https://github.com/farzonl updated
https://github.com/llvm/llvm-project/pull/122337
>From a7323e97f09db476e622bee9800e1b3e15441beb Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlo...@microsoft.com>
Date: Thu, 9 Jan 2025 13:11:52 -0500
Subject: [PATCH 1/3] [HLSL] Move length support out of the DirectX Backend
(#121611)
## Changes
- Delete DirectX length intrinsic
- Delete HLSL length lang builtin
- Implement length algorithm entirely in the header.
## History
- In the past if an HLSL intrinsic lowered to either a spirv op code or
a DXIL opcode we represented it with intrinsics
## Why we are moving away?
- To make HLSL apis more portable the team decided that it makes sense
for some intrinsics to be defined only in the header.
- Since there tends to be more SPIRV opcodes than DXIL opcodes the plan
is to support SPIRV opcodes either with target specific builtins or via
pattern matching.
---
clang/include/clang/Basic/Builtins.td | 6 -
clang/lib/CodeGen/CGBuiltin.cpp | 14 --
clang/lib/CodeGen/CGHLSLRuntime.h | 1 -
clang/lib/Headers/hlsl/hlsl_detail.h | 23 ++
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 29 +--
clang/lib/Sema/SemaHLSL.cpp | 18 --
clang/test/CodeGenHLSL/builtins/length.hlsl | 203 +++++++++++-------
.../test/SemaHLSL/BuiltIns/length-errors.hlsl | 51 +++--
llvm/include/llvm/IR/IntrinsicsDirectX.td | 1 -
.../Target/DirectX/DXILIntrinsicExpansion.cpp | 30 ---
llvm/test/CodeGen/DirectX/length.ll | 116 ----------
llvm/test/CodeGen/DirectX/length_error.ll | 10 -
.../DirectX/length_invalid_intrinsic_error.ll | 10 -
.../length_invalid_intrinsic_error_scalar.ll | 10 -
14 files changed, 198 insertions(+), 324 deletions(-)
delete mode 100644 llvm/test/CodeGen/DirectX/length.ll
delete mode 100644 llvm/test/CodeGen/DirectX/length_error.ll
delete mode 100644 llvm/test/CodeGen/DirectX/length_invalid_intrinsic_error.ll
delete mode 100644
llvm/test/CodeGen/DirectX/length_invalid_intrinsic_error_scalar.ll
diff --git a/clang/include/clang/Basic/Builtins.td
b/clang/include/clang/Basic/Builtins.td
index 468c16050e2bf0..f4216bd01a0743 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4865,12 +4865,6 @@ def HLSLIsinf : LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
-def HLSLLength : LangBuiltin<"HLSL_LANG"> {
- let Spellings = ["__builtin_hlsl_length"];
- let Attributes = [NoThrow, Const];
- let Prototype = "void(...)";
-}
-
def HLSLLerp : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_lerp"];
let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index ca03fb665d423d..2b091976699961 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19334,20 +19334,6 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned
BuiltinID,
/*ReturnType=*/X->getType(), CGM.getHLSLRuntime().getLerpIntrinsic(),
ArrayRef<Value *>{X, Y, S}, nullptr, "hlsl.lerp");
}
- case Builtin::BI__builtin_hlsl_length: {
- Value *X = EmitScalarExpr(E->getArg(0));
-
- assert(E->getArg(0)->getType()->hasFloatingRepresentation() &&
- "length operand must have a float representation");
- // if the operand is a scalar, we can use the fabs llvm intrinsic directly
- if (!E->getArg(0)->getType()->isVectorType())
- return EmitFAbs(*this, X);
-
- return Builder.CreateIntrinsic(
- /*ReturnType=*/X->getType()->getScalarType(),
- CGM.getHLSLRuntime().getLengthIntrinsic(), ArrayRef<Value *>{X},
- nullptr, "hlsl.length");
- }
case Builtin::BI__builtin_hlsl_normalize: {
Value *X = EmitScalarExpr(E->getArg(0));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 46e472f0aae213..00e110e8e6fa27 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -77,7 +77,6 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(Cross, cross)
GENERATE_HLSL_INTRINSIC_FUNCTION(Degrees, degrees)
GENERATE_HLSL_INTRINSIC_FUNCTION(Frac, frac)
- GENERATE_HLSL_INTRINSIC_FUNCTION(Length, length)
GENERATE_HLSL_INTRINSIC_FUNCTION(Lerp, lerp)
GENERATE_HLSL_INTRINSIC_FUNCTION(Normalize, normalize)
GENERATE_HLSL_INTRINSIC_FUNCTION(Rsqrt, rsqrt)
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h
b/clang/lib/Headers/hlsl/hlsl_detail.h
index 8d5fd941331531..392075d276b188 100644
--- a/clang/lib/Headers/hlsl/hlsl_detail.h
+++ b/clang/lib/Headers/hlsl/hlsl_detail.h
@@ -13,6 +13,14 @@ namespace hlsl {
namespace __detail {
+template <typename T, typename U> struct is_same {
+ static const bool value = false;
+};
+
+template <typename T> struct is_same<T, T> {
+ static const bool value = true;
+};
+
template <bool B, typename T> struct enable_if {};
template <typename T> struct enable_if<true, T> {
@@ -33,6 +41,21 @@ constexpr enable_if_t<sizeof(U) == sizeof(T), U> bit_cast(T
F) {
return __builtin_bit_cast(U, F);
}
+template <typename T>
+constexpr enable_if_t<is_same<float, T>::value || is_same<half, T>::value, T>
+length_impl(T X) {
+ return __builtin_elementwise_abs(X);
+}
+
+template <typename T, int N>
+enable_if_t<is_same<float, T>::value || is_same<half, T>::value, T>
+length_vec_impl(vector<T, N> X) {
+ vector<T, N> XSquared = X * X;
+ T XSquaredSum = XSquared[0];
+ [unroll] for (int i = 1; i < N; ++i) XSquaredSum += XSquared[i];
+ return __builtin_elementwise_sqrt(XSquaredSum);
+}
+
} // namespace __detail
} // namespace hlsl
#endif //_HLSL_HLSL_DETAILS_H_
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index b745997f1d5a2b..cf287e598f76ba 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -1297,27 +1297,16 @@ float4 lerp(float4, float4, float4);
///
/// Length is based on the following formula: sqrt(x[0]^2 + x[1]^2 + ...).
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_length)
-half length(half);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_length)
-half length(half2);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_length)
-half length(half3);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_length)
-half length(half4);
+const inline half length(half X) { return __detail::length_impl(X); }
+const inline float length(float X) { return __detail::length_impl(X); }
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_length)
-float length(float);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_length)
-float length(float2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_length)
-float length(float3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_length)
-float length(float4);
+template <int N> const inline half length(vector<half, N> X) {
+ return __detail::length_vec_impl(X);
+}
+
+template <int N> const inline float length(vector<float, N> X) {
+ return __detail::length_vec_impl(X);
+}
//===----------------------------------------------------------------------===//
// log builtins
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 83c38fad2df686..65ddee05a21512 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2112,24 +2112,6 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
return true;
break;
}
- case Builtin::BI__builtin_hlsl_length: {
- if (CheckFloatOrHalfRepresentations(&SemaRef, TheCall))
- return true;
- if (SemaRef.checkArgCount(TheCall, 1))
- return true;
-
- ExprResult A = TheCall->getArg(0);
- QualType ArgTyA = A.get()->getType();
- QualType RetTy;
-
- if (auto *VTy = ArgTyA->getAs<VectorType>())
- RetTy = VTy->getElementType();
- else
- RetTy = TheCall->getArg(0)->getType();
-
- TheCall->setType(RetTy);
- break;
- }
case Builtin::BI__builtin_hlsl_mad: {
if (SemaRef.checkArgCount(TheCall, 3))
return true;
diff --git a/clang/test/CodeGenHLSL/builtins/length.hlsl
b/clang/test/CodeGenHLSL/builtins/length.hlsl
index a24f01d275440a..ecf2edcf0b05f1 100644
--- a/clang/test/CodeGenHLSL/builtins/length.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/length.hlsl
@@ -1,73 +1,130 @@
-// 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 nofpclass(nan inf) half @
-// NATIVE_HALF: call reassoc nnan ninf nsz arcp afn half @llvm.fabs.f16(half
-// NO_HALF: call reassoc nnan ninf nsz arcp afn 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 nofpclass(nan inf) half @
-// NATIVE_HALF: %hlsl.length = call reassoc nnan ninf nsz arcp afn half
@llvm.dx.length.v2f16
-// NO_HALF: %hlsl.length = call reassoc nnan ninf nsz arcp afn 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 nofpclass(nan inf) half @
-// NATIVE_HALF: %hlsl.length = call reassoc nnan ninf nsz arcp afn half
@llvm.dx.length.v3f16
-// NO_HALF: %hlsl.length = call reassoc nnan ninf nsz arcp afn 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 nofpclass(nan inf) half @
-// NATIVE_HALF: %hlsl.length = call reassoc nnan ninf nsz arcp afn half
@llvm.dx.length.v4f16
-// NO_HALF: %hlsl.length = call reassoc nnan ninf nsz arcp afn 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 nofpclass(nan inf) float @
-// CHECK: call reassoc nnan ninf nsz arcp afn float @llvm.fabs.f32(float
-// CHECK: ret float
-float test_length_float(float p0)
-{
- return length(p0);
-}
-// CHECK: define noundef nofpclass(nan inf) float @
-// CHECK: %hlsl.length = call reassoc nnan ninf nsz arcp afn float
@llvm.dx.length.v2f32(
-// CHECK: ret float %hlsl.length
-float test_length_float2(float2 p0)
-{
- return length(p0);
-}
-// CHECK: define noundef nofpclass(nan inf) float @
-// CHECK: %hlsl.length = call reassoc nnan ninf nsz arcp afn float
@llvm.dx.length.v3f32(
-// CHECK: ret float %hlsl.length
