[clang] Implement the `fmod` intrinsic (PR #130320)
@@ -22,56 +22,144 @@ // // RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ // RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \ -// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \ +// RUN: -emit-llvm -o - | FileCheck %s \ // RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=half kmpeng wrote: Code updated with suggested changes. https://github.com/llvm/llvm-project/pull/130320 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] Implement the `fmod` intrinsic (PR #130320)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/130320
>From bdb66b5a68090e304647ddbbeb403ac408fcea65 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 25 Feb 2025 14:50:09 -0800
Subject: [PATCH 1/8] start implementation
---
clang/lib/Headers/hlsl/hlsl_detail.h | 14 ++
1 file changed, 14 insertions(+)
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h
b/clang/lib/Headers/hlsl/hlsl_detail.h
index 39254a3cc3a0a..e7e910e30957e 100644
--- a/clang/lib/Headers/hlsl/hlsl_detail.h
+++ b/clang/lib/Headers/hlsl/hlsl_detail.h
@@ -97,6 +97,20 @@ constexpr vector reflect_vec_impl(vector I,
vector N) {
#endif
}
+template
+constexpr enable_if_t::value || is_same::value, T>
+fmod_vec_impl(vector X, vector Y) {
+#if !defined(__DirectX__)
+ return __builtin_elementwise_fmod(X, Y);
+#else
+ vector div = X / Y;
+ vector result =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div)) * Y;
+ vector condition = (div >= -div);
+ vector realResult = __builtin_hlsl_select(condition, result, -result);
+ return realResult;
+#endif
+}
+
} // namespace __detail
} // namespace hlsl
#endif //_HLSL_HLSL_DETAILS_H_
>From fe73beac17b99b0f7836ffdb7050e04595d96dba Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Wed, 26 Feb 2025 15:57:59 -0800
Subject: [PATCH 2/8] finished implementation, working on fmod.hlsl tests
---
clang/lib/Headers/hlsl/hlsl_detail.h | 14 +++---
clang/test/CodeGenHLSL/builtins/fmod.hlsl | 32 +++
2 files changed, 26 insertions(+), 20 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h
b/clang/lib/Headers/hlsl/hlsl_detail.h
index e7e910e30957e..bc2dd463e0404 100644
--- a/clang/lib/Headers/hlsl/hlsl_detail.h
+++ b/clang/lib/Headers/hlsl/hlsl_detail.h
@@ -97,6 +97,12 @@ constexpr vector reflect_vec_impl(vector I,
vector N) {
#endif
}
+template
+constexpr enable_if_t::value || is_same::value, T>
+fmod_impl(T X, T Y) {
+ return __builtin_elementwise_fmod(X, Y);
+}
+
template
constexpr enable_if_t::value || is_same::value, T>
fmod_vec_impl(vector X, vector Y) {
@@ -104,10 +110,10 @@ fmod_vec_impl(vector X, vector Y) {
return __builtin_elementwise_fmod(X, Y);
#else
vector div = X / Y;
- vector result =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div)) * Y;
- vector condition = (div >= -div);
- vector realResult = __builtin_hlsl_select(condition, result, -result);
- return realResult;
+ vector ge = div >= -div;
+ vector frc =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div));
+ vector realFrc = __builtin_hlsl_select(ge, frc, -frc);
+ return realFrc * Y;
#endif
}
diff --git a/clang/test/CodeGenHLSL/builtins/fmod.hlsl
b/clang/test/CodeGenHLSL/builtins/fmod.hlsl
index b62967114d456..22376638bd093 100644
--- a/clang/test/CodeGenHLSL/builtins/fmod.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/fmod.hlsl
@@ -36,42 +36,42 @@
// CHECK: define [[FNATTRS]] [[TYPE]] @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn [[TYPE]]
-// CHECK: ret [[TYPE]] %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]]
@{{.*}}([[TYPE]] noundef nofpclass(nan inf) %{{.*}}, [[TYPE]] noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret [[TYPE]] %call
half test_fmod_half(half p0, half p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <2 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
-// CHECK: ret <2 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<2
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <2 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret <2 x [[TYPE]]> %splat.splat
half2 test_fmod_half2(half2 p0, half2 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <3 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <3 x [[TYPE]]>
-// CHECK: ret <3 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<3
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <3 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}} #{{.*}}
+// CHECK: ret <3 x [[TYPE]]> %splat.splat
half3 test_fmod_half3(half3 p0, half3 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <4 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <4 x [[TYPE]]>
-// CHECK: ret <4 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<4
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <4 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret <4 x [[TYPE]]> %splat.splat
half4 test_fmod_half4(half4 p0, half4 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] float @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn float
-// CHECK: ret float %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] float @{{.*}}(float
noundef nofpclass(nan inf) %{{.*}}, float noundef nofpclass(nan inf) %{{.*}})
#{{.*}}
+// CHECK: re
[clang] Implement the `fmod` intrinsic (PR #130320)
@@ -22,56 +22,144 @@ // // RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ // RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \ -// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \ +// RUN: -emit-llvm -o - | FileCheck %s \ // RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=half // // -- No Native Half support test --- // // RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ -// RUN: spirv-unknown-vulkan-compute %s -emit-llvm -disable-llvm-passes \ +// RUN: spirv-unknown-vulkan-compute %s -emit-llvm \ // RUN: -o - | FileCheck %s \ // RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=float +// DXCHECK: define [[FNATTRS]] [[TYPE]] @ kmpeng wrote: Code updated with suggested changes. https://github.com/llvm/llvm-project/pull/130320 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] Implement the `fmod` intrinsic (PR #130320)
https://github.com/kmpeng created
https://github.com/llvm/llvm-project/pull/130320
Replaced the current `fmod` definition with a templatized version, implemented
`fmod` algorithm for DirectX targets that matches the DXC implementation, added
corresponding tests in `clang/test/CodeGenHLSL/builtins/fmod.hlsl` and
`clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl`.
>From bdb66b5a68090e304647ddbbeb403ac408fcea65 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 25 Feb 2025 14:50:09 -0800
Subject: [PATCH 1/7] start implementation
---
clang/lib/Headers/hlsl/hlsl_detail.h | 14 ++
1 file changed, 14 insertions(+)
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h
b/clang/lib/Headers/hlsl/hlsl_detail.h
index 39254a3cc3a0a..e7e910e30957e 100644
--- a/clang/lib/Headers/hlsl/hlsl_detail.h
+++ b/clang/lib/Headers/hlsl/hlsl_detail.h
@@ -97,6 +97,20 @@ constexpr vector reflect_vec_impl(vector I,
vector N) {
#endif
}
+template
+constexpr enable_if_t::value || is_same::value, T>
+fmod_vec_impl(vector X, vector Y) {
+#if !defined(__DirectX__)
+ return __builtin_elementwise_fmod(X, Y);
+#else
+ vector div = X / Y;
+ vector result =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div)) * Y;
+ vector condition = (div >= -div);
+ vector realResult = __builtin_hlsl_select(condition, result, -result);
+ return realResult;
+#endif
+}
+
} // namespace __detail
} // namespace hlsl
#endif //_HLSL_HLSL_DETAILS_H_
>From fe73beac17b99b0f7836ffdb7050e04595d96dba Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Wed, 26 Feb 2025 15:57:59 -0800
Subject: [PATCH 2/7] finished implementation, working on fmod.hlsl tests
---
clang/lib/Headers/hlsl/hlsl_detail.h | 14 +++---
clang/test/CodeGenHLSL/builtins/fmod.hlsl | 32 +++
2 files changed, 26 insertions(+), 20 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h
b/clang/lib/Headers/hlsl/hlsl_detail.h
index e7e910e30957e..bc2dd463e0404 100644
--- a/clang/lib/Headers/hlsl/hlsl_detail.h
+++ b/clang/lib/Headers/hlsl/hlsl_detail.h
@@ -97,6 +97,12 @@ constexpr vector reflect_vec_impl(vector I,
vector N) {
#endif
}
+template
+constexpr enable_if_t::value || is_same::value, T>
+fmod_impl(T X, T Y) {
+ return __builtin_elementwise_fmod(X, Y);
+}
+
template
constexpr enable_if_t::value || is_same::value, T>
fmod_vec_impl(vector X, vector Y) {
@@ -104,10 +110,10 @@ fmod_vec_impl(vector X, vector Y) {
return __builtin_elementwise_fmod(X, Y);
#else
vector div = X / Y;
- vector result =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div)) * Y;
- vector condition = (div >= -div);
- vector realResult = __builtin_hlsl_select(condition, result, -result);
- return realResult;
+ vector ge = div >= -div;
+ vector frc =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div));
+ vector realFrc = __builtin_hlsl_select(ge, frc, -frc);
+ return realFrc * Y;
#endif
}
diff --git a/clang/test/CodeGenHLSL/builtins/fmod.hlsl
b/clang/test/CodeGenHLSL/builtins/fmod.hlsl
index b62967114d456..22376638bd093 100644
--- a/clang/test/CodeGenHLSL/builtins/fmod.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/fmod.hlsl
@@ -36,42 +36,42 @@
// CHECK: define [[FNATTRS]] [[TYPE]] @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn [[TYPE]]
-// CHECK: ret [[TYPE]] %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]]
@{{.*}}([[TYPE]] noundef nofpclass(nan inf) %{{.*}}, [[TYPE]] noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret [[TYPE]] %call
half test_fmod_half(half p0, half p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <2 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
-// CHECK: ret <2 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<2
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <2 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret <2 x [[TYPE]]> %splat.splat
half2 test_fmod_half2(half2 p0, half2 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <3 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <3 x [[TYPE]]>
-// CHECK: ret <3 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<3
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <3 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}} #{{.*}}
+// CHECK: ret <3 x [[TYPE]]> %splat.splat
half3 test_fmod_half3(half3 p0, half3 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <4 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <4 x [[TYPE]]>
-// CHECK: ret <4 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<4
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <4 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret <4 x [[TYPE]]> %splat.splat
half4 test_fmod_half4(half4 p0, half4 p1) { return fmod(p0, p1); }
// CHECK: define [[FNA
[clang] Implement the `fmod` intrinsic (PR #130320)
@@ -22,56 +22,144 @@ // // RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ // RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \ -// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \ +// RUN: -emit-llvm -o - | FileCheck %s \ // RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=half // // -- No Native Half support test --- // // RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ -// RUN: spirv-unknown-vulkan-compute %s -emit-llvm -disable-llvm-passes \ +// RUN: spirv-unknown-vulkan-compute %s -emit-llvm \ // RUN: -o - | FileCheck %s \ // RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=float +// DXCHECK: define [[FNATTRS]] [[TYPE]] @ +// DXCHECK: %div1.i = fdiv reassoc nnan ninf nsz arcp afn [[TYPE]] kmpeng wrote: @Icohedron Would you recommend running the DirectX tests with -O1 then if we want to check the operands? Currently the dependent instructions are not necessarily receiving the same operands/outputs from the instructions they depend on (e.g. `%fneg.i` depends on `%div1.i`, but `%div1.i` becomes `%7` and then `%7` is passed to `%fneg.i` instead). Running with -O1 solves this, but it also optimizes away the first `fneg` instruction I check for. https://github.com/llvm/llvm-project/pull/130320 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] Implement the `fmod` intrinsic (PR #130320)
@@ -22,56 +22,136 @@
//
// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
// RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \
-// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \
-// RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=half
+// RUN: -emit-llvm -o - | FileCheck %s \
+// RUN: -DTYPE=half --check-prefixes=CHECK,SPVCHECK
//
// -- No Native Half support test ---
//
// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
-// RUN: spirv-unknown-vulkan-compute %s -emit-llvm -disable-llvm-passes \
+// RUN: spirv-unknown-vulkan-compute %s -emit-llvm \
// RUN: -o - | FileCheck %s \
-// RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=float
+// RUN: -DTYPE=float --check-prefixes=CHECK,SPVCHECK
-// CHECK: define [[FNATTRS]] [[TYPE]] @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn [[TYPE]]
-// CHECK: ret [[TYPE]] %fmod
+// CHECK-LABEL: test_fmod_half
+// DXCHECK: %div1.i = fdiv reassoc nnan ninf nsz arcp afn [[TYPE]] [[X:%.*]],
[[Y:%.*]]
+// DXCHECK: %fneg.i = fneg reassoc nnan ninf nsz arcp afn [[TYPE]]
[[DIV1_I:%.*]]
+// DXCHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn oge [[TYPE]]
[[DIV1_I_2:%.*]], %fneg.i
+// DXNATIVE_HALF: %elt.abs.i = call reassoc nnan ninf nsz arcp afn [[TYPE]]
@llvm.fabs.f16([[TYPE]] [[DIV1_I_3:%.*]])
+// DXNO_HALF: %elt.abs.i = call reassoc nnan ninf nsz arcp afn [[TYPE]]
@llvm.fabs.f32([[TYPE]] [[DIV1_I_3:%.*]])
+// DXNATIVE_HALF: %hlsl.frac.i = call reassoc nnan ninf nsz arcp afn [[TYPE]]
@llvm.dx.frac.f16([[TYPE]] %elt.abs.i)
+// DXNO_HALF: %hlsl.frac.i = call reassoc nnan ninf nsz arcp afn [[TYPE]]
@llvm.dx.frac.f32([[TYPE]] %elt.abs.i)
+// DXCHECK: %fneg2.i = fneg reassoc nnan ninf nsz arcp afn [[TYPE]]
[[HLSL_FRAC_I:%.*]]
+// DXCHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1
[[CMP_I:%.*]], [[TYPE]] [[HLSL_FRAC_I_2:%.*]], [[TYPE]] %fneg2.i
+// DXCHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn [[TYPE]]
%hlsl.select.i, [[Y_2:%.*]]
+// DXCHECK: ret [[TYPE]] %mul.i
+// SPVCHECK: %fmod.i = frem reassoc nnan ninf nsz arcp afn [[TYPE]] [[X:%.*]],
[[Y:%.*]]
+// SPVCHECK: ret [[TYPE]] %fmod.i
half test_fmod_half(half p0, half p1) { return fmod(p0, p1); }
-// CHECK: define [[FNATTRS]] <2 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
-// CHECK: ret <2 x [[TYPE]]> %fmod
+// CHECK-LABEL: test_fmod_half2
+// DXCHECK: %div1.i = fdiv reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
[[X:%.*]], [[Y:%.*]]
+// DXCHECK: %fneg.i = fneg reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
[[DIV1_I:%.*]]
+// DXCHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn oge <2 x [[TYPE]]>
[[DIV1_I_2:%.*]], %fneg.i
+// DXNATIVE_HALF: %elt.abs.i = call reassoc nnan ninf nsz arcp afn <2 x
[[TYPE]]> @llvm.fabs.v2f16(<2 x [[TYPE]]> [[DIV1_I_3:%.*]])
+// DXNO_HALF: %elt.abs.i = call reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
@llvm.fabs.v2f32(<2 x [[TYPE]]> [[DIV1_I_3:%.*]])
+// DXNATIVE_HALF: %hlsl.frac.i = call reassoc nnan ninf nsz arcp afn <2 x
[[TYPE]]> @llvm.dx.frac.v2f16(<2 x [[TYPE]]> %elt.abs.i)
+// DXNO_HALF: %hlsl.frac.i = call reassoc nnan ninf nsz arcp afn <2 x
[[TYPE]]> @llvm.dx.frac.v2f32(<2 x [[TYPE]]> %elt.abs.i)
+// DXCHECK: %fneg2.i = fneg reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
[[HLSL_FRAC_I:%.*]]
+// DXCHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn <2 x i1>
[[CMP_I:%.*]], <2 x [[TYPE]]> [[HLSL_FRAC_I_2:%.*]], <2 x [[TYPE]]> %fneg2.i
+// DXCHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
%hlsl.select.i, [[Y_2:%.*]]
+// DXCHECK: ret <2 x [[TYPE]]> %mul.i
+// SPVCHECK: %fmod.i = frem reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
[[X:%.*]], [[Y:%.*]]
+// SPVCHECK: ret <2 x [[TYPE]]> %fmod.i
half2 test_fmod_half2(half2 p0, half2 p1) { return fmod(p0, p1); }
-// CHECK: define [[FNATTRS]] <3 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <3 x [[TYPE]]>
-// CHECK: ret <3 x [[TYPE]]> %fmod
+// CHECK-LABEL: test_fmod_half3
+// DXCHECK: %div1.i = fdiv reassoc nnan ninf nsz arcp afn <3 x [[TYPE]]>
[[X:%.*]], [[Y:%.*]]
+// DXCHECK: %fneg.i = fneg reassoc nnan ninf nsz arcp afn <3 x [[TYPE]]>
[[DIV1_I:%.*]]
kmpeng wrote:
Do you think I need to add DXCHECKs for the other operators that are like the
`DIV1_I` ones? Adding it for `HLSL_FRAC_I` wouldn't be much of a problem, but
`CMP_I` would require a lot of lines.
