvm::Function *F =
+CGM.getIntrinsic(IID, {ArgTy, Addr->getType(), Val->getType()});
+return Builder.CreateCa

Getting started

[Anshil Gandhi via cfe-commits]

Hi everyone,

My name is Anshil Gandhi and I am currently in my third year of BSc double
majoring in Computing Science and Mathematics. I am interested in
developing the clang frontend, C++ 1x features implementation in
particular. I have cloned the git repository of llvm and explored through
various features of clang, however I am not sure how to familarize myself
with the project organization. I will appreciate any pointers on how to get
started. Thanks in advance!

Kind regards,

Anshil Gandhi
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits