[clang] [clang][HLSL] Add radians intrinsic (PR #110802)
https://github.com/adam-yang created
https://github.com/llvm/llvm-project/pull/110802
partially fixes #99151
### Changes
* Implemented `radians` clang builtin
* Linked `radians` clang builtin with `hlsl_intrinsics.h`
* Added sema checks for `radians` to `CheckHLSLBuiltinFunctionCall` in
`SemaChecking.cpp`
* Add codegen for `radians` to `EmitHLSLBuiltinExpr` in `CGBuiltin.cpp`
* Add codegen tests to `clang/test/CodeGenHLSL/builtins/radians.hlsl`
* Add sema tests to `clang/test/SemaHLSL/BuiltIns/radians-errors.hlsl`
### Related PRs
* [[DXIL] Add radians intrinsic
#110616](https://github.com/llvm/llvm-project/pull/110616)
* [[SPIRV] Add radians intrinsic
#110800](https://github.com/llvm/llvm-project/pull/110800)
>From be88a591ed220714b21b73e34406bfcdb866ac17 Mon Sep 17 00:00:00 2001
From: Adam Yang
Date: Wed, 2 Oct 2024 01:21:19 -0700
Subject: [PATCH] Added radians to clang
---
clang/include/clang/Basic/Builtins.td | 6 +++
clang/lib/CodeGen/CGBuiltin.cpp | 8 +++
clang/lib/CodeGen/CGHLSLRuntime.h | 1 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 30 +++
clang/lib/Sema/SemaHLSL.cpp | 1 +
clang/test/CodeGenHLSL/builtins/radians.hlsl | 54 +++
.../SemaHLSL/BuiltIns/radians-errors.hlsl | 27 ++
7 files changed, 127 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/radians.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/radians-errors.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 8090119e512fbb..09b130e0def0d7 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4788,6 +4788,12 @@ def HLSLStep: LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLRadians : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_elementwise_radians"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index d739597de4c855..6c24c66a81e73c 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18852,6 +18852,14 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
retType, CGM.getHLSLRuntime().getSignIntrinsic(),
ArrayRef{Op0}, nullptr, "hlsl.sign");
}
+ case Builtin::BI__builtin_hlsl_elementwise_radians: {
+Value *Op0 = EmitScalarExpr(E->getArg(0));
+if (!E->getArg(0)->getType()->hasFloatingRepresentation())
+ llvm_unreachable("radians operand must have a float representation");
+return Builder.CreateIntrinsic(
+/*ReturnType=*/Op0->getType(),
CGM.getHLSLRuntime().getRadiansIntrinsic(),
+ArrayRef{Op0}, nullptr, "hlsl.radians");
+ }
}
return nullptr;
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index a8aabca7348ffb..c807a3216ec9ee 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -82,6 +82,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(Saturate, saturate)
GENERATE_HLSL_INTRINSIC_FUNCTION(Sign, sign)
GENERATE_HLSL_INTRINSIC_FUNCTION(Step, step)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(Radians, radians)
GENERATE_HLSL_INTRINSIC_FUNCTION(ThreadId, thread_id)
GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 810a16d75f0228..81b297723da3f5 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2086,5 +2086,35 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int3 sign(double3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int4 sign(double4);
+
+//===--===//
+// radians builtins
+//===--===//
+
+/// \fn T radians(T Val)
+/// \brief Converts the specified value from degrees to radians.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half radians(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half2 radians(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half3 radians(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half4 radians(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float radians(float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float2 radians(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float3 radians(float3);
+_HLSL_BUILTIN_ALIAS(__builtin
[clang] [clang][HLSL] Add radians intrinsic (PR #111883)
https://github.com/adam-yang created
https://github.com/llvm/llvm-project/pull/111883
partially fixes #70103
### Changes
* Implemented `GroupMemoryBarrierWithGroupSync` clang builtin
* Linked `GroupMemoryBarrierWithGroupSync` clang builtin with
`hlsl_intrinsics.h`
* Added sema checks for `GroupMemoryBarrierWithGroupSync` to
`CheckHLSLBuiltinFunctionCall` in
`SemaChecking.cpp`
* Add codegen for `GroupMemoryBarrierWithGroupSync` to `EmitHLSLBuiltinExpr` in
`CGBuiltin.cpp`
* Add codegen tests to
`clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync.hlsl`
* Add sema tests to
`clang/test/SemaHLSL/BuiltIns/GroupMemoryBarrierWithGroupSync-errors.hlsl`
### Related PRs
>From 0f97b92617dcadd78362f24f3ef0719c19f6f3f6 Mon Sep 17 00:00:00 2001
From: Adam Yang
Date: Thu, 10 Oct 2024 10:20:37 -0700
Subject: [PATCH 1/2] Added the intrinsic to clang
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 4
clang/lib/CodeGen/CGHLSLRuntime.h | 1 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 12 ++
clang/lib/Sema/SemaHLSL.cpp | 5 +
...roupMemoryBarrierWithGroupSync-errors.hlsl | 6 +
.../GroupMemoryBarrierWithGroupSync.hlsl | 22 +++
7 files changed, 56 insertions(+)
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 9ebee81fcb0d3d..96fea17badfe9e 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4830,6 +4830,12 @@ def HLSLRadians : LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLGroupMemoryBarrierWithGroupSync: LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_group_memory_barrier_with_group_sync"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 57705f2d2d0423..45890019ee2a64 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18901,6 +18901,10 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
CGM.getHLSLRuntime().getRadiansIntrinsic(), ArrayRef{Op0},
nullptr, "hlsl.radians");
}
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+Intrinsic::ID ID =
CGM.getHLSLRuntime().getGroupMemoryBarrierWithGroupSyncIntrinsic();
+return EmitRuntimeCall(Intrinsic::getDeclaration(&CGM.getModule(), ID));
+ }
}
return nullptr;
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 05ff325216f55b..1e095cedf58731 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -89,6 +89,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
GENERATE_HLSL_INTRINSIC_FUNCTION(WaveIsFirstLane, wave_is_first_lane)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrierWithGroupSync,
groupMemoryBarrierWithGroupSync)
//===--===//
// End of reserved area for HLSL intrinsic getters.
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index f39a68ba847e98..637df8924e2468 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2168,5 +2168,17 @@ float3 radians(float3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
float4 radians(float4);
+//===--===//
+// GroupMemoryBarrierWithGroupSync builtins
+//===--===//
+
+/// \fn void GroupMemoryBarrierWithGroupSync(void)
+/// \brief Blocks execution of all threads in a group until all group shared
+/// accesses have been completed and all threads in the group have reached this
+/// call.
+
+_HLSL_BUILTIN_ALIAS(__builtin_group_memory_barrier_with_group_sync)
+void GroupMemoryBarrierWithGroupSync(void);
+
} // namespace hlsl
#endif //_HLSL_HLSL_INTRINSICS_H_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index d2d2df829e7b11..30480f8f57c12d 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2018,6 +2018,11 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+if (SemaRef.checkArgCountAtMost(TheCall, 0))
+ return true;
+break;
+ }
}
return false;
}
diff --git
a/clang/test/CodeGenHLSL/buil
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang edited https://github.com/llvm/llvm-project/pull/111883 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang ready_for_review https://github.com/llvm/llvm-project/pull/111883 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/111883
>From 0f97b92617dcadd78362f24f3ef0719c19f6f3f6 Mon Sep 17 00:00:00 2001
From: Adam Yang
Date: Thu, 10 Oct 2024 10:20:37 -0700
Subject: [PATCH 1/3] Added the intrinsic to clang
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 4
clang/lib/CodeGen/CGHLSLRuntime.h | 1 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 12 ++
clang/lib/Sema/SemaHLSL.cpp | 5 +
...roupMemoryBarrierWithGroupSync-errors.hlsl | 6 +
.../GroupMemoryBarrierWithGroupSync.hlsl | 22 +++
7 files changed, 56 insertions(+)
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 9ebee81fcb0d3d..96fea17badfe9e 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4830,6 +4830,12 @@ def HLSLRadians : LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLGroupMemoryBarrierWithGroupSync: LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_group_memory_barrier_with_group_sync"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 57705f2d2d0423..45890019ee2a64 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18901,6 +18901,10 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
CGM.getHLSLRuntime().getRadiansIntrinsic(), ArrayRef{Op0},
nullptr, "hlsl.radians");
}
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+Intrinsic::ID ID =
CGM.getHLSLRuntime().getGroupMemoryBarrierWithGroupSyncIntrinsic();
+return EmitRuntimeCall(Intrinsic::getDeclaration(&CGM.getModule(), ID));
+ }
}
return nullptr;
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 05ff325216f55b..1e095cedf58731 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -89,6 +89,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
GENERATE_HLSL_INTRINSIC_FUNCTION(WaveIsFirstLane, wave_is_first_lane)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrierWithGroupSync,
groupMemoryBarrierWithGroupSync)
//===--===//
// End of reserved area for HLSL intrinsic getters.
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index f39a68ba847e98..637df8924e2468 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2168,5 +2168,17 @@ float3 radians(float3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
float4 radians(float4);
+//===--===//
+// GroupMemoryBarrierWithGroupSync builtins
+//===--===//
+
+/// \fn void GroupMemoryBarrierWithGroupSync(void)
+/// \brief Blocks execution of all threads in a group until all group shared
+/// accesses have been completed and all threads in the group have reached this
+/// call.
+
+_HLSL_BUILTIN_ALIAS(__builtin_group_memory_barrier_with_group_sync)
+void GroupMemoryBarrierWithGroupSync(void);
+
} // namespace hlsl
#endif //_HLSL_HLSL_INTRINSICS_H_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index d2d2df829e7b11..30480f8f57c12d 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2018,6 +2018,11 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+if (SemaRef.checkArgCountAtMost(TheCall, 0))
+ return true;
+break;
+ }
}
return false;
}
diff --git
a/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
new file mode 100644
index 00..9ffc1ebce1c909
--- /dev/null
+++
b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
@@ -0,0 +1,6 @@
+// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only
-disable-llvm-passes -verify -verify-ignore-unexpected
+
+void test_too_many_arg() {
+ __builtin_group_memory_barrier_with_group_sync(0);
+ // expected-error@-1 {{too many argume
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang edited https://github.com/llvm/llvm-project/pull/111883 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang edited https://github.com/llvm/llvm-project/pull/111883 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang edited https://github.com/llvm/llvm-project/pull/111883 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang edited https://github.com/llvm/llvm-project/pull/111883 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add radians intrinsic (PR #110802)
https://github.com/adam-yang ready_for_review https://github.com/llvm/llvm-project/pull/110802 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add radians intrinsic (PR #110802)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/110802
>From be88a591ed220714b21b73e34406bfcdb866ac17 Mon Sep 17 00:00:00 2001
From: Adam Yang
Date: Wed, 2 Oct 2024 01:21:19 -0700
Subject: [PATCH 1/2] Added radians to clang
---
clang/include/clang/Basic/Builtins.td | 6 +++
clang/lib/CodeGen/CGBuiltin.cpp | 8 +++
clang/lib/CodeGen/CGHLSLRuntime.h | 1 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 30 +++
clang/lib/Sema/SemaHLSL.cpp | 1 +
clang/test/CodeGenHLSL/builtins/radians.hlsl | 54 +++
.../SemaHLSL/BuiltIns/radians-errors.hlsl | 27 ++
7 files changed, 127 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/radians.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/radians-errors.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 8090119e512fbb..09b130e0def0d7 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4788,6 +4788,12 @@ def HLSLStep: LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLRadians : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_elementwise_radians"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index d739597de4c855..6c24c66a81e73c 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18852,6 +18852,14 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
retType, CGM.getHLSLRuntime().getSignIntrinsic(),
ArrayRef{Op0}, nullptr, "hlsl.sign");
}
+ case Builtin::BI__builtin_hlsl_elementwise_radians: {
+Value *Op0 = EmitScalarExpr(E->getArg(0));
+if (!E->getArg(0)->getType()->hasFloatingRepresentation())
+ llvm_unreachable("radians operand must have a float representation");
+return Builder.CreateIntrinsic(
+/*ReturnType=*/Op0->getType(),
CGM.getHLSLRuntime().getRadiansIntrinsic(),
+ArrayRef{Op0}, nullptr, "hlsl.radians");
+ }
}
return nullptr;
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index a8aabca7348ffb..c807a3216ec9ee 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -82,6 +82,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(Saturate, saturate)
GENERATE_HLSL_INTRINSIC_FUNCTION(Sign, sign)
GENERATE_HLSL_INTRINSIC_FUNCTION(Step, step)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(Radians, radians)
GENERATE_HLSL_INTRINSIC_FUNCTION(ThreadId, thread_id)
GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 810a16d75f0228..81b297723da3f5 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2086,5 +2086,35 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int3 sign(double3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int4 sign(double4);
+
+//===--===//
+// radians builtins
+//===--===//
+
+/// \fn T radians(T Val)
+/// \brief Converts the specified value from degrees to radians.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half radians(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half2 radians(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half3 radians(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half4 radians(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float radians(float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float2 radians(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float3 radians(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float4 radians(float4);
+
} // namespace hlsl
#endif //_HLSL_HLSL_INTRINSICS_H_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 43cc6c81ae5cb0..c0326cf280cf3f 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1861,6 +1861,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_hlsl_elementwise_radians:
case Builtin::BI__builtin_hlsl_elementwise_rsqrt:
case Builtin::BI__builtin_hlsl_elementwise_frac: {
if (CheckFloatOrHalfRepresentations
[clang] [clang][HLSL] Add radians intrinsic (PR #110802)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/110802
>From 30c96a73cc07aa8203f12157475b81eb947951fe Mon Sep 17 00:00:00 2001
From: Adam Yang
Date: Wed, 2 Oct 2024 01:21:19 -0700
Subject: [PATCH 1/4] Added radians to clang
---
clang/include/clang/Basic/Builtins.td | 6 +++
clang/lib/CodeGen/CGBuiltin.cpp | 8 +++
clang/lib/CodeGen/CGHLSLRuntime.h | 1 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 30 +++
clang/lib/Sema/SemaHLSL.cpp | 1 +
clang/test/CodeGenHLSL/builtins/radians.hlsl | 54 +++
.../SemaHLSL/BuiltIns/radians-errors.hlsl | 27 ++
7 files changed, 127 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/radians.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/radians-errors.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index a726a0ef4b4bd2..9ebee81fcb0d3d 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4824,6 +4824,12 @@ def HLSLStep: LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLRadians : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_elementwise_radians"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 43700ea9dd3cfd..ebb9dcfd933aa6 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18892,6 +18892,14 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
retType, CGM.getHLSLRuntime().getSignIntrinsic(),
ArrayRef{Op0}, nullptr, "hlsl.sign");
}
+ case Builtin::BI__builtin_hlsl_elementwise_radians: {
+Value *Op0 = EmitScalarExpr(E->getArg(0));
+if (!E->getArg(0)->getType()->hasFloatingRepresentation())
+ llvm_unreachable("radians operand must have a float representation");
+return Builder.CreateIntrinsic(
+/*ReturnType=*/Op0->getType(),
CGM.getHLSLRuntime().getRadiansIntrinsic(),
+ArrayRef{Op0}, nullptr, "hlsl.radians");
+ }
}
return nullptr;
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 6722d2c7c50a2b..05ff325216f55b 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -83,6 +83,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(Saturate, saturate)
GENERATE_HLSL_INTRINSIC_FUNCTION(Sign, sign)
GENERATE_HLSL_INTRINSIC_FUNCTION(Step, step)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(Radians, radians)
GENERATE_HLSL_INTRINSIC_FUNCTION(ThreadId, thread_id)
GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index c3ecfc7c90d433..f39a68ba847e98 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2138,5 +2138,35 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int3 sign(double3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int4 sign(double4);
+
+//===--===//
+// radians builtins
+//===--===//
+
+/// \fn T radians(T Val)
+/// \brief Converts the specified value from degrees to radians.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half radians(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half2 radians(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half3 radians(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half4 radians(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float radians(float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float2 radians(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float3 radians(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float4 radians(float4);
+
} // namespace hlsl
#endif //_HLSL_HLSL_INTRINSICS_H_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index fbcba201a351a6..d2d2df829e7b11 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1896,6 +1896,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_hlsl_elementwise_radians:
case Builtin::BI__builtin_hlsl_elementwise_rsqrt:
case Builtin::BI__builtin_hlsl_elementwise_frac: {
if (CheckFloatOrHalfRepresentations
[clang] [clang][HLSL] Add radians intrinsic (PR #110802)
adam-yang wrote: @farzonl This is also ready for merging if there's no further feedback. https://github.com/llvm/llvm-project/pull/110802 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
https://github.com/adam-yang created
https://github.com/llvm/llvm-project/pull/113394
Fixes #88052
- Added the following intrinsics:
- `int_spv_uclamp`
- `int_spv_sclamp`
- `int_spv_fclamp`
- Update the clamp.hlsl unit tests to include SPIRV
- Added the SPIRV specific tests
>From 8cf5032a5580b97c4c4965e577374f627fbe0643 Mon Sep 17 00:00:00 2001
From: Adam Yang <31109344+adam-y...@users.noreply.github.com>
Date: Mon, 14 Oct 2024 16:35:45 -0700
Subject: [PATCH 1/3] Added the intrinsics and modified the clang test
---
clang/lib/CodeGen/CGBuiltin.cpp | 25 ++-
clang/test/CodeGenHLSL/builtins/clamp.hlsl| 174 +-
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 3 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 6 +
4 files changed, 122 insertions(+), 86 deletions(-)
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 1ad950798c2118..1087537ae4ee29 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18665,10 +18665,27 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned
BuiltinID,
if (auto *VecTy = Ty->getAs())
Ty = VecTy->getElementType();
IsUnsigned = Ty->isUnsignedIntegerType();
-return Builder.CreateIntrinsic(
-/*ReturnType=*/OpX->getType(),
-IsUnsigned ? Intrinsic::dx_uclamp : Intrinsic::dx_clamp,
-ArrayRef{OpX, OpMin, OpMax}, nullptr, "dx.clamp");
+switch (CGM.getTarget().getTriple().getArch()) {
+case llvm::Triple::dxil: {
+ return Builder.CreateIntrinsic(
+ /*ReturnType=*/OpX->getType(),
+ IsUnsigned ? Intrinsic::dx_uclamp : Intrinsic::dx_clamp,
+ ArrayRef{OpX, OpMin, OpMax}, nullptr, "dx.clamp");
+} break;
+case llvm::Triple::spirv: {
+ Intrinsic::ID Intr = Intrinsic::spv_sclamp;
+ if (Ty->isFloatingType()) {
+Intr = Intrinsic::spv_fclamp;
+ } else if (IsUnsigned) {
+Intr = Intrinsic::spv_uclamp;
+ }
+ return Builder.CreateIntrinsic(OpX->getType(), Intr,
+ ArrayRef{OpX, OpMin, OpMax},
+ nullptr, "spv.clamp");
+} break;
+default:
+ llvm_unreachable("Intrinsic clamp not supported by target architecture");
+}
}
case Builtin::BI__builtin_hlsl_cross: {
Value *Op0 = EmitScalarExpr(E->getArg(0));
diff --git a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
index af8f6b9733a071..806e786ae70931 100644
--- a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
@@ -1,133 +1,143 @@
// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.3-library %s \
// RUN: -fnative-half-type -emit-llvm -disable-llvm-passes -o - | \
-// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF
+// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF \
+// RUN: -DSCLAMP="dx.clamp" -DUCLAMP="dx.uclamp" -DFCLAMP="dx.clamp"
-DFNATTR="noundef"
// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.3-library %s \
// RUN: -emit-llvm -disable-llvm-passes -o - | \
-// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF
+// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF \
+// RUN: -DSCLAMP="dx.clamp" -DUCLAMP="dx.uclamp" -DFCLAMP="dx.clamp"
-DFNATTR="noundef"
+// RUN: %clang_cc1 -finclude-default-header -triple
spirv-unknown-vulkan-compute %s \
+// RUN: -fnative-half-type -emit-llvm -disable-llvm-passes -o - | \
+// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF \
+// RUN: -DSCLAMP="spv.sclamp" -DUCLAMP="spv.uclamp" -DFCLAMP="spv.fclamp"
-DFNATTR="spir_func noundef"
+// RUN: %clang_cc1 -finclude-default-header -triple
spirv-unknown-vulkan-compute %s \
+// RUN: -emit-llvm -disable-llvm-passes -o - | \
+// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF \
+// RUN: -DSCLAMP="spv.sclamp" -DUCLAMP="spv.uclamp" -DFCLAMP="spv.fclamp"
-DFNATTR="spir_func noundef"
#ifdef __HLSL_ENABLE_16_BIT
-// NATIVE_HALF-LABEL: define noundef i16 @_Z16test_clamp_short
-// NATIVE_HALF: call i16 @llvm.dx.clamp.i16(
+// NATIVE_HALF: define [[FNATTR]] i16 @_Z16test_clamp_short
+// NATIVE_HALF: call i16 @llvm.[[SCLAMP]].i16(
int16_t test_clamp_short(int16_t p0, int16_t p1) { return clamp(p0, p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <2 x i16> @_Z17test_clamp_short2
-// NATIVE_HALF: call <2 x i16> @llvm.dx.clamp.v2i16(
+// NATIVE_HALF: define [[FNATTR]] <2 x i16> @_Z17test_clamp_short2
+// NATIVE_HALF: call <2 x i16> @llvm.[[SCLAMP]].v2i16(
int16_t2 test_clamp_short2(int16_t2 p0, int16_t2 p1) { return clamp(p0,
p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <3 x i16> @_Z17test_clamp_short3
-// NATIVE_HALF: call <3 x i16> @llvm.dx.clamp.v3i16
+// NATIVE_HALF: define [[FNATTR]] <3 x i16> @_Z17test_clamp_short3
+// NATIVE_HALF: call <3 x i16> @llvm.[[SCLAMP]].v3i16
int16_t3 test_clamp_short3(int16_t3 p0, int16_t3 p1) { return clamp(p0,
p1,p1); }
-// NATIVE_HALF-LA
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/113394
>From 8cf5032a5580b97c4c4965e577374f627fbe0643 Mon Sep 17 00:00:00 2001
From: Adam Yang <31109344+adam-y...@users.noreply.github.com>
Date: Mon, 14 Oct 2024 16:35:45 -0700
Subject: [PATCH 1/4] Added the intrinsics and modified the clang test
---
clang/lib/CodeGen/CGBuiltin.cpp | 25 ++-
clang/test/CodeGenHLSL/builtins/clamp.hlsl| 174 +-
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 3 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 6 +
4 files changed, 122 insertions(+), 86 deletions(-)
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 1ad950798c2118..1087537ae4ee29 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18665,10 +18665,27 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned
BuiltinID,
if (auto *VecTy = Ty->getAs())
Ty = VecTy->getElementType();
IsUnsigned = Ty->isUnsignedIntegerType();
-return Builder.CreateIntrinsic(
-/*ReturnType=*/OpX->getType(),
-IsUnsigned ? Intrinsic::dx_uclamp : Intrinsic::dx_clamp,
-ArrayRef{OpX, OpMin, OpMax}, nullptr, "dx.clamp");
+switch (CGM.getTarget().getTriple().getArch()) {
+case llvm::Triple::dxil: {
+ return Builder.CreateIntrinsic(
+ /*ReturnType=*/OpX->getType(),
+ IsUnsigned ? Intrinsic::dx_uclamp : Intrinsic::dx_clamp,
+ ArrayRef{OpX, OpMin, OpMax}, nullptr, "dx.clamp");
+} break;
+case llvm::Triple::spirv: {
+ Intrinsic::ID Intr = Intrinsic::spv_sclamp;
+ if (Ty->isFloatingType()) {
+Intr = Intrinsic::spv_fclamp;
+ } else if (IsUnsigned) {
+Intr = Intrinsic::spv_uclamp;
+ }
+ return Builder.CreateIntrinsic(OpX->getType(), Intr,
+ ArrayRef{OpX, OpMin, OpMax},
+ nullptr, "spv.clamp");
+} break;
+default:
+ llvm_unreachable("Intrinsic clamp not supported by target architecture");
+}
}
case Builtin::BI__builtin_hlsl_cross: {
Value *Op0 = EmitScalarExpr(E->getArg(0));
diff --git a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
index af8f6b9733a071..806e786ae70931 100644
--- a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
@@ -1,133 +1,143 @@
// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.3-library %s \
// RUN: -fnative-half-type -emit-llvm -disable-llvm-passes -o - | \
-// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF
+// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF \
+// RUN: -DSCLAMP="dx.clamp" -DUCLAMP="dx.uclamp" -DFCLAMP="dx.clamp"
-DFNATTR="noundef"
// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.3-library %s \
// RUN: -emit-llvm -disable-llvm-passes -o - | \
-// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF
+// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF \
+// RUN: -DSCLAMP="dx.clamp" -DUCLAMP="dx.uclamp" -DFCLAMP="dx.clamp"
-DFNATTR="noundef"
+// RUN: %clang_cc1 -finclude-default-header -triple
spirv-unknown-vulkan-compute %s \
+// RUN: -fnative-half-type -emit-llvm -disable-llvm-passes -o - | \
+// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF \
+// RUN: -DSCLAMP="spv.sclamp" -DUCLAMP="spv.uclamp" -DFCLAMP="spv.fclamp"
-DFNATTR="spir_func noundef"
+// RUN: %clang_cc1 -finclude-default-header -triple
spirv-unknown-vulkan-compute %s \
+// RUN: -emit-llvm -disable-llvm-passes -o - | \
+// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF \
+// RUN: -DSCLAMP="spv.sclamp" -DUCLAMP="spv.uclamp" -DFCLAMP="spv.fclamp"
-DFNATTR="spir_func noundef"
#ifdef __HLSL_ENABLE_16_BIT
-// NATIVE_HALF-LABEL: define noundef i16 @_Z16test_clamp_short
-// NATIVE_HALF: call i16 @llvm.dx.clamp.i16(
+// NATIVE_HALF: define [[FNATTR]] i16 @_Z16test_clamp_short
+// NATIVE_HALF: call i16 @llvm.[[SCLAMP]].i16(
int16_t test_clamp_short(int16_t p0, int16_t p1) { return clamp(p0, p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <2 x i16> @_Z17test_clamp_short2
-// NATIVE_HALF: call <2 x i16> @llvm.dx.clamp.v2i16(
+// NATIVE_HALF: define [[FNATTR]] <2 x i16> @_Z17test_clamp_short2
+// NATIVE_HALF: call <2 x i16> @llvm.[[SCLAMP]].v2i16(
int16_t2 test_clamp_short2(int16_t2 p0, int16_t2 p1) { return clamp(p0,
p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <3 x i16> @_Z17test_clamp_short3
-// NATIVE_HALF: call <3 x i16> @llvm.dx.clamp.v3i16
+// NATIVE_HALF: define [[FNATTR]] <3 x i16> @_Z17test_clamp_short3
+// NATIVE_HALF: call <3 x i16> @llvm.[[SCLAMP]].v3i16
int16_t3 test_clamp_short3(int16_t3 p0, int16_t3 p1) { return clamp(p0,
p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <4 x i16> @_Z17test_clamp_short4
-// NATIVE_HALF: call <4 x i16> @llvm.dx.clamp.v4i16
+// NATIVE_HALF: define [[FNATTR]] <4 x i16> @_Z17test_clamp_short4
+// NATIVE_HALF: call <
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
https://github.com/adam-yang ready_for_review https://github.com/llvm/llvm-project/pull/113394 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
@@ -83,12 +83,210 @@ entry: ret i64 %0 } -declare half @llvm.dx.clamp.f16(half, half, half) -declare float @llvm.dx.clamp.f32(float, float, float) -declare double @llvm.dx.clamp.f64(double, double, double) -declare i16 @llvm.dx.clamp.i16(i16, i16, i16) -declare i32 @llvm.dx.clamp.i32(i32, i32, i32) -declare i64 @llvm.dx.clamp.i64(i64, i64, i64) +declare half @llvm.dx.nclamp.f16(half, half, half) adam-yang wrote: Done. https://github.com/llvm/llvm-project/pull/113394 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/113394
>From 8cf5032a5580b97c4c4965e577374f627fbe0643 Mon Sep 17 00:00:00 2001
From: Adam Yang <31109344+adam-y...@users.noreply.github.com>
Date: Mon, 14 Oct 2024 16:35:45 -0700
Subject: [PATCH 1/5] Added the intrinsics and modified the clang test
---
clang/lib/CodeGen/CGBuiltin.cpp | 25 ++-
clang/test/CodeGenHLSL/builtins/clamp.hlsl| 174 +-
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 3 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 6 +
4 files changed, 122 insertions(+), 86 deletions(-)
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 1ad950798c2118..1087537ae4ee29 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18665,10 +18665,27 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned
BuiltinID,
if (auto *VecTy = Ty->getAs())
Ty = VecTy->getElementType();
IsUnsigned = Ty->isUnsignedIntegerType();
-return Builder.CreateIntrinsic(
-/*ReturnType=*/OpX->getType(),
-IsUnsigned ? Intrinsic::dx_uclamp : Intrinsic::dx_clamp,
-ArrayRef{OpX, OpMin, OpMax}, nullptr, "dx.clamp");
+switch (CGM.getTarget().getTriple().getArch()) {
+case llvm::Triple::dxil: {
+ return Builder.CreateIntrinsic(
+ /*ReturnType=*/OpX->getType(),
+ IsUnsigned ? Intrinsic::dx_uclamp : Intrinsic::dx_clamp,
+ ArrayRef{OpX, OpMin, OpMax}, nullptr, "dx.clamp");
+} break;
+case llvm::Triple::spirv: {
+ Intrinsic::ID Intr = Intrinsic::spv_sclamp;
+ if (Ty->isFloatingType()) {
+Intr = Intrinsic::spv_fclamp;
+ } else if (IsUnsigned) {
+Intr = Intrinsic::spv_uclamp;
+ }
+ return Builder.CreateIntrinsic(OpX->getType(), Intr,
+ ArrayRef{OpX, OpMin, OpMax},
+ nullptr, "spv.clamp");
+} break;
+default:
+ llvm_unreachable("Intrinsic clamp not supported by target architecture");
+}
}
case Builtin::BI__builtin_hlsl_cross: {
Value *Op0 = EmitScalarExpr(E->getArg(0));
diff --git a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
index af8f6b9733a071..806e786ae70931 100644
--- a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
@@ -1,133 +1,143 @@
// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.3-library %s \
// RUN: -fnative-half-type -emit-llvm -disable-llvm-passes -o - | \
-// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF
+// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF \
+// RUN: -DSCLAMP="dx.clamp" -DUCLAMP="dx.uclamp" -DFCLAMP="dx.clamp"
-DFNATTR="noundef"
// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.3-library %s \
// RUN: -emit-llvm -disable-llvm-passes -o - | \
-// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF
+// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF \
+// RUN: -DSCLAMP="dx.clamp" -DUCLAMP="dx.uclamp" -DFCLAMP="dx.clamp"
-DFNATTR="noundef"
+// RUN: %clang_cc1 -finclude-default-header -triple
spirv-unknown-vulkan-compute %s \
+// RUN: -fnative-half-type -emit-llvm -disable-llvm-passes -o - | \
+// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF \
+// RUN: -DSCLAMP="spv.sclamp" -DUCLAMP="spv.uclamp" -DFCLAMP="spv.fclamp"
-DFNATTR="spir_func noundef"
+// RUN: %clang_cc1 -finclude-default-header -triple
spirv-unknown-vulkan-compute %s \
+// RUN: -emit-llvm -disable-llvm-passes -o - | \
+// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF \
+// RUN: -DSCLAMP="spv.sclamp" -DUCLAMP="spv.uclamp" -DFCLAMP="spv.fclamp"
-DFNATTR="spir_func noundef"
#ifdef __HLSL_ENABLE_16_BIT
-// NATIVE_HALF-LABEL: define noundef i16 @_Z16test_clamp_short
-// NATIVE_HALF: call i16 @llvm.dx.clamp.i16(
+// NATIVE_HALF: define [[FNATTR]] i16 @_Z16test_clamp_short
+// NATIVE_HALF: call i16 @llvm.[[SCLAMP]].i16(
int16_t test_clamp_short(int16_t p0, int16_t p1) { return clamp(p0, p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <2 x i16> @_Z17test_clamp_short2
-// NATIVE_HALF: call <2 x i16> @llvm.dx.clamp.v2i16(
+// NATIVE_HALF: define [[FNATTR]] <2 x i16> @_Z17test_clamp_short2
+// NATIVE_HALF: call <2 x i16> @llvm.[[SCLAMP]].v2i16(
int16_t2 test_clamp_short2(int16_t2 p0, int16_t2 p1) { return clamp(p0,
p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <3 x i16> @_Z17test_clamp_short3
-// NATIVE_HALF: call <3 x i16> @llvm.dx.clamp.v3i16
+// NATIVE_HALF: define [[FNATTR]] <3 x i16> @_Z17test_clamp_short3
+// NATIVE_HALF: call <3 x i16> @llvm.[[SCLAMP]].v3i16
int16_t3 test_clamp_short3(int16_t3 p0, int16_t3 p1) { return clamp(p0,
p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <4 x i16> @_Z17test_clamp_short4
-// NATIVE_HALF: call <4 x i16> @llvm.dx.clamp.v4i16
+// NATIVE_HALF: define [[FNATTR]] <4 x i16> @_Z17test_clamp_short4
+// NATIVE_HALF: call <
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
@@ -0,0 +1,130 @@
+; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
adam-yang wrote:
I'm just following what was done in DX; there's clamp and clamp-vec.
https://github.com/llvm/llvm-project/pull/113394
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/113394
>From 8cf5032a5580b97c4c4965e577374f627fbe0643 Mon Sep 17 00:00:00 2001
From: Adam Yang <31109344+adam-y...@users.noreply.github.com>
Date: Mon, 14 Oct 2024 16:35:45 -0700
Subject: [PATCH 1/6] Added the intrinsics and modified the clang test
---
clang/lib/CodeGen/CGBuiltin.cpp | 25 ++-
clang/test/CodeGenHLSL/builtins/clamp.hlsl| 174 +-
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 3 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 6 +
4 files changed, 122 insertions(+), 86 deletions(-)
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 1ad950798c2118..1087537ae4ee29 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18665,10 +18665,27 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned
BuiltinID,
if (auto *VecTy = Ty->getAs())
Ty = VecTy->getElementType();
IsUnsigned = Ty->isUnsignedIntegerType();
-return Builder.CreateIntrinsic(
-/*ReturnType=*/OpX->getType(),
-IsUnsigned ? Intrinsic::dx_uclamp : Intrinsic::dx_clamp,
-ArrayRef{OpX, OpMin, OpMax}, nullptr, "dx.clamp");
+switch (CGM.getTarget().getTriple().getArch()) {
+case llvm::Triple::dxil: {
+ return Builder.CreateIntrinsic(
+ /*ReturnType=*/OpX->getType(),
+ IsUnsigned ? Intrinsic::dx_uclamp : Intrinsic::dx_clamp,
+ ArrayRef{OpX, OpMin, OpMax}, nullptr, "dx.clamp");
+} break;
+case llvm::Triple::spirv: {
+ Intrinsic::ID Intr = Intrinsic::spv_sclamp;
+ if (Ty->isFloatingType()) {
+Intr = Intrinsic::spv_fclamp;
+ } else if (IsUnsigned) {
+Intr = Intrinsic::spv_uclamp;
+ }
+ return Builder.CreateIntrinsic(OpX->getType(), Intr,
+ ArrayRef{OpX, OpMin, OpMax},
+ nullptr, "spv.clamp");
+} break;
+default:
+ llvm_unreachable("Intrinsic clamp not supported by target architecture");
+}
}
case Builtin::BI__builtin_hlsl_cross: {
Value *Op0 = EmitScalarExpr(E->getArg(0));
diff --git a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
index af8f6b9733a071..806e786ae70931 100644
--- a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
@@ -1,133 +1,143 @@
// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.3-library %s \
// RUN: -fnative-half-type -emit-llvm -disable-llvm-passes -o - | \
-// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF
+// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF \
+// RUN: -DSCLAMP="dx.clamp" -DUCLAMP="dx.uclamp" -DFCLAMP="dx.clamp"
-DFNATTR="noundef"
// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.3-library %s \
// RUN: -emit-llvm -disable-llvm-passes -o - | \
-// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF
+// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF \
+// RUN: -DSCLAMP="dx.clamp" -DUCLAMP="dx.uclamp" -DFCLAMP="dx.clamp"
-DFNATTR="noundef"
+// RUN: %clang_cc1 -finclude-default-header -triple
spirv-unknown-vulkan-compute %s \
+// RUN: -fnative-half-type -emit-llvm -disable-llvm-passes -o - | \
+// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF \
+// RUN: -DSCLAMP="spv.sclamp" -DUCLAMP="spv.uclamp" -DFCLAMP="spv.fclamp"
-DFNATTR="spir_func noundef"
+// RUN: %clang_cc1 -finclude-default-header -triple
spirv-unknown-vulkan-compute %s \
+// RUN: -emit-llvm -disable-llvm-passes -o - | \
+// RUN: FileCheck %s --check-prefixes=CHECK,NO_HALF \
+// RUN: -DSCLAMP="spv.sclamp" -DUCLAMP="spv.uclamp" -DFCLAMP="spv.fclamp"
-DFNATTR="spir_func noundef"
#ifdef __HLSL_ENABLE_16_BIT
-// NATIVE_HALF-LABEL: define noundef i16 @_Z16test_clamp_short
-// NATIVE_HALF: call i16 @llvm.dx.clamp.i16(
+// NATIVE_HALF: define [[FNATTR]] i16 @_Z16test_clamp_short
+// NATIVE_HALF: call i16 @llvm.[[SCLAMP]].i16(
int16_t test_clamp_short(int16_t p0, int16_t p1) { return clamp(p0, p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <2 x i16> @_Z17test_clamp_short2
-// NATIVE_HALF: call <2 x i16> @llvm.dx.clamp.v2i16(
+// NATIVE_HALF: define [[FNATTR]] <2 x i16> @_Z17test_clamp_short2
+// NATIVE_HALF: call <2 x i16> @llvm.[[SCLAMP]].v2i16(
int16_t2 test_clamp_short2(int16_t2 p0, int16_t2 p1) { return clamp(p0,
p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <3 x i16> @_Z17test_clamp_short3
-// NATIVE_HALF: call <3 x i16> @llvm.dx.clamp.v3i16
+// NATIVE_HALF: define [[FNATTR]] <3 x i16> @_Z17test_clamp_short3
+// NATIVE_HALF: call <3 x i16> @llvm.[[SCLAMP]].v3i16
int16_t3 test_clamp_short3(int16_t3 p0, int16_t3 p1) { return clamp(p0,
p1,p1); }
-// NATIVE_HALF-LABEL: define noundef <4 x i16> @_Z17test_clamp_short4
-// NATIVE_HALF: call <4 x i16> @llvm.dx.clamp.v4i16
+// NATIVE_HALF: define [[FNATTR]] <4 x i16> @_Z17test_clamp_short4
+// NATIVE_HALF: call <
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
@@ -18661,14 +18661,30 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned
BuiltinID,
Value *OpMax = EmitScalarExpr(E->getArg(2));
QualType Ty = E->getArg(0)->getType();
-bool IsUnsigned = false;
if (auto *VecTy = Ty->getAs())
Ty = VecTy->getElementType();
-IsUnsigned = Ty->isUnsignedIntegerType();
-return Builder.CreateIntrinsic(
-/*ReturnType=*/OpX->getType(),
-IsUnsigned ? Intrinsic::dx_uclamp : Intrinsic::dx_clamp,
-ArrayRef{OpX, OpMin, OpMax}, nullptr, "dx.clamp");
+bool IsUnsigned = Ty->isUnsignedIntegerType();
+switch (CGM.getTarget().getTriple().getArch()) {
+case llvm::Triple::dxil: {
+ return Builder.CreateIntrinsic(
+ /*ReturnType=*/OpX->getType(),
+ IsUnsigned ? Intrinsic::dx_uclamp : Intrinsic::dx_clamp,
+ ArrayRef{OpX, OpMin, OpMax}, nullptr, "dx.clamp");
+} break;
+case llvm::Triple::spirv: {
adam-yang wrote:
This is all fixed, the two sides now match nicely.
https://github.com/llvm/llvm-project/pull/113394
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
@@ -2559,6 +2559,12 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg, } break; case Intrinsic::spv_saturate: return selectSaturate(ResVReg, ResType, I); + case Intrinsic::spv_fclamp: +return selectExtInst(ResVReg, ResType, I, CL::fclamp, GL::FClamp); adam-yang wrote: Anyways, it's now doing the right thing and tests verify it. https://github.com/llvm/llvm-project/pull/113394 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
@@ -40,8 +40,9 @@ def int_dx_cast_handle : Intrinsic<[llvm_any_ty], [llvm_any_ty]>; def int_dx_all : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty], [IntrNoMem]>; def int_dx_any : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty], [IntrNoMem]>; -def int_dx_clamp : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>; def int_dx_uclamp : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>; +def int_dx_sclamp : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>; +def int_dx_nclamp : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>; adam-yang wrote: They are now updated along with getting rid of clamp-vec.ll https://github.com/llvm/llvm-project/pull/113394 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
@@ -0,0 +1,130 @@
+; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
adam-yang wrote:
clamp-vec is now gone. The vector test cases in DirectX clamp.ll now also go
through scalarizing and lowering all the way to DXIL ops.
https://github.com/llvm/llvm-project/pull/113394
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
@@ -2559,6 +2559,12 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg, } break; case Intrinsic::spv_saturate: return selectSaturate(ResVReg, ResType, I); + case Intrinsic::spv_fclamp: +return selectExtInst(ResVReg, ResType, I, CL::fclamp, GL::FClamp); adam-yang wrote: looks like `CL::nclamp` doesn't exist, but `CL::fclamp` actually has the same behaviour as `GL::NClamp`: in fmin/fmax, if one of the args is NaN, it returns the other argument. https://github.com/llvm/llvm-project/pull/113394 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
@@ -1,133 +1,143 @@ // RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.3-library %s \ // RUN: -fnative-half-type -emit-llvm -disable-llvm-passes -o - | \ -// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF +// RUN: FileCheck %s --check-prefixes=CHECK,NATIVE_HALF \ +// RUN: -DSCLAMP="dx.clamp" -DUCLAMP="dx.uclamp" -DFCLAMP="dx.clamp" -DFNATTR="noundef" adam-yang wrote: Fixed. https://github.com/llvm/llvm-project/pull/113394 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add radians intrinsic (PR #110802)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/110802
>From 08f0b8a10f3e5292fe2f91946392141c9239ec2c Mon Sep 17 00:00:00 2001
From: Adam Yang
Date: Wed, 2 Oct 2024 01:21:19 -0700
Subject: [PATCH 1/4] Added radians to clang
---
clang/include/clang/Basic/Builtins.td | 6 +++
clang/lib/CodeGen/CGBuiltin.cpp | 8 +++
clang/lib/CodeGen/CGHLSLRuntime.h | 1 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 30 +++
clang/lib/Sema/SemaHLSL.cpp | 1 +
clang/test/CodeGenHLSL/builtins/radians.hlsl | 54 +++
.../SemaHLSL/BuiltIns/radians-errors.hlsl | 27 ++
7 files changed, 127 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/radians.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/radians-errors.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 5b504666365b32..38064e32a3f655 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4818,6 +4818,12 @@ def HLSLStep: LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLRadians : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_elementwise_radians"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c864714182e019..02de763ecd3f52 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18877,6 +18877,14 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
retType, CGM.getHLSLRuntime().getSignIntrinsic(),
ArrayRef{Op0}, nullptr, "hlsl.sign");
}
+ case Builtin::BI__builtin_hlsl_elementwise_radians: {
+Value *Op0 = EmitScalarExpr(E->getArg(0));
+if (!E->getArg(0)->getType()->hasFloatingRepresentation())
+ llvm_unreachable("radians operand must have a float representation");
+return Builder.CreateIntrinsic(
+/*ReturnType=*/Op0->getType(),
CGM.getHLSLRuntime().getRadiansIntrinsic(),
+ArrayRef{Op0}, nullptr, "hlsl.radians");
+ }
}
return nullptr;
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index a8aabca7348ffb..c807a3216ec9ee 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -82,6 +82,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(Saturate, saturate)
GENERATE_HLSL_INTRINSIC_FUNCTION(Sign, sign)
GENERATE_HLSL_INTRINSIC_FUNCTION(Step, step)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(Radians, radians)
GENERATE_HLSL_INTRINSIC_FUNCTION(ThreadId, thread_id)
GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index d28f204e352de5..6393a16bc423c0 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2116,5 +2116,35 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int3 sign(double3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int4 sign(double4);
+
+//===--===//
+// radians builtins
+//===--===//
+
+/// \fn T radians(T Val)
+/// \brief Converts the specified value from degrees to radians.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half radians(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half2 radians(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half3 radians(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+half4 radians(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float radians(float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float2 radians(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float3 radians(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
+float4 radians(float4);
+
} // namespace hlsl
#endif //_HLSL_HLSL_INTRINSICS_H_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 5eda22f560c9d2..ef2721221a4e21 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1861,6 +1861,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_hlsl_elementwise_radians:
case Builtin::BI__builtin_hlsl_elementwise_rsqrt:
case Builtin::BI__builtin_hlsl_elementwise_frac: {
if (CheckFloatOrHalfRepresentations
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/111883
>From 0f97b92617dcadd78362f24f3ef0719c19f6f3f6 Mon Sep 17 00:00:00 2001
From: Adam Yang
Date: Thu, 10 Oct 2024 10:20:37 -0700
Subject: [PATCH 1/6] Added the intrinsic to clang
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 4
clang/lib/CodeGen/CGHLSLRuntime.h | 1 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 12 ++
clang/lib/Sema/SemaHLSL.cpp | 5 +
...roupMemoryBarrierWithGroupSync-errors.hlsl | 6 +
.../GroupMemoryBarrierWithGroupSync.hlsl | 22 +++
7 files changed, 56 insertions(+)
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 9ebee81fcb0d3d..96fea17badfe9e 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4830,6 +4830,12 @@ def HLSLRadians : LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLGroupMemoryBarrierWithGroupSync: LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_group_memory_barrier_with_group_sync"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 57705f2d2d0423..45890019ee2a64 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18901,6 +18901,10 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
CGM.getHLSLRuntime().getRadiansIntrinsic(), ArrayRef{Op0},
nullptr, "hlsl.radians");
}
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+Intrinsic::ID ID =
CGM.getHLSLRuntime().getGroupMemoryBarrierWithGroupSyncIntrinsic();
+return EmitRuntimeCall(Intrinsic::getDeclaration(&CGM.getModule(), ID));
+ }
}
return nullptr;
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 05ff325216f55b..1e095cedf58731 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -89,6 +89,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
GENERATE_HLSL_INTRINSIC_FUNCTION(WaveIsFirstLane, wave_is_first_lane)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrierWithGroupSync,
groupMemoryBarrierWithGroupSync)
//===--===//
// End of reserved area for HLSL intrinsic getters.
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index f39a68ba847e98..637df8924e2468 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2168,5 +2168,17 @@ float3 radians(float3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
float4 radians(float4);
+//===--===//
+// GroupMemoryBarrierWithGroupSync builtins
+//===--===//
+
+/// \fn void GroupMemoryBarrierWithGroupSync(void)
+/// \brief Blocks execution of all threads in a group until all group shared
+/// accesses have been completed and all threads in the group have reached this
+/// call.
+
+_HLSL_BUILTIN_ALIAS(__builtin_group_memory_barrier_with_group_sync)
+void GroupMemoryBarrierWithGroupSync(void);
+
} // namespace hlsl
#endif //_HLSL_HLSL_INTRINSICS_H_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index d2d2df829e7b11..30480f8f57c12d 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2018,6 +2018,11 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+if (SemaRef.checkArgCountAtMost(TheCall, 0))
+ return true;
+break;
+ }
}
return false;
}
diff --git
a/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
new file mode 100644
index 00..9ffc1ebce1c909
--- /dev/null
+++
b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
@@ -0,0 +1,6 @@
+// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only
-disable-llvm-passes -verify -verify-ignore-unexpected
+
+void test_too_many_arg() {
+ __builtin_group_memory_barrier_with_group_sync(0);
+ // expected-error@-1 {{too many argume
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
@@ -2018,6 +2018,11 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_hlsl_group_memory_barrier_with_group_sync: {
+if (SemaRef.checkArgCountAtMost(TheCall, 0))
adam-yang wrote:
Verified that removing this check the error still works.
https://github.com/llvm/llvm-project/pull/111883
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
@@ -2018,6 +2018,11 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_hlsl_group_memory_barrier_with_group_sync: {
+if (SemaRef.checkArgCountAtMost(TheCall, 0))
adam-yang wrote:
The reason is `checkArgCount` actually expects there to be at least one arg. I
opted to just call a different function rather than modifying `checkArgCount`.
https://github.com/llvm/llvm-project/pull/111883
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/111883
>From 0f97b92617dcadd78362f24f3ef0719c19f6f3f6 Mon Sep 17 00:00:00 2001
From: Adam Yang
Date: Thu, 10 Oct 2024 10:20:37 -0700
Subject: [PATCH 1/5] Added the intrinsic to clang
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 4
clang/lib/CodeGen/CGHLSLRuntime.h | 1 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 12 ++
clang/lib/Sema/SemaHLSL.cpp | 5 +
...roupMemoryBarrierWithGroupSync-errors.hlsl | 6 +
.../GroupMemoryBarrierWithGroupSync.hlsl | 22 +++
7 files changed, 56 insertions(+)
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 9ebee81fcb0d3d..96fea17badfe9e 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4830,6 +4830,12 @@ def HLSLRadians : LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLGroupMemoryBarrierWithGroupSync: LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_group_memory_barrier_with_group_sync"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 57705f2d2d0423..45890019ee2a64 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18901,6 +18901,10 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
CGM.getHLSLRuntime().getRadiansIntrinsic(), ArrayRef{Op0},
nullptr, "hlsl.radians");
}
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+Intrinsic::ID ID =
CGM.getHLSLRuntime().getGroupMemoryBarrierWithGroupSyncIntrinsic();
+return EmitRuntimeCall(Intrinsic::getDeclaration(&CGM.getModule(), ID));
+ }
}
return nullptr;
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 05ff325216f55b..1e095cedf58731 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -89,6 +89,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
GENERATE_HLSL_INTRINSIC_FUNCTION(WaveIsFirstLane, wave_is_first_lane)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrierWithGroupSync,
groupMemoryBarrierWithGroupSync)
//===--===//
// End of reserved area for HLSL intrinsic getters.
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index f39a68ba847e98..637df8924e2468 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2168,5 +2168,17 @@ float3 radians(float3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
float4 radians(float4);
+//===--===//
+// GroupMemoryBarrierWithGroupSync builtins
+//===--===//
+
+/// \fn void GroupMemoryBarrierWithGroupSync(void)
+/// \brief Blocks execution of all threads in a group until all group shared
+/// accesses have been completed and all threads in the group have reached this
+/// call.
+
+_HLSL_BUILTIN_ALIAS(__builtin_group_memory_barrier_with_group_sync)
+void GroupMemoryBarrierWithGroupSync(void);
+
} // namespace hlsl
#endif //_HLSL_HLSL_INTRINSICS_H_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index d2d2df829e7b11..30480f8f57c12d 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2018,6 +2018,11 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+if (SemaRef.checkArgCountAtMost(TheCall, 0))
+ return true;
+break;
+ }
}
return false;
}
diff --git
a/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
new file mode 100644
index 00..9ffc1ebce1c909
--- /dev/null
+++
b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
@@ -0,0 +1,6 @@
+// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only
-disable-llvm-passes -verify -verify-ignore-unexpected
+
+void test_too_many_arg() {
+ __builtin_group_memory_barrier_with_group_sync(0);
+ // expected-error@-1 {{too many argume
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
@@ -0,0 +1,6 @@ +// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -emit-llvm-only -disable-llvm-passes -verify -verify-ignore-unexpected adam-yang wrote: It's removed now. https://github.com/llvm/llvm-project/pull/111883 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Added clamp intrinsic (PR #113394)
https://github.com/adam-yang closed https://github.com/llvm/llvm-project/pull/113394 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/111883
>From b7d329a60d6de140df9d7a2484e7a998c3a7b532 Mon Sep 17 00:00:00 2001
From: Adam Yang
Date: Thu, 10 Oct 2024 10:20:37 -0700
Subject: [PATCH 1/7] Added the intrinsic to clang
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 4
clang/lib/CodeGen/CGHLSLRuntime.h | 1 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 12 ++
clang/lib/Sema/SemaHLSL.cpp | 4
...roupMemoryBarrierWithGroupSync-errors.hlsl | 6 +
.../GroupMemoryBarrierWithGroupSync.hlsl | 22 +++
7 files changed, 55 insertions(+)
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index db5cd73fba8ad1..dcde6b005c9c3e 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4924,6 +4924,12 @@ def HLSLClip: LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLGroupMemoryBarrierWithGroupSync: LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_group_memory_barrier_with_group_sync"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index cb9c23b8e0a0d0..3b6ba23a85a010 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19446,6 +19446,10 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
assert(E->getArg(0)->getType()->hasFloatingRepresentation() &&
"clip operands types mismatch");
return handleHlslClip(E, this);
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+Intrinsic::ID ID =
CGM.getHLSLRuntime().getGroupMemoryBarrierWithGroupSyncIntrinsic();
+return EmitRuntimeCall(Intrinsic::getDeclaration(&CGM.getModule(), ID));
+ }
}
return nullptr;
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 854214d6bc0677..767baa441e25a2 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -103,6 +103,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(CreateHandleFromBinding, handle_fromBinding)
GENERATE_HLSL_INTRINSIC_FUNCTION(BufferUpdateCounter, bufferUpdateCounter)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrierWithGroupSync,
groupMemoryBarrierWithGroupSync)
//===--===//
// End of reserved area for HLSL intrinsic getters.
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index a3e0b5c65a6f52..aacf3ed31ef64b 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2481,5 +2481,17 @@ float3 radians(float3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
float4 radians(float4);
+//===--===//
+// GroupMemoryBarrierWithGroupSync builtins
+//===--===//
+
+/// \fn void GroupMemoryBarrierWithGroupSync(void)
+/// \brief Blocks execution of all threads in a group until all group shared
+/// accesses have been completed and all threads in the group have reached this
+/// call.
+
+_HLSL_BUILTIN_ALIAS(__builtin_group_memory_barrier_with_group_sync)
+void GroupMemoryBarrierWithGroupSync(void);
+
} // namespace hlsl
#endif //_HLSL_HLSL_INTRINSICS_H_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 8b2f24a8e4be0a..154f39f716bff7 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2239,6 +2239,10 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
<< 1;
return true;
}
+ }
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+if (SemaRef.checkArgCountAtMost(TheCall, 0))
+ return true;
break;
}
}
diff --git
a/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
new file mode 100644
index 00..9ffc1ebce1c909
--- /dev/null
+++
b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
@@ -0,0 +1,6 @@
+// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only
-disable-llvm-passes -verify -verify-ignore-unexpected
+
+void test_too_many_arg() {
+ __builtin_group_memory_barrier_with_group_sync(0);
+ // expected-err
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/111883
>From b7d329a60d6de140df9d7a2484e7a998c3a7b532 Mon Sep 17 00:00:00 2001
From: Adam Yang
Date: Thu, 10 Oct 2024 10:20:37 -0700
Subject: [PATCH 1/8] Added the intrinsic to clang
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 4
clang/lib/CodeGen/CGHLSLRuntime.h | 1 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 12 ++
clang/lib/Sema/SemaHLSL.cpp | 4
...roupMemoryBarrierWithGroupSync-errors.hlsl | 6 +
.../GroupMemoryBarrierWithGroupSync.hlsl | 22 +++
7 files changed, 55 insertions(+)
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
create mode 100644
clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index db5cd73fba8ad1..dcde6b005c9c3e 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4924,6 +4924,12 @@ def HLSLClip: LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLGroupMemoryBarrierWithGroupSync: LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_group_memory_barrier_with_group_sync"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index cb9c23b8e0a0d0..3b6ba23a85a010 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19446,6 +19446,10 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
assert(E->getArg(0)->getType()->hasFloatingRepresentation() &&
"clip operands types mismatch");
return handleHlslClip(E, this);
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+Intrinsic::ID ID =
CGM.getHLSLRuntime().getGroupMemoryBarrierWithGroupSyncIntrinsic();
+return EmitRuntimeCall(Intrinsic::getDeclaration(&CGM.getModule(), ID));
+ }
}
return nullptr;
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 854214d6bc0677..767baa441e25a2 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -103,6 +103,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(CreateHandleFromBinding, handle_fromBinding)
GENERATE_HLSL_INTRINSIC_FUNCTION(BufferUpdateCounter, bufferUpdateCounter)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrierWithGroupSync,
groupMemoryBarrierWithGroupSync)
//===--===//
// End of reserved area for HLSL intrinsic getters.
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index a3e0b5c65a6f52..aacf3ed31ef64b 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2481,5 +2481,17 @@ float3 radians(float3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians)
float4 radians(float4);
+//===--===//
+// GroupMemoryBarrierWithGroupSync builtins
+//===--===//
+
+/// \fn void GroupMemoryBarrierWithGroupSync(void)
+/// \brief Blocks execution of all threads in a group until all group shared
+/// accesses have been completed and all threads in the group have reached this
+/// call.
+
+_HLSL_BUILTIN_ALIAS(__builtin_group_memory_barrier_with_group_sync)
+void GroupMemoryBarrierWithGroupSync(void);
+
} // namespace hlsl
#endif //_HLSL_HLSL_INTRINSICS_H_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 8b2f24a8e4be0a..154f39f716bff7 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2239,6 +2239,10 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
<< 1;
return true;
}
+ }
+ case Builtin::BI__builtin_group_memory_barrier_with_group_sync: {
+if (SemaRef.checkArgCountAtMost(TheCall, 0))
+ return true;
break;
}
}
diff --git
a/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
new file mode 100644
index 00..9ffc1ebce1c909
--- /dev/null
+++
b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrierWithGroupSync-errors.hlsl
@@ -0,0 +1,6 @@
+// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only
-disable-llvm-passes -verify -verify-ignore-unexpected
+
+void test_too_many_arg() {
+ __builtin_group_memory_barrier_with_group_sync(0);
+ // expected-err
[clang] [clang][HLSL] Add GroupMemoryBarrierWithGroupSync intrinsic (PR #111883)
https://github.com/adam-yang closed https://github.com/llvm/llvm-project/pull/111883 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From e97cf4d2158fff06e7a541da73c442a9daadc8f5 Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/2] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index e2c3d3c535571c..bf8d7241e4b7bc 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4774,6 +4774,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index 8020be6c57bcfb..e431d686b11f7b 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9272,6 +9272,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 7588f8427cdd38..2fa044c0439e66 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19053,6 +19053,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19362,6 +19379,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang ready_for_review https://github.com/llvm/llvm-project/pull/118580 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
adam-yang wrote: Originally based on https://github.com/llvm/llvm-project/pull/112400. https://github.com/llvm/llvm-project/pull/118580 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From aa2c7eda19f3a89112e0f6eb8630cef58d833abb Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/2] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index e2c3d3c535571c..bf8d7241e4b7bc 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4774,6 +4774,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index 2137cb713164ad..17ebc54ba485d2 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9272,6 +9272,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 497a0b3952af8c..d4db04bad50726 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19117,6 +19117,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19426,6 +19443,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang created
https://github.com/llvm/llvm-project/pull/118580
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in DXIL.td
- add test cases to illustrate passes
>From e97cf4d2158fff06e7a541da73c442a9daadc8f5 Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index e2c3d3c535571c..bf8d7241e4b7bc 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4774,6 +4774,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index 8020be6c57bcfb..e431d686b11f7b 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9272,6 +9272,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 7588f8427cdd38..2fa044c0439e66 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19053,6 +19053,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnV
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From d435e26eb327b4d6c1e3530586b97372292ce214 Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/2] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 32a09e2ceb3857..5d825eba1ab7f4 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4774,6 +4774,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index f185a98720f5b9..d33c760fd214ca 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9272,6 +9272,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c2e983eebebc10..68f52ff2c36190 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19126,6 +19126,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19435,6 +19452,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From d435e26eb327b4d6c1e3530586b97372292ce214 Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/5] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 32a09e2ceb3857..5d825eba1ab7f4 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4774,6 +4774,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index f185a98720f5b9..d33c760fd214ca 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9272,6 +9272,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c2e983eebebc10..68f52ff2c36190 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19126,6 +19126,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19435,6 +19452,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
@@ -9272,6 +9272,8 @@ def err_typecheck_expect_scalar_or_vector : Error< "a vector of such type is required">; def err_typecheck_expect_any_scalar_or_vector : Error< "invalid operand of type %0 where a scalar or vector is required">; +def err_typecheck_expect_scalar_or_vector_not_type : Error< adam-yang wrote: Merged `err_typecheck_expect_scalar_or_vector_not_type` and `err_typecheck_expect_any_scalar_or_vector` by putting ` where a scalar or vector is required` behind a `%select`. https://github.com/llvm/llvm-project/pull/118580 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From d435e26eb327b4d6c1e3530586b97372292ce214 Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/5] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 32a09e2ceb3857..5d825eba1ab7f4 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4774,6 +4774,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index f185a98720f5b9..d33c760fd214ca 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9272,6 +9272,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c2e983eebebc10..68f52ff2c36190 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19126,6 +19126,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19435,6 +19452,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From d435e26eb327b4d6c1e3530586b97372292ce214 Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/4] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 32a09e2ceb3857..5d825eba1ab7f4 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4774,6 +4774,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index f185a98720f5b9..d33c760fd214ca 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9272,6 +9272,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c2e983eebebc10..68f52ff2c36190 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19126,6 +19126,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19435,6 +19452,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
@@ -1843,6 +1843,23 @@ static bool CheckAnyScalarOrVector(Sema *S, CallExpr
*TheCall,
return false;
}
+static bool CheckNotBoolType(Sema *S, CallExpr *TheCall, unsigned ArgIndex) {
adam-yang wrote:
Addressed both these comments.
https://github.com/llvm/llvm-project/pull/118580
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From d435e26eb327b4d6c1e3530586b97372292ce214 Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/3] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 32a09e2ceb3857..5d825eba1ab7f4 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4774,6 +4774,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index f185a98720f5b9..d33c760fd214ca 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9272,6 +9272,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c2e983eebebc10..68f52ff2c36190 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19126,6 +19126,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19435,6 +19452,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang edited https://github.com/llvm/llvm-project/pull/118580 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang edited https://github.com/llvm/llvm-project/pull/118580 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From f98e5113bc83cae2118552466288a874d5e5d63d Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/8] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 468c16050e2bf0..12b96672ce466f 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4793,6 +4793,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index d4e897868f1a9a..e74efd560fd329 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9287,6 +9287,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index ca03fb665d423d..8770c8f4953fce 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19186,6 +19186,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19505,6 +19522,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From f98e5113bc83cae2118552466288a874d5e5d63d Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/7] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 468c16050e2bf0..12b96672ce466f 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4793,6 +4793,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index d4e897868f1a9a..e74efd560fd329 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9287,6 +9287,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index ca03fb665d423d..8770c8f4953fce 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19186,6 +19186,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19505,6 +19522,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From 9a1ad541ec6e2fbd72bd88727a5910da39a5b840 Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/6] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 468c16050e2bf0..12b96672ce466f 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4793,6 +4793,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index ab2d6237c1cab8..d6295e31d2548e 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9287,6 +9287,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 573be932f8b1aa..2f754f595041ed 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19198,6 +19198,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19517,6 +19534,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveMax` intrinsic (PR #123428)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/123428
>From 9a0cf138d99ebc9ae18db054b4d8eaa34e8174a8 Mon Sep 17 00:00:00 2001
From: Adam Yang <31109344+adam-y...@users.noreply.github.com>
Date: Fri, 17 Jan 2025 16:14:44 -0800
Subject: [PATCH 1/3] Added WaveActiveMax
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 36 +
clang/lib/Sema/SemaHLSL.cpp | 1 +
.../CodeGenHLSL/builtins/WaveActiveMax.hlsl | 46 ++
.../BuiltIns/WaveActiveMax-errors.hlsl| 29
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 2 +
llvm/lib/Target/DirectX/DXIL.td | 6 +
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 33
llvm/test/CodeGen/DirectX/WaveActiveMax.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveMax.ll| 55 +++
12 files changed, 361 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveMax.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveMax-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveMax.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveMax.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index bbf4886b5cf053..a7f17ec4c8ef8f 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4795,6 +4795,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveMax : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_max"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_active_sum"];
let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index b80833fd91884d..57d8eec24a5a0b 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19203,6 +19203,25 @@ static Intrinsic::ID
getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
}
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveMaxIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::spv_wave_reduce_umax;
+return llvm::Intrinsic::spv_wave_reduce_max;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_reduce_umax;
+return llvm::Intrinsic::dx_wave_reduce_max;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveMax"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19532,6 +19551,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
/*AssumeConvergent=*/true),
ArrayRef{OpExpr}, "hlsl.wave.active.sum");
}
+ case Builtin::BI__builtin_hlsl_wave_active_max: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+OpExpr->getType(), ArrayRef{OpExpr->getType()}, false);
+Intrinsic::ID IID = getWaveActiveMaxIntrinsic(
+getTarget().getTriple().getArch(), CGM.getHLSLRuntime(),
+E->getArg(0)->getType());
+
+// Get overloaded name
+std::string Name =
+Intrinsic::getName(IID, ArrayRef{OpExpr->getType()}, &CGM.getModule());
+return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
+ /*Local=*/false,
+
/*AssumeConvergent=*/true),
+ ArrayRef{OpExpr}, "hlsl.wave.active.max");
+ }
case Builtin::BI__builtin_hlsl_wave_get_lane_index: {
// We don't define a SPIR-V intrinsic, instead it is a SPIR-V built-in
// defined in SPIRVBuiltins.td. So instead we manually get the matching
name
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 5001883003ee2d..fd7435caeb000e 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2183,6 +2183,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
TheCall->setType(ArgTyA);
break;
}
+ case Builtin::BI__builtin_hlsl_wave_active_max:
case Builtin::BI__builtin_h
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveMax` intrinsic (PR #123428)
@@ -0,0 +1,55 @@
+; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv-vulkan-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-vulkan-unknown %s -o -
-filetype=obj | spirv-val %}
+
+; Test lowering to spir-v backend for various types and scalar/vector
+
adam-yang wrote:
Done.
https://github.com/llvm/llvm-project/pull/123428
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveMax` intrinsic (PR #123428)
@@ -2132,6 +2135,33 @@ bool SPIRVInstructionSelector::selectWaveActiveCountBits(
return Result;
}
+bool SPIRVInstructionSelector::selectWaveReduceMax(Register ResVReg,
+ const SPIRVType *ResType,
+ MachineInstr &I,
+ bool IsUnsigned) const {
+ assert(I.getNumOperands() == 3);
+ assert(I.getOperand(2).isReg());
+ MachineBasicBlock &BB = *I.getParent();
+ Register InputRegister = I.getOperand(2).getReg();
+ SPIRVType *InputType = GR.getSPIRVTypeForVReg(InputRegister);
+
+ if (!InputType)
+report_fatal_error("Input Type could not be determined.");
+
+ SPIRVType *IntTy = GR.getOrCreateSPIRVIntegerType(32, I, TII);
+ // Retreive the operation to use based on input type
+ bool IsFloatTy = GR.isScalarOrVectorOfType(InputRegister,
SPIRV::OpTypeFloat);
+ auto IntegerOpcodeType =
+ IsUnsigned ? SPIRV::OpGroupNonUniformUMax : SPIRV::OpGroupNonUniformSMax;
+ auto Opcode = IsFloatTy ? SPIRV::OpGroupNonUniformFMax : IntegerOpcodeType;
+ return BuildMI(BB, I, I.getDebugLoc(), TII.get(Opcode))
+ .addDef(ResVReg)
+ .addUse(GR.getSPIRVTypeID(ResType))
+ .addUse(GR.getOrCreateConstInt(SPIRV::Scope::Subgroup, I, IntTy, TII))
+ .addImm(SPIRV::GroupOperation::Reduce)
+ .addUse(I.getOperand(2).getReg());
adam-yang wrote:
Added.
https://github.com/llvm/llvm-project/pull/123428
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveMax` intrinsic (PR #123428)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/123428
>From 9a0cf138d99ebc9ae18db054b4d8eaa34e8174a8 Mon Sep 17 00:00:00 2001
From: Adam Yang <31109344+adam-y...@users.noreply.github.com>
Date: Fri, 17 Jan 2025 16:14:44 -0800
Subject: [PATCH 1/4] Added WaveActiveMax
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 36 +
clang/lib/Sema/SemaHLSL.cpp | 1 +
.../CodeGenHLSL/builtins/WaveActiveMax.hlsl | 46 ++
.../BuiltIns/WaveActiveMax-errors.hlsl| 29
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 2 +
llvm/lib/Target/DirectX/DXIL.td | 6 +
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 33
llvm/test/CodeGen/DirectX/WaveActiveMax.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveMax.ll| 55 +++
12 files changed, 361 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveMax.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveMax-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveMax.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveMax.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index bbf4886b5cf053..a7f17ec4c8ef8f 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4795,6 +4795,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveMax : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_max"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_active_sum"];
let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index b80833fd91884d..57d8eec24a5a0b 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19203,6 +19203,25 @@ static Intrinsic::ID
getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
}
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveMaxIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::spv_wave_reduce_umax;
+return llvm::Intrinsic::spv_wave_reduce_max;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_reduce_umax;
+return llvm::Intrinsic::dx_wave_reduce_max;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveMax"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19532,6 +19551,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
/*AssumeConvergent=*/true),
ArrayRef{OpExpr}, "hlsl.wave.active.sum");
}
+ case Builtin::BI__builtin_hlsl_wave_active_max: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+OpExpr->getType(), ArrayRef{OpExpr->getType()}, false);
+Intrinsic::ID IID = getWaveActiveMaxIntrinsic(
+getTarget().getTriple().getArch(), CGM.getHLSLRuntime(),
+E->getArg(0)->getType());
+
+// Get overloaded name
+std::string Name =
+Intrinsic::getName(IID, ArrayRef{OpExpr->getType()}, &CGM.getModule());
+return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
+ /*Local=*/false,
+
/*AssumeConvergent=*/true),
+ ArrayRef{OpExpr}, "hlsl.wave.active.max");
+ }
case Builtin::BI__builtin_hlsl_wave_get_lane_index: {
// We don't define a SPIR-V intrinsic, instead it is a SPIR-V built-in
// defined in SPIRVBuiltins.td. So instead we manually get the matching
name
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 5001883003ee2d..fd7435caeb000e 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2183,6 +2183,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
TheCall->setType(ArgTyA);
break;
}
+ case Builtin::BI__builtin_hlsl_wave_active_max:
case Builtin::BI__builtin_h
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveMax` intrinsic (PR #123428)
@@ -0,0 +1,143 @@
+; RUN: opt -S -scalarizer -dxil-op-lower
-mtriple=dxil-pc-shadermodel6.3-library < %s | FileCheck %s
+
+; Test that for scalar values, WaveActiveMax maps down to the DirectX op
+
+define noundef half @wave_active_max_half(half noundef %expr) {
+entry:
+; CHECK: call half @dx.op.waveActiveOp.f16(i32 119, half %expr, i8 3, i8 0)
adam-yang wrote:
Right now waveActiveOp is marked with `memory(none)`. With dxcompiler, there's
no memory related attribute. Maybe we should update the DxilOP attribute to
match that.
https://github.com/llvm/llvm-project/pull/123428
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveMax` intrinsic (PR #123428)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/123428
>From 055412fd8d01d796ca070fafeca4375fd442986a Mon Sep 17 00:00:00 2001
From: Adam Yang <31109344+adam-y...@users.noreply.github.com>
Date: Fri, 17 Jan 2025 16:14:44 -0800
Subject: [PATCH 1/5] Added WaveActiveMax
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 36 +
clang/lib/Sema/SemaHLSL.cpp | 1 +
.../CodeGenHLSL/builtins/WaveActiveMax.hlsl | 46 ++
.../BuiltIns/WaveActiveMax-errors.hlsl| 29
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 2 +
llvm/lib/Target/DirectX/DXIL.td | 10 ++
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 33
llvm/test/CodeGen/DirectX/WaveActiveMax.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveMax.ll| 55 +++
12 files changed, 365 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveMax.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveMax-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveMax.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveMax.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 60c360d4a9e075..29939242596ba4 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4795,6 +4795,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveMax : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_max"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_active_sum"];
let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 26bdc5e36e..94c5a2fa4fda40 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19098,6 +19098,25 @@ static Intrinsic::ID
getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
}
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveMaxIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::spv_wave_reduce_umax;
+return llvm::Intrinsic::spv_wave_reduce_max;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_reduce_umax;
+return llvm::Intrinsic::dx_wave_reduce_max;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveMax"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19427,6 +19446,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
/*AssumeConvergent=*/true),
ArrayRef{OpExpr}, "hlsl.wave.active.sum");
}
+ case Builtin::BI__builtin_hlsl_wave_active_max: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+OpExpr->getType(), ArrayRef{OpExpr->getType()}, false);
+Intrinsic::ID IID = getWaveActiveMaxIntrinsic(
+getTarget().getTriple().getArch(), CGM.getHLSLRuntime(),
+E->getArg(0)->getType());
+
+// Get overloaded name
+std::string Name =
+Intrinsic::getName(IID, ArrayRef{OpExpr->getType()}, &CGM.getModule());
+return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
+ /*Local=*/false,
+
/*AssumeConvergent=*/true),
+ ArrayRef{OpExpr}, "hlsl.wave.active.max");
+ }
case Builtin::BI__builtin_hlsl_wave_get_lane_index: {
// We don't define a SPIR-V intrinsic, instead it is a SPIR-V built-in
// defined in SPIRVBuiltins.td. So instead we manually get the matching
name
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index f26469e6a2f1d7..5c58ac1f9167ee 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2422,6 +2422,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
TheCall->setType(ArgTyA);
break;
}
+ case Builtin::BI__builtin_hlsl_wave_active_max:
case Builtin::BI__builtin_
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveMax` intrinsic (PR #123428)
https://github.com/adam-yang created
https://github.com/llvm/llvm-project/pull/123428
```- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passespendent pr merges.
```
Resolves #99170
>From 9a0cf138d99ebc9ae18db054b4d8eaa34e8174a8 Mon Sep 17 00:00:00 2001
From: Adam Yang <31109344+adam-y...@users.noreply.github.com>
Date: Fri, 17 Jan 2025 16:14:44 -0800
Subject: [PATCH] Added WaveActiveMax
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 36 +
clang/lib/Sema/SemaHLSL.cpp | 1 +
.../CodeGenHLSL/builtins/WaveActiveMax.hlsl | 46 ++
.../BuiltIns/WaveActiveMax-errors.hlsl| 29
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 2 +
llvm/lib/Target/DirectX/DXIL.td | 6 +
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 33
llvm/test/CodeGen/DirectX/WaveActiveMax.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveMax.ll| 55 +++
12 files changed, 361 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveMax.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveMax-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveMax.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveMax.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index bbf4886b5cf053..a7f17ec4c8ef8f 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4795,6 +4795,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveMax : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_max"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_active_sum"];
let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index b80833fd91884d..57d8eec24a5a0b 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19203,6 +19203,25 @@ static Intrinsic::ID
getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
}
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveMaxIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::spv_wave_reduce_umax;
+return llvm::Intrinsic::spv_wave_reduce_max;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_reduce_umax;
+return llvm::Intrinsic::dx_wave_reduce_max;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveMax"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19532,6 +19551,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
/*AssumeConvergent=*/true),
ArrayRef{OpExpr}, "hlsl.wave.active.sum");
}
+ case Builtin::BI__builtin_hlsl_wave_active_max: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+OpExpr->getType(), ArrayRef{OpExpr->getType()}, false);
+Intrinsic::ID IID = getWaveActiveMaxIntrinsic(
+getTarget().getTriple().getArch(), CGM.getHLSLRuntime(),
+E->getArg(0)->getType());
+
+// Get overloaded name
+std::string Name =
+Intrinsic::getName(IID, ArrayRef{OpExpr->getType()}, &CGM.getModule());
+return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
+ /*Local=*/false,
+
/*AssumeConvergent=*/true),
+ ArrayRef{OpExpr}, "hlsl.wave.active.max");
+ }
case Builtin::BI__builtin_hlsl_wave_get_lane_index: {
// We don't define a SPIR-V intrinsic, ins
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveMax` intrinsic (PR #123428)
@@ -19203,6 +19203,25 @@ static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch, } } +// Return wave active sum that corresponds to the QT scalar type +static Intrinsic::ID getWaveActiveMaxIntrinsic(llvm::Triple::ArchType Arch, adam-yang wrote: I gave it a bit of try, it's kind of just shuffling things around and not really making things any easier to read. I think it okay as is. https://github.com/llvm/llvm-project/pull/123428 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveMax` intrinsic (PR #123428)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/123428
>From cedd93faaa75c343fda7a21ecf813249cf08609a Mon Sep 17 00:00:00 2001
From: Adam Yang <31109344+adam-y...@users.noreply.github.com>
Date: Fri, 17 Jan 2025 16:14:44 -0800
Subject: [PATCH 1/6] Added WaveActiveMax
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 36 +
clang/lib/Sema/SemaHLSL.cpp | 1 +
.../CodeGenHLSL/builtins/WaveActiveMax.hlsl | 46 ++
.../BuiltIns/WaveActiveMax-errors.hlsl| 29
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 2 +
llvm/lib/Target/DirectX/DXIL.td | 10 ++
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 33
llvm/test/CodeGen/DirectX/WaveActiveMax.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveMax.ll| 55 +++
12 files changed, 365 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveMax.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveMax-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveMax.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveMax.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 60c360d4a9e075..29939242596ba4 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4795,6 +4795,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveMax : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_max"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_active_sum"];
let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 9ae133421d8dac..27dfab11c5531c 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19295,6 +19295,25 @@ static Intrinsic::ID
getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
}
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveMaxIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::spv_wave_reduce_umax;
+return llvm::Intrinsic::spv_wave_reduce_max;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_reduce_umax;
+return llvm::Intrinsic::dx_wave_reduce_max;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveMax"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19624,6 +19643,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
/*AssumeConvergent=*/true),
ArrayRef{OpExpr}, "hlsl.wave.active.sum");
}
+ case Builtin::BI__builtin_hlsl_wave_active_max: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+OpExpr->getType(), ArrayRef{OpExpr->getType()}, false);
+Intrinsic::ID IID = getWaveActiveMaxIntrinsic(
+getTarget().getTriple().getArch(), CGM.getHLSLRuntime(),
+E->getArg(0)->getType());
+
+// Get overloaded name
+std::string Name =
+Intrinsic::getName(IID, ArrayRef{OpExpr->getType()}, &CGM.getModule());
+return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
+ /*Local=*/false,
+
/*AssumeConvergent=*/true),
+ ArrayRef{OpExpr}, "hlsl.wave.active.max");
+ }
case Builtin::BI__builtin_hlsl_wave_get_lane_index: {
// We don't define a SPIR-V intrinsic, instead it is a SPIR-V built-in
// defined in SPIRVBuiltins.td. So instead we manually get the matching
name
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index aa99b44958eafd..d748c10455289b 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2430,6 +2430,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
TheCall->setType(ArgTyA);
break;
}
+ case Builtin::BI__builtin_hlsl_wave_active_max:
case Builtin::BI__builtin_
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveMax` intrinsic (PR #123428)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/123428
>From 9a0cf138d99ebc9ae18db054b4d8eaa34e8174a8 Mon Sep 17 00:00:00 2001
From: Adam Yang <31109344+adam-y...@users.noreply.github.com>
Date: Fri, 17 Jan 2025 16:14:44 -0800
Subject: [PATCH 1/2] Added WaveActiveMax
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 36 +
clang/lib/Sema/SemaHLSL.cpp | 1 +
.../CodeGenHLSL/builtins/WaveActiveMax.hlsl | 46 ++
.../BuiltIns/WaveActiveMax-errors.hlsl| 29
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 2 +
llvm/lib/Target/DirectX/DXIL.td | 6 +
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 33
llvm/test/CodeGen/DirectX/WaveActiveMax.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveMax.ll| 55 +++
12 files changed, 361 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveMax.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveMax-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveMax.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveMax.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index bbf4886b5cf053..a7f17ec4c8ef8f 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4795,6 +4795,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveMax : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_max"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_active_sum"];
let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index b80833fd91884d..57d8eec24a5a0b 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19203,6 +19203,25 @@ static Intrinsic::ID
getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
}
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveMaxIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::spv_wave_reduce_umax;
+return llvm::Intrinsic::spv_wave_reduce_max;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_reduce_umax;
+return llvm::Intrinsic::dx_wave_reduce_max;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveMax"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19532,6 +19551,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
/*AssumeConvergent=*/true),
ArrayRef{OpExpr}, "hlsl.wave.active.sum");
}
+ case Builtin::BI__builtin_hlsl_wave_active_max: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+OpExpr->getType(), ArrayRef{OpExpr->getType()}, false);
+Intrinsic::ID IID = getWaveActiveMaxIntrinsic(
+getTarget().getTriple().getArch(), CGM.getHLSLRuntime(),
+E->getArg(0)->getType());
+
+// Get overloaded name
+std::string Name =
+Intrinsic::getName(IID, ArrayRef{OpExpr->getType()}, &CGM.getModule());
+return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
+ /*Local=*/false,
+
/*AssumeConvergent=*/true),
+ ArrayRef{OpExpr}, "hlsl.wave.active.max");
+ }
case Builtin::BI__builtin_hlsl_wave_get_lane_index: {
// We don't define a SPIR-V intrinsic, instead it is a SPIR-V built-in
// defined in SPIRVBuiltins.td. So instead we manually get the matching
name
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 5001883003ee2d..fd7435caeb000e 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2183,6 +2183,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
TheCall->setType(ArgTyA);
break;
}
+ case Builtin::BI__builtin_hlsl_wave_active_max:
case Builtin::BI__builtin_h
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From 9498523511a53c43b6282500fed2659e12ddd752 Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/7] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index ea22690ce4f5cc..ec8e2420469011 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4795,6 +4795,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index 8be4f946dce1cc..56d882efdc9ace 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9299,6 +9299,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 1b25d365932c30..503fcc07eb33c1 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19186,6 +19186,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19491,6 +19508,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang closed https://github.com/llvm/llvm-project/pull/118580 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic (PR #118580)
https://github.com/adam-yang updated
https://github.com/llvm/llvm-project/pull/118580
>From 9a1ad541ec6e2fbd72bd88727a5910da39a5b840 Mon Sep 17 00:00:00 2001
From: Finn Plummer
Date: Fri, 18 Oct 2024 10:49:18 -0700
Subject: [PATCH 1/5] [HLSL][SPIRV][DXIL] Implement `WaveActiveSum` intrinsic
- add clang builtin to Builtins.td
- link builtin in hlsl_intrinsics
- add codegen for spirv intrinsic and two directx intrinsics to retain
signedness information of the operands in CGBuiltin.cpp
- add semantic analysis in SemaHLSL.cpp
- add lowering of spirv intrinsic to spirv backend in
SPIRVInstructionSelector.cpp
- add lowering of directx intrinsics to WaveActiveOp dxil op in
DXIL.td
- add test cases to illustrate passes
---
clang/include/clang/Basic/Builtins.td | 6 +
.../clang/Basic/DiagnosticSemaKinds.td| 3 +
clang/lib/CodeGen/CGBuiltin.cpp | 34 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 99
clang/lib/Sema/SemaHLSL.cpp | 31
.../CodeGenHLSL/builtins/WaveActiveSum.hlsl | 45 ++
.../BuiltIns/WaveActiveSum-errors.hlsl| 28
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 26
.../DirectX/DirectXTargetTransformInfo.cpp| 2 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 30
llvm/test/CodeGen/DirectX/WaveActiveSum.ll| 143 ++
.../SPIRV/hlsl-intrinsics/WaveActiveSum.ll| 41 +
14 files changed, 491 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveActiveSum-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/WaveActiveSum.ll
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveActiveSum.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 468c16050e2bf0..12b96672ce466f 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4793,6 +4793,12 @@ def HLSLWaveActiveCountBits : LangBuiltin<"HLSL_LANG"> {
let Prototype = "unsigned int(bool)";
}
+def HLSLWaveActiveSum : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_wave_active_sum"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void (...)";
+}
+
def HLSLWaveGetLaneIndex : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_wave_get_lane_index"];
let Attributes = [NoThrow, Const];
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td
b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index ab2d6237c1cab8..d6295e31d2548e 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9287,6 +9287,9 @@ def err_typecheck_expect_scalar_or_vector : Error<
"a vector of such type is required">;
def err_typecheck_expect_any_scalar_or_vector : Error<
"invalid operand of type %0 where a scalar or vector is required">;
+def err_typecheck_expect_scalar_or_vector_not_type : Error<
+ "invalid operand of type %0 where %1 or "
+ "a vector of such type is not allowed">;
def err_typecheck_expect_flt_or_vector : Error<
"invalid operand of type %0 where floating, complex or "
"a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 573be932f8b1aa..2f754f595041ed 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19198,6 +19198,23 @@ static Intrinsic::ID
getFirstBitHighIntrinsic(CGHLSLRuntime &RT, QualType QT) {
return RT.getFirstBitUHighIntrinsic();
}
+// Return wave active sum that corresponds to the QT scalar type
+static Intrinsic::ID getWaveActiveSumIntrinsic(llvm::Triple::ArchType Arch,
+ CGHLSLRuntime &RT, QualType QT)
{
+ switch (Arch) {
+ case llvm::Triple::spirv:
+return llvm::Intrinsic::spv_wave_active_sum;
+ case llvm::Triple::dxil: {
+if (QT->isUnsignedIntegerType())
+ return llvm::Intrinsic::dx_wave_active_usum;
+return llvm::Intrinsic::dx_wave_active_sum;
+ }
+ default:
+llvm_unreachable("Intrinsic WaveActiveSum"
+ " not supported by target architecture");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E,
ReturnValueSlot ReturnValue) {
@@ -19517,6 +19534,23 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID),
ArrayRef{OpExpr});
}
+ case Builtin::BI__builtin_hlsl_wave_active_sum: {
+// Due to the use of variadic arguments, explicitly retreive argument
+Value *OpExpr = EmitScalarExpr(E->getArg(0));
+llvm::FunctionType *FT = llvm::FunctionType::get(
+