```llvm
%cmp.i = fcmp reassoc nnan ninf nsz arcp afn oge <2 x half> %6, %fneg.i
%sext.i = sext <2 x i1> %cmp.i to <2 x i16>
%insertvec.i = shufflevector <2 x i1> %cmp.i, <2 x i1> poison, <8 x i32>
%8 = bitcast <8 x i1> %insertvec.i to i8
...
%load_bits.i = load i8, ptr %ge.i, align 1
%10 = bitcast i8 %load_bits.i to <8 x i1>
%extractvec.i = shufflevector <8 x i1> %10, <8 x i1> poison, <2 x i32>
...
%hlsl.select.i = select reassoc nnan ninf nsz arcp afn <2 x i1>
%extractvec.i, <2 x h
[clang] Implement the `fmod` intrinsic (PR #130320)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/130320
>From bdb66b5a68090e304647ddbbeb403ac408fcea65 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 25 Feb 2025 14:50:09 -0800
Subject: [PATCH 1/9] start implementation
---
clang/lib/Headers/hlsl/hlsl_detail.h | 14 ++
1 file changed, 14 insertions(+)
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h
b/clang/lib/Headers/hlsl/hlsl_detail.h
index 39254a3cc3a0a..e7e910e30957e 100644
--- a/clang/lib/Headers/hlsl/hlsl_detail.h
+++ b/clang/lib/Headers/hlsl/hlsl_detail.h
@@ -97,6 +97,20 @@ constexpr vector reflect_vec_impl(vector I,
vector N) {
#endif
}
+template
+constexpr enable_if_t::value || is_same::value, T>
+fmod_vec_impl(vector X, vector Y) {
+#if !defined(__DirectX__)
+ return __builtin_elementwise_fmod(X, Y);
+#else
+ vector div = X / Y;
+ vector result =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div)) * Y;
+ vector condition = (div >= -div);
+ vector realResult = __builtin_hlsl_select(condition, result, -result);
+ return realResult;
+#endif
+}
+
} // namespace __detail
} // namespace hlsl
#endif //_HLSL_HLSL_DETAILS_H_
>From fe73beac17b99b0f7836ffdb7050e04595d96dba Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Wed, 26 Feb 2025 15:57:59 -0800
Subject: [PATCH 2/9] finished implementation, working on fmod.hlsl tests
---
clang/lib/Headers/hlsl/hlsl_detail.h | 14 +++---
clang/test/CodeGenHLSL/builtins/fmod.hlsl | 32 +++
2 files changed, 26 insertions(+), 20 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h
b/clang/lib/Headers/hlsl/hlsl_detail.h
index e7e910e30957e..bc2dd463e0404 100644
--- a/clang/lib/Headers/hlsl/hlsl_detail.h
+++ b/clang/lib/Headers/hlsl/hlsl_detail.h
@@ -97,6 +97,12 @@ constexpr vector reflect_vec_impl(vector I,
vector N) {
#endif
}
+template
+constexpr enable_if_t::value || is_same::value, T>
+fmod_impl(T X, T Y) {
+ return __builtin_elementwise_fmod(X, Y);
+}
+
template
constexpr enable_if_t::value || is_same::value, T>
fmod_vec_impl(vector X, vector Y) {
@@ -104,10 +110,10 @@ fmod_vec_impl(vector X, vector Y) {
return __builtin_elementwise_fmod(X, Y);
#else
vector div = X / Y;
- vector result =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div)) * Y;
- vector condition = (div >= -div);
- vector realResult = __builtin_hlsl_select(condition, result, -result);
- return realResult;
+ vector ge = div >= -div;
+ vector frc =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div));
+ vector realFrc = __builtin_hlsl_select(ge, frc, -frc);
+ return realFrc * Y;
#endif
}
diff --git a/clang/test/CodeGenHLSL/builtins/fmod.hlsl
b/clang/test/CodeGenHLSL/builtins/fmod.hlsl
index b62967114d456..22376638bd093 100644
--- a/clang/test/CodeGenHLSL/builtins/fmod.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/fmod.hlsl
@@ -36,42 +36,42 @@
// CHECK: define [[FNATTRS]] [[TYPE]] @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn [[TYPE]]
-// CHECK: ret [[TYPE]] %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]]
@{{.*}}([[TYPE]] noundef nofpclass(nan inf) %{{.*}}, [[TYPE]] noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret [[TYPE]] %call
half test_fmod_half(half p0, half p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <2 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
-// CHECK: ret <2 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<2
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <2 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret <2 x [[TYPE]]> %splat.splat
half2 test_fmod_half2(half2 p0, half2 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <3 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <3 x [[TYPE]]>
-// CHECK: ret <3 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<3
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <3 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}} #{{.*}}
+// CHECK: ret <3 x [[TYPE]]> %splat.splat
half3 test_fmod_half3(half3 p0, half3 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <4 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <4 x [[TYPE]]>
-// CHECK: ret <4 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<4
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <4 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret <4 x [[TYPE]]> %splat.splat
half4 test_fmod_half4(half4 p0, half4 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] float @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn float
-// CHECK: ret float %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] float @{{.*}}(float
noundef nofpclass(nan inf) %{{.*}}, float noundef nofpclass(nan inf) %{{.*}})
#{{.*}}
+// CHECK: re
[clang] Implement the `fmod` intrinsic (PR #130320)
@@ -22,56 +22,144 @@ // // RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ // RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \ -// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \ +// RUN: -emit-llvm -o - | FileCheck %s \ // RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=half // // -- No Native Half support test --- // // RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ -// RUN: spirv-unknown-vulkan-compute %s -emit-llvm -disable-llvm-passes \ +// RUN: spirv-unknown-vulkan-compute %s -emit-llvm \ // RUN: -o - | FileCheck %s \ // RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=float +// DXCHECK: define [[FNATTRS]] [[TYPE]] @ +// DXCHECK: %div1.i = fdiv reassoc nnan ninf nsz arcp afn [[TYPE]] kmpeng wrote: Code updated with suggested changes. https://github.com/llvm/llvm-project/pull/130320 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] Implement the `fmod` intrinsic (PR #130320)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/130320 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] Implement the `fmod` intrinsic (PR #130320)
@@ -22,56 +22,144 @@ // // RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ // RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \ -// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \ +// RUN: -emit-llvm -o - | FileCheck %s \ // RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=half // // -- No Native Half support test --- // // RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ -// RUN: spirv-unknown-vulkan-compute %s -emit-llvm -disable-llvm-passes \ +// RUN: spirv-unknown-vulkan-compute %s -emit-llvm \ // RUN: -o - | FileCheck %s \ // RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=float +// DXCHECK: define [[FNATTRS]] [[TYPE]] @ +// DXCHECK: %div1.i = fdiv reassoc nnan ninf nsz arcp afn [[TYPE]] kmpeng wrote: `%div1.i` is used as an operand in a lot following instructions and a different register is used for each instruction (`%7`, `%8`, `%9`). Would it be better practice to replace all of these register names with `[[DIV1_I:%.*]]` so it's clear they're all `%div1.i`, or would it be better to replace the first one with `[[DIV1_I:%.*]]`, the second with `[[DIV1_I_2:%.*]]`, etc.? https://github.com/llvm/llvm-project/pull/130320 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] Implement the `fmod` intrinsic (PR #130320)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/130320 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] Implement the `fmod` intrinsic (PR #130320)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/130320 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] Implement the `fmod` intrinsic (PR #130320)
@@ -22,56 +22,136 @@
//
// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
// RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \
-// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \
-// RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=half
+// RUN: -emit-llvm -o - | FileCheck %s \
+// RUN: -DTYPE=half --check-prefixes=CHECK,SPVCHECK
//
// -- No Native Half support test ---
//
// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
-// RUN: spirv-unknown-vulkan-compute %s -emit-llvm -disable-llvm-passes \
+// RUN: spirv-unknown-vulkan-compute %s -emit-llvm \
// RUN: -o - | FileCheck %s \
-// RUN: -DFNATTRS="spir_func noundef nofpclass(nan inf)" -DTYPE=float
+// RUN: -DTYPE=float --check-prefixes=CHECK,SPVCHECK
-// CHECK: define [[FNATTRS]] [[TYPE]] @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn [[TYPE]]
-// CHECK: ret [[TYPE]] %fmod
+// CHECK-LABEL: test_fmod_half
+// DXCHECK: %div1.i = fdiv reassoc nnan ninf nsz arcp afn [[TYPE]] [[X:%.*]],
[[Y:%.*]]
+// DXCHECK: %fneg.i = fneg reassoc nnan ninf nsz arcp afn [[TYPE]]
[[DIV1_I:%.*]]
+// DXCHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn oge [[TYPE]]
[[DIV1_I_2:%.*]], %fneg.i
+// DXNATIVE_HALF: %elt.abs.i = call reassoc nnan ninf nsz arcp afn [[TYPE]]
@llvm.fabs.f16([[TYPE]] [[DIV1_I_3:%.*]])
+// DXNO_HALF: %elt.abs.i = call reassoc nnan ninf nsz arcp afn [[TYPE]]
@llvm.fabs.f32([[TYPE]] [[DIV1_I_3:%.*]])
+// DXNATIVE_HALF: %hlsl.frac.i = call reassoc nnan ninf nsz arcp afn [[TYPE]]
@llvm.dx.frac.f16([[TYPE]] %elt.abs.i)
+// DXNO_HALF: %hlsl.frac.i = call reassoc nnan ninf nsz arcp afn [[TYPE]]
@llvm.dx.frac.f32([[TYPE]] %elt.abs.i)
+// DXCHECK: %fneg2.i = fneg reassoc nnan ninf nsz arcp afn [[TYPE]]
[[HLSL_FRAC_I:%.*]]
+// DXCHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1
[[CMP_I:%.*]], [[TYPE]] [[HLSL_FRAC_I_2:%.*]], [[TYPE]] %fneg2.i
+// DXCHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn [[TYPE]]
%hlsl.select.i, [[Y_2:%.*]]
+// DXCHECK: ret [[TYPE]] %mul.i
+// SPVCHECK: %fmod.i = frem reassoc nnan ninf nsz arcp afn [[TYPE]] [[X:%.*]],
[[Y:%.*]]
+// SPVCHECK: ret [[TYPE]] %fmod.i
half test_fmod_half(half p0, half p1) { return fmod(p0, p1); }
-// CHECK: define [[FNATTRS]] <2 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
-// CHECK: ret <2 x [[TYPE]]> %fmod
+// CHECK-LABEL: test_fmod_half2
+// DXCHECK: %div1.i = fdiv reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
[[X:%.*]], [[Y:%.*]]
+// DXCHECK: %fneg.i = fneg reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
[[DIV1_I:%.*]]
+// DXCHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn oge <2 x [[TYPE]]>
[[DIV1_I_2:%.*]], %fneg.i
+// DXNATIVE_HALF: %elt.abs.i = call reassoc nnan ninf nsz arcp afn <2 x
[[TYPE]]> @llvm.fabs.v2f16(<2 x [[TYPE]]> [[DIV1_I_3:%.*]])
+// DXNO_HALF: %elt.abs.i = call reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
@llvm.fabs.v2f32(<2 x [[TYPE]]> [[DIV1_I_3:%.*]])
+// DXNATIVE_HALF: %hlsl.frac.i = call reassoc nnan ninf nsz arcp afn <2 x
[[TYPE]]> @llvm.dx.frac.v2f16(<2 x [[TYPE]]> %elt.abs.i)
+// DXNO_HALF: %hlsl.frac.i = call reassoc nnan ninf nsz arcp afn <2 x
[[TYPE]]> @llvm.dx.frac.v2f32(<2 x [[TYPE]]> %elt.abs.i)
+// DXCHECK: %fneg2.i = fneg reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
[[HLSL_FRAC_I:%.*]]
+// DXCHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn <2 x i1>
[[CMP_I:%.*]], <2 x [[TYPE]]> [[HLSL_FRAC_I_2:%.*]], <2 x [[TYPE]]> %fneg2.i
+// DXCHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
%hlsl.select.i, [[Y_2:%.*]]
+// DXCHECK: ret <2 x [[TYPE]]> %mul.i
+// SPVCHECK: %fmod.i = frem reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
[[X:%.*]], [[Y:%.*]]
+// SPVCHECK: ret <2 x [[TYPE]]> %fmod.i
half2 test_fmod_half2(half2 p0, half2 p1) { return fmod(p0, p1); }
-// CHECK: define [[FNATTRS]] <3 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <3 x [[TYPE]]>
-// CHECK: ret <3 x [[TYPE]]> %fmod
+// CHECK-LABEL: test_fmod_half3
+// DXCHECK: %div1.i = fdiv reassoc nnan ninf nsz arcp afn <3 x [[TYPE]]>
[[X:%.*]], [[Y:%.*]]
+// DXCHECK: %fneg.i = fneg reassoc nnan ninf nsz arcp afn <3 x [[TYPE]]>
[[DIV1_I:%.*]]
kmpeng wrote:
Changes updated with Farzon's suggestions to include just the [relevant
instructions for
fmod](https://github.com/llvm/llvm-project/issues/99118#:~:text=%251%20%3D%20fdiv%20fast,5%2C%20%25p2)
and their exact matches with the exception of `-D`.
https://github.com/llvm/llvm-project/pull/130320
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] Implement the `fmod` intrinsic (PR #130320)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/130320
>From fd3db87977a52d7560d9eec15522c8afb85c9171 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 25 Feb 2025 14:50:09 -0800
Subject: [PATCH 01/10] start implementation
---
clang/lib/Headers/hlsl/hlsl_detail.h | 14 ++
1 file changed, 14 insertions(+)
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h
b/clang/lib/Headers/hlsl/hlsl_detail.h
index c691d85283de4..c6dd5494f149f 100644
--- a/clang/lib/Headers/hlsl/hlsl_detail.h
+++ b/clang/lib/Headers/hlsl/hlsl_detail.h
@@ -45,6 +45,20 @@ template struct is_arithmetic {
static const bool Value = __is_arithmetic(T);
};
+template
+constexpr enable_if_t::value || is_same::value, T>
+fmod_vec_impl(vector X, vector Y) {
+#if !defined(__DirectX__)
+ return __builtin_elementwise_fmod(X, Y);
+#else
+ vector div = X / Y;
+ vector result =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div)) * Y;
+ vector condition = (div >= -div);
+ vector realResult = __builtin_hlsl_select(condition, result, -result);
+ return realResult;
+#endif
+}
+
} // namespace __detail
} // namespace hlsl
#endif //_HLSL_HLSL_DETAILS_H_
>From 69e1d874bb2d679151e2a44e55a8a9d38dafa581 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Wed, 26 Feb 2025 15:57:59 -0800
Subject: [PATCH 02/10] finished implementation, working on fmod.hlsl tests
---
clang/lib/Headers/hlsl/hlsl_detail.h | 14 +++---
clang/test/CodeGenHLSL/builtins/fmod.hlsl | 32 +++
2 files changed, 26 insertions(+), 20 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h
b/clang/lib/Headers/hlsl/hlsl_detail.h
index c6dd5494f149f..f43d449ce8b7b 100644
--- a/clang/lib/Headers/hlsl/hlsl_detail.h
+++ b/clang/lib/Headers/hlsl/hlsl_detail.h
@@ -45,6 +45,12 @@ template struct is_arithmetic {
static const bool Value = __is_arithmetic(T);
};
+template
+constexpr enable_if_t::value || is_same::value, T>
+fmod_impl(T X, T Y) {
+ return __builtin_elementwise_fmod(X, Y);
+}
+
template
constexpr enable_if_t::value || is_same::value, T>
fmod_vec_impl(vector X, vector Y) {
@@ -52,10 +58,10 @@ fmod_vec_impl(vector X, vector Y) {
return __builtin_elementwise_fmod(X, Y);
#else
vector div = X / Y;
- vector result =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div)) * Y;
- vector condition = (div >= -div);
- vector realResult = __builtin_hlsl_select(condition, result, -result);
- return realResult;
+ vector ge = div >= -div;
+ vector frc =
__builtin_hlsl_elementwise_frac(__builtin_elementwise_abs(div));
+ vector realFrc = __builtin_hlsl_select(ge, frc, -frc);
+ return realFrc * Y;
#endif
}
diff --git a/clang/test/CodeGenHLSL/builtins/fmod.hlsl
b/clang/test/CodeGenHLSL/builtins/fmod.hlsl
index b62967114d456..22376638bd093 100644
--- a/clang/test/CodeGenHLSL/builtins/fmod.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/fmod.hlsl
@@ -36,42 +36,42 @@
// CHECK: define [[FNATTRS]] [[TYPE]] @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn [[TYPE]]
-// CHECK: ret [[TYPE]] %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]]
@{{.*}}([[TYPE]] noundef nofpclass(nan inf) %{{.*}}, [[TYPE]] noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret [[TYPE]] %call
half test_fmod_half(half p0, half p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <2 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <2 x [[TYPE]]>
-// CHECK: ret <2 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<2
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <2 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret <2 x [[TYPE]]> %splat.splat
half2 test_fmod_half2(half2 p0, half2 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <3 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <3 x [[TYPE]]>
-// CHECK: ret <3 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<3
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <3 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}} #{{.*}}
+// CHECK: ret <3 x [[TYPE]]> %splat.splat
half3 test_fmod_half3(half3 p0, half3 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] <4 x [[TYPE]]> @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn <4 x [[TYPE]]>
-// CHECK: ret <4 x [[TYPE]]> %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] [[TYPE]] @{{.*}}(<4
x [[TYPE]]> noundef nofpclass(nan inf) %{{.*}}, <4 x [[TYPE]]> noundef
nofpclass(nan inf) %{{.*}}) #{{.*}}
+// CHECK: ret <4 x [[TYPE]]> %splat.splat
half4 test_fmod_half4(half4 p0, half4 p1) { return fmod(p0, p1); }
// CHECK: define [[FNATTRS]] float @
-// CHECK: %fmod = frem reassoc nnan ninf nsz arcp afn float
-// CHECK: ret float %fmod
+// CHECK: call reassoc nnan ninf nsz arcp afn [[FNATTRS]] float @{{.*}}(float
noundef nofpclass(nan inf) %{{.*}}, float noundef nofpclass(na
[clang] [HLSL] Add bounds checks for the HLSL `fmod` vector arguments and return types (PR #131035)
https://github.com/kmpeng created
https://github.com/llvm/llvm-project/pull/131035
Fixes #131024.
- Fixes template for scalar and vector `fmod` intrinsic overloads
- Fixes `fmod` Sema test
>From b33ab9d6a3c87c59ca9b2d3e8e5001f1d79bb620 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 11 Mar 2025 16:47:27 -0700
Subject: [PATCH] add bounds checks for the hlsl fmod vector arguments and
return types
---
.../lib/Headers/hlsl/hlsl_intrinsic_helpers.h | 4 +-
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 22 --
clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl | 44 ++-
3 files changed, 53 insertions(+), 17 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
index 5f7c047dbf340..89ab664e90ba9 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
@@ -58,9 +58,7 @@ constexpr vector reflect_vec_impl(vector I,
vector N) {
#endif
}
-template
-constexpr enable_if_t::value || is_same::value, T>
-fmod_impl(T X, T Y) {
+template constexpr T fmod_impl(T X, T Y) {
#if !defined(__DIRECTX__)
return __builtin_elementwise_fmod(X, Y);
#else
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..a48a8e998a015 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -129,19 +129,33 @@ const inline float
distance(__detail::HLSL_FIXED_VECTOR X,
/// Return the floating-point remainder of the x parameter divided by the y
/// parameter.
+template
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-const inline half fmod(half X, half Y) { return __detail::fmod_impl(X, Y); }
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ T> fmod(T X, T Y) {
+ return __detail::fmod_impl(X, Y);
+}
-const inline float fmod(float X, float Y) { return __detail::fmod_impl(X, Y); }
+template
+const inline __detail::enable_if_t<
+__detail::is_arithmetic::Value && __detail::is_same::value, T>
+fmod(T X, T Y) {
+ return __detail::fmod_impl(X, Y);
+}
template
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-const inline vector fmod(vector X, vector Y) {
+const inline __detail::HLSL_FIXED_VECTOR fmod(
+__detail::HLSL_FIXED_VECTOR X,
+__detail::HLSL_FIXED_VECTOR Y) {
return __detail::fmod_vec_impl(X, Y);
}
template
-const inline vector fmod(vector X, vector Y) {
+const inline __detail::HLSL_FIXED_VECTOR
+fmod(__detail::HLSL_FIXED_VECTOR X,
+ __detail::HLSL_FIXED_VECTOR Y) {
return __detail::fmod_vec_impl(X, Y);
}
diff --git a/clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl
b/clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl
index 86f5a6f7bea9c..fc931139e523d 100644
--- a/clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl
@@ -3,8 +3,8 @@
float test_no_second_arg(float2 p0) {
return fmod(p0);
// expected-error@-1 {{no matching function for call to 'fmod'}}
- // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires 2 arguments, but 1 was provided}}
- // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires 2 arguments, but 1 was provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 1 was provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 1 was provided}}
// expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 1 was provided}}
// expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 1 was provided}}
}
@@ -12,22 +12,46 @@ float test_no_second_arg(float2 p0) {
float test_too_many_arg(float2 p0) {
return fmod(p0, p0, p0);
// expected-error@-1 {{no matching function for call to 'fmod'}}
- // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires 2 arguments, but 3 were provided}}
- // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires 2 arguments, but 3 were provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 3 were provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 3 were provided}}
// expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 3 were provided}}
// expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 3 were provided}}
}
float test_double_inputs(double p0, double p1) {
return fmod(p0, p1);
- // expected-error@-1 {{call to 'fmod' is
[clang] [HLSL] Add bounds checks for the HLSL `fmod` vector arguments and return types (PR #131035)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/131035
>From 72625f987846d33c11e57ec4e42e98bf211f3389 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 11 Mar 2025 16:47:27 -0700
Subject: [PATCH] add bounds checks for the hlsl fmod vector arguments and
return types
---
.../lib/Headers/hlsl/hlsl_intrinsic_helpers.h | 4 +-
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 22 --
clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl | 44 ++-
3 files changed, 53 insertions(+), 17 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
index 5f7c047dbf340..89ab664e90ba9 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
@@ -58,9 +58,7 @@ constexpr vector reflect_vec_impl(vector I,
vector N) {
#endif
}
-template
-constexpr enable_if_t::value || is_same::value, T>
-fmod_impl(T X, T Y) {
+template constexpr T fmod_impl(T X, T Y) {
#if !defined(__DIRECTX__)
return __builtin_elementwise_fmod(X, Y);
#else
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..a48a8e998a015 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -129,19 +129,33 @@ const inline float
distance(__detail::HLSL_FIXED_VECTOR X,
/// Return the floating-point remainder of the x parameter divided by the y
/// parameter.
+template
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-const inline half fmod(half X, half Y) { return __detail::fmod_impl(X, Y); }
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ T> fmod(T X, T Y) {
+ return __detail::fmod_impl(X, Y);
+}
-const inline float fmod(float X, float Y) { return __detail::fmod_impl(X, Y); }
+template
+const inline __detail::enable_if_t<
+__detail::is_arithmetic::Value && __detail::is_same::value, T>
+fmod(T X, T Y) {
+ return __detail::fmod_impl(X, Y);
+}
template
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-const inline vector fmod(vector X, vector Y) {
+const inline __detail::HLSL_FIXED_VECTOR fmod(
+__detail::HLSL_FIXED_VECTOR X,
+__detail::HLSL_FIXED_VECTOR Y) {
return __detail::fmod_vec_impl(X, Y);
}
template
-const inline vector fmod(vector X, vector Y) {
+const inline __detail::HLSL_FIXED_VECTOR
+fmod(__detail::HLSL_FIXED_VECTOR X,
+ __detail::HLSL_FIXED_VECTOR Y) {
return __detail::fmod_vec_impl(X, Y);
}
diff --git a/clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl
b/clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl
index 86f5a6f7bea9c..fc931139e523d 100644
--- a/clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/fmod-errors.hlsl
@@ -3,8 +3,8 @@
float test_no_second_arg(float2 p0) {
return fmod(p0);
// expected-error@-1 {{no matching function for call to 'fmod'}}
- // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires 2 arguments, but 1 was provided}}
- // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires 2 arguments, but 1 was provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 1 was provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 1 was provided}}
// expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 1 was provided}}
// expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 1 was provided}}
}
@@ -12,22 +12,46 @@ float test_no_second_arg(float2 p0) {
float test_too_many_arg(float2 p0) {
return fmod(p0, p0, p0);
// expected-error@-1 {{no matching function for call to 'fmod'}}
- // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires 2 arguments, but 3 were provided}}
- // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires 2 arguments, but 3 were provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 3 were provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 3 were provided}}
// expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 3 were provided}}
// expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires 2 arguments, but 3 were provided}}
}
float test_double_inputs(double p0, double p1) {
return fmod(p0, p1);
- // expected-error@-1 {{call to 'fmod' is ambiguous}}
- // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function}}
- // expected-note@hlsl/hl
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng deleted https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng created
https://github.com/llvm/llvm-project/pull/132288
Closes #99156.
Tasks completed:
- Implement `smoothstep` using HLSL source in `hlsl_intrinsics.h`
- Implement the `smoothstep` SPIR-V target built-in in
`clang/include/clang/Basic/BuiltinsSPIRV.td`
- Add sema checks for `smoothstep` to `CheckSPIRVBuiltinFunctionCall` in
`clang/lib/Sema/SemaSPIRV.cpp`
- Add codegen for spv `smoothstep` to `EmitSPIRVBuiltinExpr` in `CGBuiltin.cpp`
- Add codegen tests to `clang/test/CodeGenHLSL/builtins/smoothstep.hlsl`
- Add spv codegen test to `clang/test/CodeGenSPIRV/Builtins/smoothstep.c`
- Add sema tests to `clang/test/SemaHLSL/BuiltIns/smoothstep-errors.hlsl`
- Add spv sema tests to `clang/test/SemaSPIRV/BuiltIns/smoothstep-errors.c`
- Create the `int_spv_smoothstep` intrinsic in `IntrinsicsSPIRV.td`
- In SPIRVInstructionSelector.cpp create the `smoothstep` lowering and map it
to `int_spv_smoothstep` in `SPIRVInstructionSelector::selectIntrinsic`
- Create SPIR-V backend test case in
`llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll`
- Create SPIR-V backend test case in
`llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll`
>From 0a5da660c5aae053d87d556e59f98c121d916b79 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 18 Mar 2025 13:25:10 -0700
Subject: [PATCH 1/5] create int_spv_smoothstep intrinsic, create smoothstep
lowering & map to int_spv_smoothstep, create SPIR-V backend test cases
---
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 2 +
.../SPIRV/hlsl-intrinsics/smoothstep.ll | 60 ++
llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll | 61 +++
4 files changed, 124 insertions(+)
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
create mode 100644 llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 4a0e10db2f1e4..7760961de7b6c 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -75,6 +75,7 @@ let TargetPrefix = "spv" in {
def int_spv_reflect : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
[LLVMMatchType<0>], [IntrNoMem]>;
+ def int_spv_smoothstep : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[LLVMMatchType<0>, llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_fdot :
DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 2aba950037ec3..73b06027823f3 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -3121,6 +3121,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register
ResVReg,
return selectExtInst(ResVReg, ResType, I, CL::rsqrt, GL::InverseSqrt);
case Intrinsic::spv_sign:
return selectSign(ResVReg, ResType, I);
+ case Intrinsic::spv_smoothstep:
+return selectExtInst(ResVReg, ResType, I, CL::smoothstep, GL::SmoothStep);
case Intrinsic::spv_firstbituhigh: // There is no CL equivalent of FindUMsb
return selectFirstBitHigh(ResVReg, ResType, I, /*IsSigned=*/false);
case Intrinsic::spv_firstbitshigh: // There is no CL equivalent of FindSMsb
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
new file mode 100644
index 0..09f93ab7955d3
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
+
+; Make sure SPIRV operation function calls for smoothstep are lowered
correctly.
+
+; CHECK-DAG: %[[#op_ext_glsl:]] = OpExtInstImport "GLSL.std.450"
+; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16
+; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4
+; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32
+; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4
+
+define noundef half @smoothstep_half(half noundef %a, half noundef %b, half
noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_16]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call half @
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -101,6 +101,44 @@ bool SemaSPIRV::CheckSPIRVBuiltinFunctionCall(unsigned
BuiltinID,
TheCall->setType(RetTy);
break;
}
+ case SPIRV::BI__builtin_spirv_smoothstep: {
+if (SemaRef.checkArgCount(TheCall, 3))
+ return true;
+
+ExprResult A = TheCall->getArg(0);
+QualType ArgTyA = A.get()->getType();
+auto *VTyA = ArgTyA->getAs();
+if (!(ArgTyA->isScalarType() || VTyA)) {
kmpeng wrote:
Would using `SemaRef.BuiltinElementwiseTernaryMath` with
`ArgTyRestr=Sema::EltwiseBuiltinArgTyRestriction::FloatTy` be appropriate here
for the first check?
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -0,0 +1,44 @@ +// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5 + +// RUN: %clang_cc1 -O1 -triple spirv-pc-vulkan-compute %s -emit-llvm -o - | FileCheck %s + +typedef float float2 __attribute__((ext_vector_type(2))); +typedef float float3 __attribute__((ext_vector_type(3))); +typedef float float4 __attribute__((ext_vector_type(4))); + +// CHECK-LABEL: define spir_func float @test_smoothstep_float( kmpeng wrote: Yes. It also supports int types so will add both to the tests. https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -101,6 +101,44 @@ bool SemaSPIRV::CheckSPIRVBuiltinFunctionCall(unsigned
BuiltinID,
TheCall->setType(RetTy);
break;
}
+ case SPIRV::BI__builtin_spirv_smoothstep: {
+if (SemaRef.checkArgCount(TheCall, 3))
+ return true;
+
+ExprResult A = TheCall->getArg(0);
+QualType ArgTyA = A.get()->getType();
+auto *VTyA = ArgTyA->getAs();
+if (!(ArgTyA->isScalarType() || VTyA)) {
kmpeng wrote:
Code updated with suggested changes.
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng deleted https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/132288
>From 0a5da660c5aae053d87d556e59f98c121d916b79 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 18 Mar 2025 13:25:10 -0700
Subject: [PATCH 1/6] create int_spv_smoothstep intrinsic, create smoothstep
lowering & map to int_spv_smoothstep, create SPIR-V backend test cases
---
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 2 +
.../SPIRV/hlsl-intrinsics/smoothstep.ll | 60 ++
llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll | 61 +++
4 files changed, 124 insertions(+)
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
create mode 100644 llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 4a0e10db2f1e4..7760961de7b6c 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -75,6 +75,7 @@ let TargetPrefix = "spv" in {
def int_spv_reflect : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
[LLVMMatchType<0>], [IntrNoMem]>;
+ def int_spv_smoothstep : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[LLVMMatchType<0>, llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_fdot :
DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 2aba950037ec3..73b06027823f3 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -3121,6 +3121,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register
ResVReg,
return selectExtInst(ResVReg, ResType, I, CL::rsqrt, GL::InverseSqrt);
case Intrinsic::spv_sign:
return selectSign(ResVReg, ResType, I);
+ case Intrinsic::spv_smoothstep:
+return selectExtInst(ResVReg, ResType, I, CL::smoothstep, GL::SmoothStep);
case Intrinsic::spv_firstbituhigh: // There is no CL equivalent of FindUMsb
return selectFirstBitHigh(ResVReg, ResType, I, /*IsSigned=*/false);
case Intrinsic::spv_firstbitshigh: // There is no CL equivalent of FindSMsb
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
new file mode 100644
index 0..09f93ab7955d3
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
+
+; Make sure SPIRV operation function calls for smoothstep are lowered
correctly.
+
+; CHECK-DAG: %[[#op_ext_glsl:]] = OpExtInstImport "GLSL.std.450"
+; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16
+; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4
+; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32
+; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4
+
+define noundef half @smoothstep_half(half noundef %a, half noundef %b, half
noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_16]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call half @llvm.spv.smoothstep.f16(half %a, half %b, half
%c)
+ ret half %spv.smoothstep
+}
+
+define noundef float @smoothstep_float(float noundef %a, float noundef %b,
float noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_32]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_32]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call float @llvm.spv.smoothstep.f32(float %a, float %b,
float %c)
+ ret float %spv.smoothstep
+}
+
+define noundef <4 x half> @smoothstep_half4(<4 x half> noundef %a, <4 x half>
noundef %b, <4 x half> noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#vec4_float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#]] = OpExtInst
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -20955,6 +20955,25 @@ Value *CodeGenFunction::EmitSPIRVBuiltinExpr(unsigned
BuiltinID,
/*ReturnType=*/I->getType(), Intrinsic::spv_reflect,
ArrayRef{I, N}, nullptr, "spv.reflect");
}
+ case SPIRV::BI__builtin_spirv_smoothstep: {
+Value *Min = EmitScalarExpr(E->getArg(0));
+Value *Max = EmitScalarExpr(E->getArg(1));
+Value *X = EmitScalarExpr(E->getArg(2));
+assert(E->getArg(0)->getType()->hasFloatingRepresentation() &&
+ E->getArg(1)->getType()->hasFloatingRepresentation() &&
+ E->getArg(2)->getType()->hasFloatingRepresentation() &&
+ "SmoothStep operands must have a float representation");
+assert((E->getArg(0)->getType()->isScalarType() ||
kmpeng wrote:
Code updated with suggested changes.
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -20955,6 +20955,25 @@ Value *CodeGenFunction::EmitSPIRVBuiltinExpr(unsigned
BuiltinID,
/*ReturnType=*/I->getType(), Intrinsic::spv_reflect,
ArrayRef{I, N}, nullptr, "spv.reflect");
}
+ case SPIRV::BI__builtin_spirv_smoothstep: {
+Value *Min = EmitScalarExpr(E->getArg(0));
+Value *Max = EmitScalarExpr(E->getArg(1));
+Value *X = EmitScalarExpr(E->getArg(2));
+assert(E->getArg(0)->getType()->hasFloatingRepresentation() &&
+ E->getArg(1)->getType()->hasFloatingRepresentation() &&
+ E->getArg(2)->getType()->hasFloatingRepresentation() &&
+ "SmoothStep operands must have a float representation");
+assert((E->getArg(0)->getType()->isScalarType() ||
+E->getArg(0)->getType()->isVectorType()) &&
+ (E->getArg(1)->getType()->isScalarType() ||
+E->getArg(1)->getType()->isVectorType()) &&
+ (E->getArg(2)->getType()->isScalarType() ||
+E->getArg(2)->getType()->isVectorType()) &&
+ "SmoothStep operands must be a scalar or vector");
+return Builder.CreateIntrinsic(
+/*ReturnType=*/Min->getType(), Intrinsic::spv_smoothstep,
+ArrayRef{Min, Max, X}, nullptr, "spv.smoothstep");
kmpeng wrote:
Code updated with suggested changes.
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -0,0 +1,18 @@
+// RUN: %clang_cc1 %s -triple spirv-pc-vulkan-compute -verify
+
+typedef float float2 __attribute__((ext_vector_type(2)));
+
+float2 test_no_second_arg(float2 p0) {
+ return __builtin_spirv_smoothstep(p0);
+ // expected-error@-1 {{too few arguments to function call, expected 3, have
1}}
+}
+
+float2 test_no_third_arg(float2 p0) {
+ return __builtin_spirv_smoothstep(p0, p0);
+ // expected-error@-1 {{too few arguments to function call, expected 3, have
2}}
+}
+
+float2 test_too_many_arg(float2 p0) {
+ return __builtin_spirv_smoothstep(p0, p0, p0, p0);
+ // expected-error@-1 {{too many arguments to function call, expected 3, have
4}}
+}
kmpeng wrote:
Code updated with suggested changes.
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/132288
>From 0a5da660c5aae053d87d556e59f98c121d916b79 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 18 Mar 2025 13:25:10 -0700
Subject: [PATCH 1/7] create int_spv_smoothstep intrinsic, create smoothstep
lowering & map to int_spv_smoothstep, create SPIR-V backend test cases
---
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 2 +
.../SPIRV/hlsl-intrinsics/smoothstep.ll | 60 ++
llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll | 61 +++
4 files changed, 124 insertions(+)
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
create mode 100644 llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 4a0e10db2f1e4..7760961de7b6c 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -75,6 +75,7 @@ let TargetPrefix = "spv" in {
def int_spv_reflect : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
[LLVMMatchType<0>], [IntrNoMem]>;
+ def int_spv_smoothstep : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[LLVMMatchType<0>, llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_fdot :
DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 2aba950037ec3..73b06027823f3 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -3121,6 +3121,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register
ResVReg,
return selectExtInst(ResVReg, ResType, I, CL::rsqrt, GL::InverseSqrt);
case Intrinsic::spv_sign:
return selectSign(ResVReg, ResType, I);
+ case Intrinsic::spv_smoothstep:
+return selectExtInst(ResVReg, ResType, I, CL::smoothstep, GL::SmoothStep);
case Intrinsic::spv_firstbituhigh: // There is no CL equivalent of FindUMsb
return selectFirstBitHigh(ResVReg, ResType, I, /*IsSigned=*/false);
case Intrinsic::spv_firstbitshigh: // There is no CL equivalent of FindSMsb
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
new file mode 100644
index 0..09f93ab7955d3
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
+
+; Make sure SPIRV operation function calls for smoothstep are lowered
correctly.
+
+; CHECK-DAG: %[[#op_ext_glsl:]] = OpExtInstImport "GLSL.std.450"
+; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16
+; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4
+; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32
+; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4
+
+define noundef half @smoothstep_half(half noundef %a, half noundef %b, half
noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_16]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call half @llvm.spv.smoothstep.f16(half %a, half %b, half
%c)
+ ret half %spv.smoothstep
+}
+
+define noundef float @smoothstep_float(float noundef %a, float noundef %b,
float noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_32]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_32]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call float @llvm.spv.smoothstep.f32(float %a, float %b,
float %c)
+ ret float %spv.smoothstep
+}
+
+define noundef <4 x half> @smoothstep_half4(<4 x half> noundef %a, <4 x half>
noundef %b, <4 x half> noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#vec4_float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#]] = OpExtInst
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng created
https://github.com/llvm/llvm-project/pull/134171
Closes #99135.
Tasks completed:
- Wrote implementation in `hlsl_intrinsics.h`/`hlsl_intrinsic_helpers.h` that
matches DXC
- Created overloads in `hlsl_compat_overloads.h` that take doubles/ints and
explicitly converts them to floats, like DXC does
- Added codegen tests to `clang/test/CodeGenHLSL/builtins/lit.hlsl` and
`lit-overloads.hlsl`
- Added sema tests to `clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl` and
`lit-errors-16bit.hlsl`
>From 719bb94279f64f134c826faa22898e4e549bb23c Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Thu, 27 Mar 2025 14:39:27 -0700
Subject: [PATCH 1/6] finished lit implementation, added codegen and sema tests
---
.../lib/Headers/hlsl/hlsl_intrinsic_helpers.h | 12 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 31
clang/test/CodeGenHLSL/builtins/lit.hlsl | 36 +
clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl | 50 +++
4 files changed, 129 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/lit.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
index 8cdd63d7e07bb..619d4c59f8074 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
@@ -101,6 +101,18 @@ constexpr vector smoothstep_vec_impl(vector
Min, vector Max,
#endif
}
+template
+constexpr vector lit_impl(T N_dot_l, T N_dot_h, T M) {
+ bool Cond1 = N_dot_l < 0;
+ T ClampedP1 = select(Cond1, 0, N_dot_l);
+ vector Result = {1, ClampedP1, 0, 1};
+ bool CombinedCond = or (Cond1, (N_dot_h < 0));
+ T LogP2 = log(N_dot_h);
+ T Exp = exp(LogP2 * M);
+ Result[2] = select(CombinedCond, 0, Exp);
+ return Result;
+}
+
} // namespace __detail
} // namespace hlsl
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index fd799b8d874ae..5b6692d2a4281 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -253,6 +253,37 @@ const inline float
length(__detail::HLSL_FIXED_VECTOR X) {
return __detail::length_vec_impl(X);
}
+//===--===//
+// lit builtins
+//===--===//
+
+/// \fn vector lit(T x, T y)
+/// \brief Returns a lighting coefficient vector.
+/// \param N_dot_l The dot product of the normalized surface normal and the
+/// light vector.
+/// \param N_dot_h The dot product of the half-angle vector and the surface
+/// normal.
+/// \param M A specular exponent.
+///
+/// This function returns a lighting coefficient vector (ambient, diffuse,
+/// specular, 1).
+
+template
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ vector> lit(T N_dot_l, T N_dot_h, T M) {
+ return __detail::lit_impl(N_dot_l, N_dot_h, M);
+}
+
+template
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ vector>
+lit(T N_dot_l, T N_dot_h, T M) {
+ return __detail::lit_impl(N_dot_l, N_dot_h, M);
+}
+
//===--===//
// D3DCOLORtoUBYTE4 builtin
//===--===//
diff --git a/clang/test/CodeGenHLSL/builtins/lit.hlsl
b/clang/test/CodeGenHLSL/builtins/lit.hlsl
new file mode 100644
index 0..3fb0a2c5b5d6f
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/lit.hlsl
@@ -0,0 +1,36 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
UTC_ARGS: --version 5
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple
dxil-pc-shadermodel6.3-library %s -fnative-half-type -emit-llvm -O1 -o - |
FileCheck %s
+
+// CHECK-LABEL: define noundef nofpclass(nan inf) <4 x half>
@_Z13test_lit_halfDhDhDh(
+// CHECK-SAME: half noundef nofpclass(nan inf) [[N_DOT_L:%.*]], half noundef
nofpclass(nan inf) [[N_DOT_H:%.*]], half noundef nofpclass(nan inf) [[M:%.*]])
local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT:[[CMP_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt half
[[N_DOT_L]], 0xH
+// CHECK-NEXT:[[HLSL_SELECT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn half @llvm.maxnum.f16(half [[N_DOT_L]], half 0xH)
+// CHECK-NEXT:[[VECINIT2_I:%.*]] = insertelement <4 x half> , half [[HLSL_SELECT_I]], i64 1
+// CHECK-NEXT:[[CMP4_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt
half [[N_DOT_H]], 0xH
+// CHECK-NEXT:[[HLSL_OR_I:%.*]] = or i1 [[CMP_I]], [[CMP4_I]]
+// CHECK-NEXT:[
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng deleted https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
@@ -280,6 +280,22 @@ constexpr bool4 isinf(double4 V) { return
isinf((float4)V); }
_DXC_COMPAT_TERNARY_DOUBLE_OVERLOADS(lerp)
_DXC_COMPAT_TERNARY_INTEGER_OVERLOADS(lerp)
+//===--===//
+// lit builtins overloads
+//===--===//
+
+template
+constexpr __detail::enable_if_t<__detail::is_arithmetic::Value &&
+(__detail::is_same::value ||
kmpeng wrote:
The ambiguous error doesn't appear anymore. But removing the templates lets
vector half and float inputs pass so I don't think we can do that.
https://github.com/llvm/llvm-project/pull/134171
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/134171
>From 719bb94279f64f134c826faa22898e4e549bb23c Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Thu, 27 Mar 2025 14:39:27 -0700
Subject: [PATCH 1/7] finished lit implementation, added codegen and sema tests
---
.../lib/Headers/hlsl/hlsl_intrinsic_helpers.h | 12 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 31
clang/test/CodeGenHLSL/builtins/lit.hlsl | 36 +
clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl | 50 +++
4 files changed, 129 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/lit.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
index 8cdd63d7e07bb..619d4c59f8074 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
@@ -101,6 +101,18 @@ constexpr vector smoothstep_vec_impl(vector
Min, vector Max,
#endif
}
+template
+constexpr vector lit_impl(T N_dot_l, T N_dot_h, T M) {
+ bool Cond1 = N_dot_l < 0;
+ T ClampedP1 = select(Cond1, 0, N_dot_l);
+ vector Result = {1, ClampedP1, 0, 1};
+ bool CombinedCond = or (Cond1, (N_dot_h < 0));
+ T LogP2 = log(N_dot_h);
+ T Exp = exp(LogP2 * M);
+ Result[2] = select(CombinedCond, 0, Exp);
+ return Result;
+}
+
} // namespace __detail
} // namespace hlsl
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index fd799b8d874ae..5b6692d2a4281 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -253,6 +253,37 @@ const inline float
length(__detail::HLSL_FIXED_VECTOR X) {
return __detail::length_vec_impl(X);
}
+//===--===//
+// lit builtins
+//===--===//
+
+/// \fn vector lit(T x, T y)
+/// \brief Returns a lighting coefficient vector.
+/// \param N_dot_l The dot product of the normalized surface normal and the
+/// light vector.
+/// \param N_dot_h The dot product of the half-angle vector and the surface
+/// normal.
+/// \param M A specular exponent.
+///
+/// This function returns a lighting coefficient vector (ambient, diffuse,
+/// specular, 1).
+
+template
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ vector> lit(T N_dot_l, T N_dot_h, T M) {
+ return __detail::lit_impl(N_dot_l, N_dot_h, M);
+}
+
+template
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ vector>
+lit(T N_dot_l, T N_dot_h, T M) {
+ return __detail::lit_impl(N_dot_l, N_dot_h, M);
+}
+
//===--===//
// D3DCOLORtoUBYTE4 builtin
//===--===//
diff --git a/clang/test/CodeGenHLSL/builtins/lit.hlsl
b/clang/test/CodeGenHLSL/builtins/lit.hlsl
new file mode 100644
index 0..3fb0a2c5b5d6f
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/lit.hlsl
@@ -0,0 +1,36 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
UTC_ARGS: --version 5
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple
dxil-pc-shadermodel6.3-library %s -fnative-half-type -emit-llvm -O1 -o - |
FileCheck %s
+
+// CHECK-LABEL: define noundef nofpclass(nan inf) <4 x half>
@_Z13test_lit_halfDhDhDh(
+// CHECK-SAME: half noundef nofpclass(nan inf) [[N_DOT_L:%.*]], half noundef
nofpclass(nan inf) [[N_DOT_H:%.*]], half noundef nofpclass(nan inf) [[M:%.*]])
local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT:[[CMP_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt half
[[N_DOT_L]], 0xH
+// CHECK-NEXT:[[HLSL_SELECT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn half @llvm.maxnum.f16(half [[N_DOT_L]], half 0xH)
+// CHECK-NEXT:[[VECINIT2_I:%.*]] = insertelement <4 x half> , half [[HLSL_SELECT_I]], i64 1
+// CHECK-NEXT:[[CMP4_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt
half [[N_DOT_H]], 0xH
+// CHECK-NEXT:[[HLSL_OR_I:%.*]] = or i1 [[CMP_I]], [[CMP4_I]]
+// CHECK-NEXT:[[ELT_LOG_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
half @llvm.log.f16(half [[N_DOT_H]])
+// CHECK-NEXT:[[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn half
[[ELT_LOG_I]], [[M]]
+// CHECK-NEXT:[[ELT_EXP_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
half @llvm.exp.f16(half [[MUL_I]])
+// CHECK-NEXT:[[HLSL_SELECT7_I:%.*]] = select reassoc nnan ninf nsz arcp
afn i1 [[HLSL_OR_I]], half 0xH, half [[ELT_EXP_I]]
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
@@ -253,6 +253,37 @@ const inline float
length(__detail::HLSL_FIXED_VECTOR X) {
return __detail::length_vec_impl(X);
}
+//===--===//
+// lit builtins
+//===--===//
+
+/// \fn vector lit(T NDotL, T NDotH, T M)
+/// \brief Returns a lighting coefficient vector.
+/// \param NDotL The dot product of the normalized surface normal and the
+/// light vector.
+/// \param NDotH The dot product of the half-angle vector and the surface
+/// normal.
+/// \param M A specular exponent.
+///
+/// This function returns a lighting coefficient vector (ambient, diffuse,
+/// specular, 1).
+
+template
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
kmpeng wrote:
> I believe you should not worry about implicit truncation here.
Hm ok, curious what other people's thoughts are on this. I was just trying to
match the behavior of DXC since calling with vector inputs on DXC doesn't
compile. I still think `_DXC_COMPAT_TERNARY_DOUBLE_OVERLOADS(lit)` doesn't work
regardless because the return type is always a vector of 4.
https://github.com/llvm/llvm-project/pull/134171
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
@@ -0,0 +1,36 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
UTC_ARGS: --version 5
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple
dxil-pc-shadermodel6.3-library %s -fnative-half-type -emit-llvm -O1 -o - |
FileCheck %s
+
+// CHECK-LABEL: test_lit_half
+// CHECK-SAME: half noundef nofpclass(nan inf) [[NDOTL:%.*]], half noundef
nofpclass(nan inf) [[NDOTH:%.*]], half noundef nofpclass(nan inf) [[M:%.*]])
local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT:[[CMP_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt half
[[NDOTL]], 0xH
+// CHECK-NEXT:[[HLSL_SELECT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn half @llvm.maxnum.f16(half [[NDOTL]], half 0xH)
+// CHECK-NEXT:[[VECINIT2_I:%.*]] = insertelement <4 x half> , half [[HLSL_SELECT_I]], i64 1
+// CHECK-NEXT:[[CMP4_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt
half [[NDOTH]], 0xH
+// CHECK-NEXT:[[HLSL_OR_I:%.*]] = or i1 [[CMP_I]], [[CMP4_I]]
+// CHECK-NEXT:[[ELT_LOG_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
half @llvm.log.f16(half [[NDOTH]])
+// CHECK-NEXT:[[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn half
[[ELT_LOG_I]], [[M]]
+// CHECK-NEXT:[[ELT_EXP_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
half @llvm.exp.f16(half [[MUL_I]])
+// CHECK-NEXT:[[HLSL_SELECT7_I:%.*]] = select reassoc nnan ninf nsz arcp
afn i1 [[HLSL_OR_I]], half 0xH, half [[ELT_EXP_I]]
+// CHECK-NEXT:[[VECINS_I:%.*]] = insertelement <4 x half> [[VECINIT2_I]],
half [[HLSL_SELECT7_I]], i64 2
+// CHECK-NEXT:ret <4 x half> [[VECINS_I]]
+//
kmpeng wrote:
Code updated
https://github.com/llvm/llvm-project/pull/134171
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
@@ -101,6 +101,16 @@ constexpr vector smoothstep_vec_impl(vector
Min, vector Max,
#endif
}
+template constexpr vector lit_impl(T NDotL, T NDotH, T M) {
+ bool DiffuseCond = NDotL < 0;
+ T Diffuse = select(DiffuseCond, 0, NDotL);
+ vector Result = {1, Diffuse, 0, 1};
+ bool SpecularCond = or (DiffuseCond, (NDotH < 0));
kmpeng wrote:
Code updated
https://github.com/llvm/llvm-project/pull/134171
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/134171
>From 719bb94279f64f134c826faa22898e4e549bb23c Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Thu, 27 Mar 2025 14:39:27 -0700
Subject: [PATCH 1/9] finished lit implementation, added codegen and sema tests
---
.../lib/Headers/hlsl/hlsl_intrinsic_helpers.h | 12 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 31
clang/test/CodeGenHLSL/builtins/lit.hlsl | 36 +
clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl | 50 +++
4 files changed, 129 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/lit.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
index 8cdd63d7e07bb..619d4c59f8074 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
@@ -101,6 +101,18 @@ constexpr vector smoothstep_vec_impl(vector
Min, vector Max,
#endif
}
+template
+constexpr vector lit_impl(T N_dot_l, T N_dot_h, T M) {
+ bool Cond1 = N_dot_l < 0;
+ T ClampedP1 = select(Cond1, 0, N_dot_l);
+ vector Result = {1, ClampedP1, 0, 1};
+ bool CombinedCond = or (Cond1, (N_dot_h < 0));
+ T LogP2 = log(N_dot_h);
+ T Exp = exp(LogP2 * M);
+ Result[2] = select(CombinedCond, 0, Exp);
+ return Result;
+}
+
} // namespace __detail
} // namespace hlsl
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index fd799b8d874ae..5b6692d2a4281 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -253,6 +253,37 @@ const inline float
length(__detail::HLSL_FIXED_VECTOR X) {
return __detail::length_vec_impl(X);
}
+//===--===//
+// lit builtins
+//===--===//
+
+/// \fn vector lit(T x, T y)
+/// \brief Returns a lighting coefficient vector.
+/// \param N_dot_l The dot product of the normalized surface normal and the
+/// light vector.
+/// \param N_dot_h The dot product of the half-angle vector and the surface
+/// normal.
+/// \param M A specular exponent.
+///
+/// This function returns a lighting coefficient vector (ambient, diffuse,
+/// specular, 1).
+
+template
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ vector> lit(T N_dot_l, T N_dot_h, T M) {
+ return __detail::lit_impl(N_dot_l, N_dot_h, M);
+}
+
+template
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ vector>
+lit(T N_dot_l, T N_dot_h, T M) {
+ return __detail::lit_impl(N_dot_l, N_dot_h, M);
+}
+
//===--===//
// D3DCOLORtoUBYTE4 builtin
//===--===//
diff --git a/clang/test/CodeGenHLSL/builtins/lit.hlsl
b/clang/test/CodeGenHLSL/builtins/lit.hlsl
new file mode 100644
index 0..3fb0a2c5b5d6f
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/lit.hlsl
@@ -0,0 +1,36 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
UTC_ARGS: --version 5
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple
dxil-pc-shadermodel6.3-library %s -fnative-half-type -emit-llvm -O1 -o - |
FileCheck %s
+
+// CHECK-LABEL: define noundef nofpclass(nan inf) <4 x half>
@_Z13test_lit_halfDhDhDh(
+// CHECK-SAME: half noundef nofpclass(nan inf) [[N_DOT_L:%.*]], half noundef
nofpclass(nan inf) [[N_DOT_H:%.*]], half noundef nofpclass(nan inf) [[M:%.*]])
local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT:[[CMP_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt half
[[N_DOT_L]], 0xH
+// CHECK-NEXT:[[HLSL_SELECT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn half @llvm.maxnum.f16(half [[N_DOT_L]], half 0xH)
+// CHECK-NEXT:[[VECINIT2_I:%.*]] = insertelement <4 x half> , half [[HLSL_SELECT_I]], i64 1
+// CHECK-NEXT:[[CMP4_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt
half [[N_DOT_H]], 0xH
+// CHECK-NEXT:[[HLSL_OR_I:%.*]] = or i1 [[CMP_I]], [[CMP4_I]]
+// CHECK-NEXT:[[ELT_LOG_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
half @llvm.log.f16(half [[N_DOT_H]])
+// CHECK-NEXT:[[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn half
[[ELT_LOG_I]], [[M]]
+// CHECK-NEXT:[[ELT_EXP_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
half @llvm.exp.f16(half [[MUL_I]])
+// CHECK-NEXT:[[HLSL_SELECT7_I:%.*]] = select reassoc nnan ninf nsz arcp
afn i1 [[HLSL_OR_I]], half 0xH, half [[ELT_EXP_I]]
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
@@ -253,6 +253,37 @@ const inline float
length(__detail::HLSL_FIXED_VECTOR X) {
return __detail::length_vec_impl(X);
}
+//===--===//
+// lit builtins
+//===--===//
+
+/// \fn vector lit(T NDotL, T NDotH, T M)
+/// \brief Returns a lighting coefficient vector.
+/// \param NDotL The dot product of the normalized surface normal and the
+/// light vector.
+/// \param NDotH The dot product of the half-angle vector and the surface
+/// normal.
+/// \param M A specular exponent.
+///
+/// This function returns a lighting coefficient vector (ambient, diffuse,
+/// specular, 1).
+
+template
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
kmpeng wrote:
> If DXC does implicit cast then i think we should as well.
DXC doesn't implicit cast so going to leave the code as is
https://github.com/llvm/llvm-project/pull/134171
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
@@ -280,6 +280,22 @@ constexpr bool4 isinf(double4 V) { return
isinf((float4)V); }
_DXC_COMPAT_TERNARY_DOUBLE_OVERLOADS(lerp)
_DXC_COMPAT_TERNARY_INTEGER_OVERLOADS(lerp)
+//===--===//
+// lit builtins overloads
+//===--===//
+
+template
+constexpr __detail::enable_if_t<__detail::is_arithmetic::Value &&
+(__detail::is_same::value ||
kmpeng wrote:
Code updated to allow the bool overload
https://github.com/llvm/llvm-project/pull/134171
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
@@ -280,6 +280,22 @@ constexpr bool4 isinf(double4 V) { return
isinf((float4)V); }
_DXC_COMPAT_TERNARY_DOUBLE_OVERLOADS(lerp)
_DXC_COMPAT_TERNARY_INTEGER_OVERLOADS(lerp)
+//===--===//
+// lit builtins overloads
+//===--===//
+
+template
+constexpr __detail::enable_if_t<__detail::is_arithmetic::Value &&
+(__detail::is_same::value ||
kmpeng wrote:
The ambiguous error doesn't appear, but the half definition in
`hlsl_instrinsics.h` doesn't seem to be used anymore when we're only checking
`__detail::is_arithmetic` in the overloads. The tablegen shows half inputs
being casted to floats and then the float definition of `lit` being called.
https://github.com/llvm/llvm-project/pull/134171
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -101,6 +101,52 @@ bool SemaSPIRV::CheckSPIRVBuiltinFunctionCall(unsigned
BuiltinID,
TheCall->setType(RetTy);
break;
}
+ case SPIRV::BI__builtin_spirv_smoothstep: {
+if (SemaRef.checkArgCount(TheCall, 3))
+ return true;
+
+// check if the all arguments have floating representation
+ExprResult A = TheCall->getArg(0);
+QualType ArgTyA = A.get()->getType();
+if (!ArgTyA->hasFloatingRepresentation()) {
+ SemaRef.Diag(A.get()->getBeginLoc(),
+ diag::err_typecheck_convert_incompatible)
kmpeng wrote:
Don't we also want to say the [api supports
matrices](https://github.com/llvm/llvm-project/pull/132288#discussion_r2008116154)
though? I don't think I can do that with `err_builtin_invalid_arg_type` unless
I edit `DiagnosticSemaKinds.td`—the closest I can get is this message:
```
1st argument must be a scalar or vector of floating-point types (was 'int')
```
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/132288
>From 0a5da660c5aae053d87d556e59f98c121d916b79 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 18 Mar 2025 13:25:10 -0700
Subject: [PATCH 1/8] create int_spv_smoothstep intrinsic, create smoothstep
lowering & map to int_spv_smoothstep, create SPIR-V backend test cases
---
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 2 +
.../SPIRV/hlsl-intrinsics/smoothstep.ll | 60 ++
llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll | 61 +++
4 files changed, 124 insertions(+)
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
create mode 100644 llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 4a0e10db2f1e4..7760961de7b6c 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -75,6 +75,7 @@ let TargetPrefix = "spv" in {
def int_spv_reflect : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
[LLVMMatchType<0>], [IntrNoMem]>;
+ def int_spv_smoothstep : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[LLVMMatchType<0>, llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_fdot :
DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 2aba950037ec3..73b06027823f3 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -3121,6 +3121,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register
ResVReg,
return selectExtInst(ResVReg, ResType, I, CL::rsqrt, GL::InverseSqrt);
case Intrinsic::spv_sign:
return selectSign(ResVReg, ResType, I);
+ case Intrinsic::spv_smoothstep:
+return selectExtInst(ResVReg, ResType, I, CL::smoothstep, GL::SmoothStep);
case Intrinsic::spv_firstbituhigh: // There is no CL equivalent of FindUMsb
return selectFirstBitHigh(ResVReg, ResType, I, /*IsSigned=*/false);
case Intrinsic::spv_firstbitshigh: // There is no CL equivalent of FindSMsb
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
new file mode 100644
index 0..09f93ab7955d3
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
+
+; Make sure SPIRV operation function calls for smoothstep are lowered
correctly.
+
+; CHECK-DAG: %[[#op_ext_glsl:]] = OpExtInstImport "GLSL.std.450"
+; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16
+; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4
+; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32
+; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4
+
+define noundef half @smoothstep_half(half noundef %a, half noundef %b, half
noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_16]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call half @llvm.spv.smoothstep.f16(half %a, half %b, half
%c)
+ ret half %spv.smoothstep
+}
+
+define noundef float @smoothstep_float(float noundef %a, float noundef %b,
float noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_32]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_32]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call float @llvm.spv.smoothstep.f32(float %a, float %b,
float %c)
+ ret float %spv.smoothstep
+}
+
+define noundef <4 x half> @smoothstep_half4(<4 x half> noundef %a, <4 x half>
noundef %b, <4 x half> noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#vec4_float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#]] = OpExtInst
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -0,0 +1,183 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
UTC_ARGS: --version 5
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
+// RUN: dxil-pc-shadermodel6.3-library %s -fnative-half-type \
+// RUN: -emit-llvm -O1 -o - | FileCheck %s
+// RUN: %clang_cc1 -finclude-default-header -triple \
+// RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \
+// RUN: -emit-llvm -O1 -o - | FileCheck %s --check-prefix=SPVCHECK
+
+// CHECK-LABEL: define noundef nofpclass(nan inf) half
@_Z20test_smoothstep_halfDhDhDh(
+// CHECK-SAME: half noundef nofpclass(nan inf) [[MIN:%.*]], half noundef
nofpclass(nan inf) [[MAX:%.*]], half noundef nofpclass(nan inf) [[X:%.*]])
local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT:[[SUB_I:%.*]] = fsub reassoc nnan ninf nsz arcp afn half
[[X]], [[MIN]]
+// CHECK-NEXT:[[SUB1_I:%.*]] = fsub reassoc nnan ninf nsz arcp afn half
[[MAX]], [[MIN]]
+// CHECK-NEXT:[[DIV_I:%.*]] = fdiv reassoc nnan ninf nsz arcp afn half
[[SUB_I]], [[SUB1_I]]
+// CHECK-NEXT:[[HLSL_SATURATE_I:%.*]] = tail call reassoc nnan ninf nsz
arcp afn half @llvm.dx.saturate.f16(half [[DIV_I]])
+// CHECK-NEXT:[[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn half
[[HLSL_SATURATE_I]], 0xH4000
+// CHECK-NEXT:[[SUB2_I:%.*]] = fsub reassoc nnan ninf nsz arcp afn half
0xH4200, [[MUL_I]]
+// CHECK-NEXT:[[TMP0:%.*]] = fmul reassoc nnan ninf nsz arcp afn half
[[HLSL_SATURATE_I]], [[HLSL_SATURATE_I]]
+// CHECK-NEXT:[[MUL4_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn half
[[TMP0]], [[SUB2_I]]
+// CHECK-NEXT:ret half [[MUL4_I]]
+//
+// SPVCHECK-LABEL: define spir_func noundef nofpclass(nan inf) half
@_Z20test_smoothstep_halfDhDhDh(
+// SPVCHECK-SAME: half noundef nofpclass(nan inf) [[MIN:%.*]], half noundef
nofpclass(nan inf) [[MAX:%.*]], half noundef nofpclass(nan inf) [[X:%.*]])
local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// SPVCHECK-NEXT: [[ENTRY:.*:]]
+// SPVCHECK-NEXT:[[CONV_I:%.*]] = fpext reassoc nnan ninf nsz arcp afn
half [[MIN]] to double
+// SPVCHECK-NEXT:[[CONV1_I:%.*]] = fpext reassoc nnan ninf nsz arcp afn
half [[MAX]] to double
+// SPVCHECK-NEXT:[[CONV2_I:%.*]] = fpext reassoc nnan ninf nsz arcp afn
half [[X]] to double
+// SPVCHECK-NEXT:[[SPV_SMOOTHSTEP_I:%.*]] = tail call reassoc nnan ninf
nsz arcp afn double @llvm.spv.smoothstep.f64(double [[CONV_I]], double
[[CONV1_I]], double [[CONV2_I]])
kmpeng wrote:
Yes, I agree. This is what I was asking about earlier. I'm not sure why this is
happening.
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -101,6 +101,52 @@ bool SemaSPIRV::CheckSPIRVBuiltinFunctionCall(unsigned
BuiltinID,
TheCall->setType(RetTy);
break;
}
+ case SPIRV::BI__builtin_spirv_smoothstep: {
+if (SemaRef.checkArgCount(TheCall, 3))
+ return true;
+
+// check if the all arguments have floating representation
+ExprResult A = TheCall->getArg(0);
kmpeng wrote:
I'm already checking `hasFloatingRepresentation` for all of the args and
emitting the corresponding diag. Do you want me to move the individual checks
into a for loop?
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/132288
>From 0a5da660c5aae053d87d556e59f98c121d916b79 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 18 Mar 2025 13:25:10 -0700
Subject: [PATCH 1/7] create int_spv_smoothstep intrinsic, create smoothstep
lowering & map to int_spv_smoothstep, create SPIR-V backend test cases
---
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 2 +
.../SPIRV/hlsl-intrinsics/smoothstep.ll | 60 ++
llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll | 61 +++
4 files changed, 124 insertions(+)
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
create mode 100644 llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 4a0e10db2f1e4..7760961de7b6c 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -75,6 +75,7 @@ let TargetPrefix = "spv" in {
def int_spv_reflect : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
[LLVMMatchType<0>], [IntrNoMem]>;
+ def int_spv_smoothstep : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[LLVMMatchType<0>, llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_fdot :
DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 2aba950037ec3..73b06027823f3 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -3121,6 +3121,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register
ResVReg,
return selectExtInst(ResVReg, ResType, I, CL::rsqrt, GL::InverseSqrt);
case Intrinsic::spv_sign:
return selectSign(ResVReg, ResType, I);
+ case Intrinsic::spv_smoothstep:
+return selectExtInst(ResVReg, ResType, I, CL::smoothstep, GL::SmoothStep);
case Intrinsic::spv_firstbituhigh: // There is no CL equivalent of FindUMsb
return selectFirstBitHigh(ResVReg, ResType, I, /*IsSigned=*/false);
case Intrinsic::spv_firstbitshigh: // There is no CL equivalent of FindSMsb
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
new file mode 100644
index 0..09f93ab7955d3
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
+
+; Make sure SPIRV operation function calls for smoothstep are lowered
correctly.
+
+; CHECK-DAG: %[[#op_ext_glsl:]] = OpExtInstImport "GLSL.std.450"
+; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16
+; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4
+; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32
+; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4
+
+define noundef half @smoothstep_half(half noundef %a, half noundef %b, half
noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_16]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call half @llvm.spv.smoothstep.f16(half %a, half %b, half
%c)
+ ret half %spv.smoothstep
+}
+
+define noundef float @smoothstep_float(float noundef %a, float noundef %b,
float noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_32]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_32]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call float @llvm.spv.smoothstep.f32(float %a, float %b,
float %c)
+ ret float %spv.smoothstep
+}
+
+define noundef <4 x half> @smoothstep_half4(<4 x half> noundef %a, <4 x half>
noundef %b, <4 x half> noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#vec4_float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#]] = OpExtInst
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng deleted https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
kmpeng wrote:
Seems like this change produces this error in the SPIR-V tests.
```
error: line 1: Capability Kernel is not allowed by Vulkan 1.2 specification (or
requires extension)
OpCapability Kernel
```
Still kind of new to this—do you know what's going on here?
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/132288
>From 1d28510653b39fcabe45ad37197674bdd0217add Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 18 Mar 2025 13:25:10 -0700
Subject: [PATCH 01/10] create int_spv_smoothstep intrinsic, create smoothstep
lowering & map to int_spv_smoothstep, create SPIR-V backend test cases
---
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 2 +
.../SPIRV/hlsl-intrinsics/smoothstep.ll | 60 ++
llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll | 61 +++
4 files changed, 124 insertions(+)
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
create mode 100644 llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 906bdfdcd4eb1..4389b86745d7f 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -75,6 +75,7 @@ let TargetPrefix = "spv" in {
def int_spv_reflect : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
[LLVMMatchType<0>], [IntrNoMem]>;
+ def int_spv_smoothstep : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[LLVMMatchType<0>, llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_fdot :
DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index fd598904d7ba5..644223ccdc0a7 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -3083,6 +3083,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register
ResVReg,
return selectExtInst(ResVReg, ResType, I, CL::rsqrt, GL::InverseSqrt);
case Intrinsic::spv_sign:
return selectSign(ResVReg, ResType, I);
+ case Intrinsic::spv_smoothstep:
+return selectExtInst(ResVReg, ResType, I, CL::smoothstep, GL::SmoothStep);
case Intrinsic::spv_firstbituhigh: // There is no CL equivalent of FindUMsb
return selectFirstBitHigh(ResVReg, ResType, I, /*IsSigned=*/false);
case Intrinsic::spv_firstbitshigh: // There is no CL equivalent of FindSMsb
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
new file mode 100644
index 0..09f93ab7955d3
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
+
+; Make sure SPIRV operation function calls for smoothstep are lowered
correctly.
+
+; CHECK-DAG: %[[#op_ext_glsl:]] = OpExtInstImport "GLSL.std.450"
+; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16
+; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4
+; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32
+; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4
+
+define noundef half @smoothstep_half(half noundef %a, half noundef %b, half
noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_16]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call half @llvm.spv.smoothstep.f16(half %a, half %b, half
%c)
+ ret half %spv.smoothstep
+}
+
+define noundef float @smoothstep_float(float noundef %a, float noundef %b,
float noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_32]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_32]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call float @llvm.spv.smoothstep.f32(float %a, float %b,
float %c)
+ ret float %spv.smoothstep
+}
+
+define noundef <4 x half> @smoothstep_half4(<4 x half> noundef %a, <4 x half>
noundef %b, <4 x half> noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#vec4_float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#]] = OpExtIns
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
kmpeng wrote:
Code updated with `--target-env vulkan1.2` added to both
`hlsl-intrinsics/smoothstep.ll` and `opencl/smoothstep.ll`. Let me know if that
looks good!
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -101,6 +101,52 @@ bool SemaSPIRV::CheckSPIRVBuiltinFunctionCall(unsigned
BuiltinID,
TheCall->setType(RetTy);
break;
}
+ case SPIRV::BI__builtin_spirv_smoothstep: {
+if (SemaRef.checkArgCount(TheCall, 3))
+ return true;
+
+// check if the all arguments have floating representation
+ExprResult A = TheCall->getArg(0);
+QualType ArgTyA = A.get()->getType();
+if (!ArgTyA->hasFloatingRepresentation()) {
+ SemaRef.Diag(A.get()->getBeginLoc(),
+ diag::err_typecheck_convert_incompatible)
kmpeng wrote:
Code updated.
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
kmpeng wrote:
Would you also like this added to the runs in `opencl/smoothstep.ll`?
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/132288
>From 7853b27b8fffa6d2c0393a4004abaac9a7954608 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 18 Mar 2025 13:25:10 -0700
Subject: [PATCH 01/10] create int_spv_smoothstep intrinsic, create smoothstep
lowering & map to int_spv_smoothstep, create SPIR-V backend test cases
---
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 2 +
.../SPIRV/hlsl-intrinsics/smoothstep.ll | 60 ++
llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll | 61 +++
4 files changed, 124 insertions(+)
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
create mode 100644 llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 4a0e10db2f1e4..7760961de7b6c 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -75,6 +75,7 @@ let TargetPrefix = "spv" in {
def int_spv_reflect : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
[LLVMMatchType<0>], [IntrNoMem]>;
+ def int_spv_smoothstep : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[LLVMMatchType<0>, llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_fdot :
DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index bd54590c87cac..1ba022b416808 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -3121,6 +3121,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register
ResVReg,
return selectExtInst(ResVReg, ResType, I, CL::rsqrt, GL::InverseSqrt);
case Intrinsic::spv_sign:
return selectSign(ResVReg, ResType, I);
+ case Intrinsic::spv_smoothstep:
+return selectExtInst(ResVReg, ResType, I, CL::smoothstep, GL::SmoothStep);
case Intrinsic::spv_firstbituhigh: // There is no CL equivalent of FindUMsb
return selectFirstBitHigh(ResVReg, ResType, I, /*IsSigned=*/false);
case Intrinsic::spv_firstbitshigh: // There is no CL equivalent of FindSMsb
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
new file mode 100644
index 0..09f93ab7955d3
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
+
+; Make sure SPIRV operation function calls for smoothstep are lowered
correctly.
+
+; CHECK-DAG: %[[#op_ext_glsl:]] = OpExtInstImport "GLSL.std.450"
+; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16
+; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4
+; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32
+; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4
+
+define noundef half @smoothstep_half(half noundef %a, half noundef %b, half
noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_16]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call half @llvm.spv.smoothstep.f16(half %a, half %b, half
%c)
+ ret half %spv.smoothstep
+}
+
+define noundef float @smoothstep_float(float noundef %a, float noundef %b,
float noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_32]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_32]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call float @llvm.spv.smoothstep.f32(float %a, float %b,
float %c)
+ ret float %spv.smoothstep
+}
+
+define noundef <4 x half> @smoothstep_half4(<4 x half> noundef %a, <4 x half>
noundef %b, <4 x half> noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#vec4_float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#]] = OpExtIns
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -101,6 +101,52 @@ bool SemaSPIRV::CheckSPIRVBuiltinFunctionCall(unsigned
BuiltinID,
TheCall->setType(RetTy);
break;
}
+ case SPIRV::BI__builtin_spirv_smoothstep: {
+if (SemaRef.checkArgCount(TheCall, 3))
+ return true;
+
+// check if the all arguments have floating representation
+ExprResult A = TheCall->getArg(0);
kmpeng wrote:
Code updated to move the checks into a for loop.
