================
@@ -1,73 +1,151 @@
-// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
-// RUN: dxil-pc-shadermodel6.3-library %s -fnative-half-type \
-// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \
-// RUN: --check-prefixes=CHECK,NATIVE_HALF
-// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
-// RUN: dxil-pc-shadermodel6.3-library %s -emit-llvm -disable-llvm-passes \
-// RUN: -o - | FileCheck %s --check-prefixes=CHECK,NO_HALF
-
-// NATIVE_HALF: define noundef half @
-// NATIVE_HALF: call half @llvm.fabs.f16(half
-// NO_HALF: call float @llvm.fabs.f32(float
-// NATIVE_HALF: ret half
-// NO_HALF: ret float
-half test_length_half(half p0)
-{
- return length(p0);
-}
-// NATIVE_HALF: define noundef half @
-// NATIVE_HALF: %hlsl.length = call half @llvm.dx.length.v2f16
-// NO_HALF: %hlsl.length = call float @llvm.dx.length.v2f32(
-// NATIVE_HALF: ret half %hlsl.length
-// NO_HALF: ret float %hlsl.length
-half test_length_half2(half2 p0)
-{
- return length(p0);
-}
-// NATIVE_HALF: define noundef half @
-// NATIVE_HALF: %hlsl.length = call half @llvm.dx.length.v3f16
-// NO_HALF: %hlsl.length = call float @llvm.dx.length.v3f32(
-// NATIVE_HALF: ret half %hlsl.length
-// NO_HALF: ret float %hlsl.length
-half test_length_half3(half3 p0)
-{
- return length(p0);
-}
-// NATIVE_HALF: define noundef half @
-// NATIVE_HALF: %hlsl.length = call half @llvm.dx.length.v4f16
-// NO_HALF: %hlsl.length = call float @llvm.dx.length.v4f32(
-// NATIVE_HALF: ret half %hlsl.length
-// NO_HALF: ret float %hlsl.length
-half test_length_half4(half4 p0)
-{
- return length(p0);
-}
-
-// CHECK: define noundef float @
-// CHECK: call float @llvm.fabs.f32(float
-// CHECK: ret float
-float test_length_float(float p0)
-{
- return length(p0);
-}
-// CHECK: define noundef float @
-// CHECK: %hlsl.length = call float @llvm.dx.length.v2f32(
-// CHECK: ret float %hlsl.length
-float test_length_float2(float2 p0)
-{
- return length(p0);
-}
-// CHECK: define noundef float @
-// CHECK: %hlsl.length = call float @llvm.dx.length.v3f32(
-// CHECK: ret float %hlsl.length
-float test_length_float3(float3 p0)
-{
- return length(p0);
-}
-// CHECK: define noundef float @
-// CHECK: %hlsl.length = call float @llvm.dx.length.v4f32(
-// CHECK: ret float %hlsl.length
-float test_length_float4(float4 p0)
-{
- return length(p0);
-}
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
UTC_ARGS: --version 5
+// RUN: %clang_cc1 -finclude-default-header -triple \
+// RUN: dxil-pc-shadermodel6.3-library %s -fnative-half-type \
+// RUN: -emit-llvm -O2 -o - | FileCheck %s
+
+// CHECK-LABEL: define noundef half @_Z16test_length_halfDh(
+// CHECK-SAME: half noundef [[P0:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[ELT_ABS_I:%.*]] = tail call noundef half
@llvm.fabs.f16(half [[P0]])
+// CHECK-NEXT: ret half [[ELT_ABS_I]]
+//
+half test_length_half(half p0)
+{
+ return length(p0);
+}
+
+// CHECK-LABEL: define noundef half @_Z17test_length_half2Dv2_Dh(
+// CHECK-SAME: <2 x half> noundef [[P0:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul <2 x half> [[P0]], [[P0]]
+// CHECK-NEXT: [[VECEXT_I:%.*]] = extractelement <2 x half> [[MUL_I]], i64 0
+// CHECK-NEXT: [[VECEXT1_I:%.*]] = extractelement <2 x half> [[MUL_I]], i64
1
+// CHECK-NEXT: [[ADD_I:%.*]] = fadd half [[VECEXT1_I]], [[VECEXT_I]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call noundef half @llvm.sqrt.f16(half
[[ADD_I]])
+// CHECK-NEXT: ret half [[TMP0]]
+//
+half test_length_half2(half2 p0)
+{
+ return length(p0);
+}
+
+// CHECK-LABEL: define noundef half @_Z17test_length_half3Dv3_Dh(
+// CHECK-SAME: <3 x half> noundef [[P0:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul <3 x half> [[P0]], [[P0]]
+// CHECK-NEXT: [[VECEXT_I:%.*]] = extractelement <3 x half> [[MUL_I]], i64 0
+// CHECK-NEXT: [[VECEXT1_I:%.*]] = extractelement <3 x half> [[MUL_I]], i64
1
+// CHECK-NEXT: [[ADD_I:%.*]] = fadd half [[VECEXT_I]], [[VECEXT1_I]]
+// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <3 x half> [[MUL_I]],
i64 2
+// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd half [[ADD_I]], [[VECEXT1_I_1]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call noundef half @llvm.sqrt.f16(half
[[ADD_I_1]])
+// CHECK-NEXT: ret half [[TMP0]]
+//
+half test_length_half3(half3 p0)
+{
+ return length(p0);
+}
+
+// CHECK-LABEL: define noundef half @_Z17test_length_half4Dv4_Dh(
+// CHECK-SAME: <4 x half> noundef [[P0:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul <4 x half> [[P0]], [[P0]]
+// CHECK-NEXT: [[VECEXT_I:%.*]] = extractelement <4 x half> [[MUL_I]], i64 0
+// CHECK-NEXT: [[VECEXT1_I:%.*]] = extractelement <4 x half> [[MUL_I]], i64
1
+// CHECK-NEXT: [[ADD_I:%.*]] = fadd half [[VECEXT_I]], [[VECEXT1_I]]
+// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <4 x half> [[MUL_I]],
i64 2
+// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd half [[ADD_I]], [[VECEXT1_I_1]]
+// CHECK-NEXT: [[VECEXT1_I_2:%.*]] = extractelement <4 x half> [[MUL_I]],
i64 3
+// CHECK-NEXT: [[ADD_I_2:%.*]] = fadd half [[ADD_I_1]], [[VECEXT1_I_2]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call noundef half @llvm.sqrt.f16(half
[[ADD_I_2]])
+// CHECK-NEXT: ret half [[TMP0]]
+//
+half test_length_half4(half4 p0)
+{
+ return length(p0);
+}
+
+
+// CHECK-LABEL: define noundef float @_Z17test_length_floatf(
+// CHECK-SAME: float noundef [[P0:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[ELT_ABS_I:%.*]] = tail call noundef float
@llvm.fabs.f32(float [[P0]])
+// CHECK-NEXT: ret float [[ELT_ABS_I]]
+//
+float test_length_float(float p0)
+{
+ return length(p0);
+}
+
+// CHECK-LABEL: define noundef float @_Z18test_length_float2Dv2_f(
+// CHECK-SAME: <2 x float> noundef [[P0:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul <2 x float> [[P0]], [[P0]]
+// CHECK-NEXT: [[VECEXT_I:%.*]] = extractelement <2 x float> [[MUL_I]], i64 0
+// CHECK-NEXT: [[VECEXT1_I:%.*]] = extractelement <2 x float> [[MUL_I]],
i64 1
+// CHECK-NEXT: [[ADD_I:%.*]] = fadd float [[VECEXT1_I]], [[VECEXT_I]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call noundef float @llvm.sqrt.f32(float
[[ADD_I]])
+// CHECK-NEXT: ret float [[TMP0]]
+//
+float test_length_float2(float2 p0)
+{
+ return length(p0);
+}
+
+// CHECK-LABEL: define noundef float @_Z18test_length_float3Dv3_f(
+// CHECK-SAME: <3 x float> noundef [[P0:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul <3 x float> [[P0]], [[P0]]
+// CHECK-NEXT: [[VECEXT_I:%.*]] = extractelement <3 x float> [[MUL_I]], i64 0
+// CHECK-NEXT: [[VECEXT1_I:%.*]] = extractelement <3 x float> [[MUL_I]],
i64 1
+// CHECK-NEXT: [[ADD_I:%.*]] = fadd float [[VECEXT_I]], [[VECEXT1_I]]
+// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <3 x float> [[MUL_I]],
i64 2
+// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd float [[ADD_I]], [[VECEXT1_I_1]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call noundef float @llvm.sqrt.f32(float
[[ADD_I_1]])
+// CHECK-NEXT: ret float [[TMP0]]
+//
+float test_length_float3(float3 p0)
+{
+ return length(p0);
+}
+
+
+// CHECK-LABEL: define noundef float @_Z18test_length_float4Dv4_f(
+// CHECK-SAME: <4 x float> noundef [[P0:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul <4 x float> [[P0]], [[P0]]
+// CHECK-NEXT: [[VECEXT_I:%.*]] = extractelement <4 x float> [[MUL_I]], i64 0
+// CHECK-NEXT: [[VECEXT1_I:%.*]] = extractelement <4 x float> [[MUL_I]],
i64 1
+// CHECK-NEXT: [[ADD_I:%.*]] = fadd float [[VECEXT_I]], [[VECEXT1_I]]
+// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <4 x float> [[MUL_I]],
i64 2
+// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd float [[ADD_I]], [[VECEXT1_I_1]]
+// CHECK-NEXT: [[VECEXT1_I_2:%.*]] = extractelement <4 x float> [[MUL_I]],
i64 3
+// CHECK-NEXT: [[ADD_I_2:%.*]] = fadd float [[ADD_I_1]], [[VECEXT1_I_2]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call noundef float @llvm.sqrt.f32(float
[[ADD_I_2]])
+// CHECK-NEXT: ret float [[TMP0]]
+//
+float test_length_float4(float4 p0)
+{
+ return length(p0);
+}
+
+
+// CHECK-LABEL: define noundef float @_Z26test_length_float4_extractDv4_f(
+// CHECK-SAME: <4 x float> noundef [[P0:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul <4 x float> [[P0]], [[P0]]
+// CHECK-NEXT: [[VECEXT_I:%.*]] = extractelement <4 x float> [[MUL_I]], i64 0
+// CHECK-NEXT: [[VECEXT1_I:%.*]] = extractelement <4 x float> [[MUL_I]],
i64 1
+// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <4 x float> [[MUL_I]],
i64 2
+// CHECK-NEXT: [[VECEXT1_I_2:%.*]] = extractelement <4 x float> [[MUL_I]],
i64 3
+// CHECK-NEXT: [[ADD_I12:%.*]] = fadd float [[VECEXT_I]], [[VECEXT1_I]]
+// CHECK-NEXT: [[ADD_I12_1:%.*]] = fadd float [[ADD_I12]], [[VECEXT1_I_1]]
+// CHECK-NEXT: [[ADD_I12_2:%.*]] = fadd float [[ADD_I12_1]], [[VECEXT1_I_2]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call noundef float @llvm.sqrt.f32(float
[[ADD_I12_2]])
+// CHECK-NEXT: [[ADD:%.*]] = fadd float [[TMP0]], [[TMP0]]
+// CHECK-NEXT: ret float [[ADD]]
+//
+float test_length_float4_extract(float4 p0)
+{
+ return length(p0) + length(p0);
----------------
farzonl wrote:
Adding this case because made me realize that we need `O2` so that we don't
have duplicate extracts and adds when calling length twice. The vec_reduce_add
implementation used `O1` and didn't have this problem so I think this might be
a side effect of not being able to do a constexpr.
https://github.com/llvm/llvm-project/pull/121611
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