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/132288
>From 7853b27b8fffa6d2c0393a4004abaac9a7954608 Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 18 Mar 2025 13:25:10 -0700
Subject: [PATCH 1/8] create int_spv_smoothstep intrinsic, create smoothstep
lowering & map to int_spv_smoothstep, create SPIR-V backend test cases
---
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 2 +
.../SPIRV/hlsl-intrinsics/smoothstep.ll | 60 ++
llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll | 61 +++
4 files changed, 124 insertions(+)
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
create mode 100644 llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 4a0e10db2f1e4..7760961de7b6c 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -75,6 +75,7 @@ let TargetPrefix = "spv" in {
def int_spv_reflect : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
[LLVMMatchType<0>], [IntrNoMem]>;
+ def int_spv_smoothstep : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[LLVMMatchType<0>, llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_fdot :
DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index bd54590c87cac..1ba022b416808 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -3121,6 +3121,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register
ResVReg,
return selectExtInst(ResVReg, ResType, I, CL::rsqrt, GL::InverseSqrt);
case Intrinsic::spv_sign:
return selectSign(ResVReg, ResType, I);
+ case Intrinsic::spv_smoothstep:
+return selectExtInst(ResVReg, ResType, I, CL::smoothstep, GL::SmoothStep);
case Intrinsic::spv_firstbituhigh: // There is no CL equivalent of FindUMsb
return selectFirstBitHigh(ResVReg, ResType, I, /*IsSigned=*/false);
case Intrinsic::spv_firstbitshigh: // There is no CL equivalent of FindSMsb
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
new file mode 100644
index 0..09f93ab7955d3
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
+
+; Make sure SPIRV operation function calls for smoothstep are lowered
correctly.
+
+; CHECK-DAG: %[[#op_ext_glsl:]] = OpExtInstImport "GLSL.std.450"
+; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16
+; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4
+; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32
+; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4
+
+define noundef half @smoothstep_half(half noundef %a, half noundef %b, half
noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_16]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call half @llvm.spv.smoothstep.f16(half %a, half %b, half
%c)
+ ret half %spv.smoothstep
+}
+
+define noundef float @smoothstep_float(float noundef %a, float noundef %b,
float noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_32]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_32]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call float @llvm.spv.smoothstep.f32(float %a, float %b,
float %c)
+ ret float %spv.smoothstep
+}
+
+define noundef <4 x half> @smoothstep_half4(<4 x half> noundef %a, <4 x half>
noundef %b, <4 x half> noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#vec4_float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#]] = OpExtInst
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
kmpeng wrote:
@s-perron Changing the target environments to `--target-env vulkan1.3` for
hlsl-intrinsics and `--target-env spv1.6` for opencl seems to fix the issues.
Do these changes work for you?
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/132288 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/132288
>From 1d28510653b39fcabe45ad37197674bdd0217add Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Tue, 18 Mar 2025 13:25:10 -0700
Subject: [PATCH 01/11] create int_spv_smoothstep intrinsic, create smoothstep
lowering & map to int_spv_smoothstep, create SPIR-V backend test cases
---
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 +
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 2 +
.../SPIRV/hlsl-intrinsics/smoothstep.ll | 60 ++
llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll | 61 +++
4 files changed, 124 insertions(+)
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
create mode 100644 llvm/test/CodeGen/SPIRV/opencl/smoothstep.ll
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 906bdfdcd4eb1..4389b86745d7f 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -75,6 +75,7 @@ let TargetPrefix = "spv" in {
def int_spv_reflect : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
[LLVMMatchType<0>], [IntrNoMem]>;
+ def int_spv_smoothstep : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_spv_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[LLVMMatchType<0>, llvm_anyfloat_ty], [IntrNoMem]>;
def int_spv_fdot :
DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index fd598904d7ba5..644223ccdc0a7 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -3083,6 +3083,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register
ResVReg,
return selectExtInst(ResVReg, ResType, I, CL::rsqrt, GL::InverseSqrt);
case Intrinsic::spv_sign:
return selectSign(ResVReg, ResType, I);
+ case Intrinsic::spv_smoothstep:
+return selectExtInst(ResVReg, ResType, I, CL::smoothstep, GL::SmoothStep);
case Intrinsic::spv_firstbituhigh: // There is no CL equivalent of FindUMsb
return selectFirstBitHigh(ResVReg, ResType, I, /*IsSigned=*/false);
case Intrinsic::spv_firstbitshigh: // There is no CL equivalent of FindSMsb
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
new file mode 100644
index 0..09f93ab7955d3
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/smoothstep.ll
@@ -0,0 +1,60 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-unknown %s -o - |
FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o -
-filetype=obj | spirv-val %}
+
+; Make sure SPIRV operation function calls for smoothstep are lowered
correctly.
+
+; CHECK-DAG: %[[#op_ext_glsl:]] = OpExtInstImport "GLSL.std.450"
+; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16
+; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4
+; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32
+; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4
+
+define noundef half @smoothstep_half(half noundef %a, half noundef %b, half
noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_16]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_16]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call half @llvm.spv.smoothstep.f16(half %a, half %b, half
%c)
+ ret half %spv.smoothstep
+}
+
+define noundef float @smoothstep_float(float noundef %a, float noundef %b,
float noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#float_32]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#float_32]]
+ ; CHECK: %[[#]] = OpExtInst %[[#float_32]] %[[#op_ext_glsl]] SmoothStep
%[[#arg0]] %[[#arg1]] %[[#arg2]]
+ %spv.smoothstep = call float @llvm.spv.smoothstep.f32(float %a, float %b,
float %c)
+ ret float %spv.smoothstep
+}
+
+define noundef <4 x half> @smoothstep_half4(<4 x half> noundef %a, <4 x half>
noundef %b, <4 x half> noundef %c) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#vec4_float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg2:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#]] = OpExtIns
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng updated
https://github.com/llvm/llvm-project/pull/134171
>From 719bb94279f64f134c826faa22898e4e549bb23c Mon Sep 17 00:00:00 2001
From: kmpeng
Date: Thu, 27 Mar 2025 14:39:27 -0700
Subject: [PATCH 1/8] finished lit implementation, added codegen and sema tests
---
.../lib/Headers/hlsl/hlsl_intrinsic_helpers.h | 12 +
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 31
clang/test/CodeGenHLSL/builtins/lit.hlsl | 36 +
clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl | 50 +++
4 files changed, 129 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/lit.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
index 8cdd63d7e07bb..619d4c59f8074 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
@@ -101,6 +101,18 @@ constexpr vector smoothstep_vec_impl(vector
Min, vector Max,
#endif
}
+template
+constexpr vector lit_impl(T N_dot_l, T N_dot_h, T M) {
+ bool Cond1 = N_dot_l < 0;
+ T ClampedP1 = select(Cond1, 0, N_dot_l);
+ vector Result = {1, ClampedP1, 0, 1};
+ bool CombinedCond = or (Cond1, (N_dot_h < 0));
+ T LogP2 = log(N_dot_h);
+ T Exp = exp(LogP2 * M);
+ Result[2] = select(CombinedCond, 0, Exp);
+ return Result;
+}
+
} // namespace __detail
} // namespace hlsl
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index fd799b8d874ae..5b6692d2a4281 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -253,6 +253,37 @@ const inline float
length(__detail::HLSL_FIXED_VECTOR X) {
return __detail::length_vec_impl(X);
}
+//===--===//
+// lit builtins
+//===--===//
+
+/// \fn vector lit(T x, T y)
+/// \brief Returns a lighting coefficient vector.
+/// \param N_dot_l The dot product of the normalized surface normal and the
+/// light vector.
+/// \param N_dot_h The dot product of the half-angle vector and the surface
+/// normal.
+/// \param M A specular exponent.
+///
+/// This function returns a lighting coefficient vector (ambient, diffuse,
+/// specular, 1).
+
+template
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ vector> lit(T N_dot_l, T N_dot_h, T M) {
+ return __detail::lit_impl(N_dot_l, N_dot_h, M);
+}
+
+template
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ vector>
+lit(T N_dot_l, T N_dot_h, T M) {
+ return __detail::lit_impl(N_dot_l, N_dot_h, M);
+}
+
//===--===//
// D3DCOLORtoUBYTE4 builtin
//===--===//
diff --git a/clang/test/CodeGenHLSL/builtins/lit.hlsl
b/clang/test/CodeGenHLSL/builtins/lit.hlsl
new file mode 100644
index 0..3fb0a2c5b5d6f
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/lit.hlsl
@@ -0,0 +1,36 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
UTC_ARGS: --version 5
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple
dxil-pc-shadermodel6.3-library %s -fnative-half-type -emit-llvm -O1 -o - |
FileCheck %s
+
+// CHECK-LABEL: define noundef nofpclass(nan inf) <4 x half>
@_Z13test_lit_halfDhDhDh(
+// CHECK-SAME: half noundef nofpclass(nan inf) [[N_DOT_L:%.*]], half noundef
nofpclass(nan inf) [[N_DOT_H:%.*]], half noundef nofpclass(nan inf) [[M:%.*]])
local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT:[[CMP_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt half
[[N_DOT_L]], 0xH
+// CHECK-NEXT:[[HLSL_SELECT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn half @llvm.maxnum.f16(half [[N_DOT_L]], half 0xH)
+// CHECK-NEXT:[[VECINIT2_I:%.*]] = insertelement <4 x half> , half [[HLSL_SELECT_I]], i64 1
+// CHECK-NEXT:[[CMP4_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt
half [[N_DOT_H]], 0xH
+// CHECK-NEXT:[[HLSL_OR_I:%.*]] = or i1 [[CMP_I]], [[CMP4_I]]
+// CHECK-NEXT:[[ELT_LOG_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
half @llvm.log.f16(half [[N_DOT_H]])
+// CHECK-NEXT:[[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn half
[[ELT_LOG_I]], [[M]]
+// CHECK-NEXT:[[ELT_EXP_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
half @llvm.exp.f16(half [[MUL_I]])
+// CHECK-NEXT:[[HLSL_SELECT7_I:%.*]] = select reassoc nnan ninf nsz arcp
afn i1 [[HLSL_OR_I]], half 0xH, half [[ELT_EXP_I]]
[clang] [llvm] [HLSL] Implement the `smoothstep` intrinsic (PR #132288)
@@ -322,5 +322,53 @@ reflect(__detail::HLSL_FIXED_VECTOR I,
__detail::HLSL_FIXED_VECTOR N) {
return __detail::reflect_vec_impl(I, N);
}
+
+//===--===//
+// smoothstep builtin
+//===--===//
+
+/// \fn T smoothstep(T Min, T Max, T X)
+/// \brief Returns a smooth Hermite interpolation between 0 and 1, if \a X is
in
+/// the range [\a Min, \a Max].
+/// \param Min The minimum range of the x parameter.
+/// \param Max The maximum range of the x parameter.
+/// \param X The specified value to be interpolated.
+///
+/// The return value is 0.0 if \a X ≤ \a Min and 1.0 if \a X ≥ \a Max. When \a
+/// Min < \a X < \a Max, the function performs smooth Hermite interpolation
+/// between 0 and 1.
+
+template
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ T> smoothstep(T Min, T Max, T X) {
+ return __detail::smoothstep_impl(Min, Max, X);
+}
+
+template
+const inline __detail::enable_if_t<
kmpeng wrote:
I'm not sure how to do this when clang-format is an automatic process. The
other intrinsics are also formatted the same way as mine.
https://github.com/llvm/llvm-project/pull/132288
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
@@ -253,6 +253,37 @@ const inline float
length(__detail::HLSL_FIXED_VECTOR X) {
return __detail::length_vec_impl(X);
}
+//===--===//
+// lit builtins
+//===--===//
+
+/// \fn vector lit(T NDotL, T NDotH, T M)
+/// \brief Returns a lighting coefficient vector.
+/// \param NDotL The dot product of the normalized surface normal and the
+/// light vector.
+/// \param NDotH The dot product of the half-angle vector and the surface
+/// normal.
+/// \param M A specular exponent.
+///
+/// This function returns a lighting coefficient vector (ambient, diffuse,
+/// specular, 1).
+
+template
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ vector> lit(T NDotL, T NDotH, T M) {
+ return __detail::lit_impl(NDotL, NDotH, M);
+}
+
+template
+const inline __detail::enable_if_t<__detail::is_arithmetic::Value &&
+ __detail::is_same::value,
+ vector>
kmpeng wrote:
I've tried but clang-format keeps putting it back to this :\
https://github.com/llvm/llvm-project/pull/134171
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
@@ -101,6 +101,16 @@ constexpr vector smoothstep_vec_impl(vector
Min, vector Max,
#endif
}
+template constexpr vector lit_impl(T NDotL, T NDotH, T M) {
+ bool DiffuseCond = NDotL < 0;
+ T Diffuse = select(DiffuseCond, 0, NDotL);
+ vector Result = {1, Diffuse, 0, 1};
+ bool SpecularCond = or (DiffuseCond, (NDotH < 0));
kmpeng wrote:
Adding the manual space is making it fail :\ Should I tell clang-format to
ignore it? And how do I do that?
https://github.com/llvm/llvm-project/pull/134171
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
kmpeng wrote: I currently don't have `-fnative-half-type` enabled in that file, so yes this test case ends up having `float` input/return types (though it's not necessarily being converted in the function, if that's what you're asking). With `-fnative-half-type` enabled it doesn't convert to `float` https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL] Implement the `lit` intrinsic (PR #134171)
https://github.com/kmpeng edited https://github.com/llvm/llvm-project/pull/134171 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