-float test_length_float3(float3 p0)
-{
- return length(p0);
-}
-// CHECK: define noundef nofpclass(nan inf) float @
-// CHECK: %hlsl.length = call reassoc nnan ninf nsz arcp afn 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 -O1 -o - | FileCheck %s
+
+// CHECK-LABEL: define noundef nofpclass(nan inf) half @_Z16test_length_halfDh(
+// CHECK-SAME: half noundef nofpclass(nan inf) [[P0:%.*]]) local_unnamed_addr
#[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[ELT_ABS_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
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 nofpclass(nan inf) half
@_Z17test_length_half2Dv2_Dh(
+// CHECK-SAME: <2 x half> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn half
[[VECEXT_I]], [[VECEXT1_I]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
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 nofpclass(nan inf) half
@_Z17test_length_half3Dv3_Dh(
+// CHECK-SAME: <3 x half> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn half
[[VECEXT1_I]], [[VECEXT_I]]
+// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <3 x half> [[MUL_I]],
i64 2
+// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd reassoc nnan ninf nsz arcp afn half
[[VECEXT1_I_1]], [[ADD_I]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
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 nofpclass(nan inf) half
@_Z17test_length_half4Dv4_Dh(
+// CHECK-SAME: <4 x half> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn half
[[VECEXT1_I]], [[VECEXT_I]]
+// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <4 x half> [[MUL_I]],
i64 2
+// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd reassoc nnan ninf nsz arcp afn half
[[VECEXT1_I_1]], [[ADD_I]]
+// CHECK-NEXT: [[VECEXT1_I_2:%.*]] = extractelement <4 x half> [[MUL_I]],
i64 3
+// CHECK-NEXT: [[ADD_I_2:%.*]] = fadd reassoc nnan ninf nsz arcp afn half
[[VECEXT1_I_2]], [[ADD_I_1]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
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 nofpclass(nan inf) float
@_Z17test_length_floatf(
+// CHECK-SAME: float noundef nofpclass(nan inf) [[P0:%.*]]) local_unnamed_addr
#[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[ELT_ABS_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
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 nofpclass(nan inf) float
@_Z18test_length_float2Dv2_f(
+// CHECK-SAME: <2 x float> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn float
[[VECEXT_I]], [[VECEXT1_I]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
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 nofpclass(nan inf) float
@_Z18test_length_float3Dv3_f(
+// CHECK-SAME: <3 x float> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn float
[[VECEXT1_I]], [[VECEXT_I]]
+// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <3 x float> [[MUL_I]],
i64 2
+// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd reassoc nnan ninf nsz arcp afn float
[[VECEXT1_I_1]], [[ADD_I]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
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 nofpclass(nan inf) float
@_Z18test_length_float4Dv4_f(
+// CHECK-SAME: <4 x float> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn float
[[VECEXT1_I]], [[VECEXT_I]]
+// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <4 x float> [[MUL_I]],
i64 2
+// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd reassoc nnan ninf nsz arcp afn float
[[VECEXT1_I_1]], [[ADD_I]]
+// CHECK-NEXT: [[VECEXT1_I_2:%.*]] = extractelement <4 x float> [[MUL_I]],
i64 3
+// CHECK-NEXT: [[ADD_I_2:%.*]] = fadd reassoc nnan ninf nsz arcp afn float
[[VECEXT1_I_2]], [[ADD_I_1]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef float @llvm.sqrt.f32(float [[ADD_I_2]])
+// CHECK-NEXT: ret float [[TMP0]]
+//
+float test_length_float4(float4 p0)
+{
+ return length(p0);
+}
diff --git a/clang/test/SemaHLSL/BuiltIns/length-errors.hlsl
b/clang/test/SemaHLSL/BuiltIns/length-errors.hlsl
index 281faada6f5e94..a191f0419fbbaf 100644
--- a/clang/test/SemaHLSL/BuiltIns/length-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/length-errors.hlsl
@@ -1,32 +1,53 @@
-// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.6-library %s -fnative-half-type -disable-llvm-passes
-verify -verify-ignore-unexpected
-
+// RUN: %clang_cc1 -finclude-default-header -triple
dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only
-disable-llvm-passes -verify
void test_too_few_arg()
{
- return __builtin_hlsl_length();
- // expected-error@-1 {{too few arguments to function call, expected 1, have
0}}
+ return length();
+ // expected-error@-1 {{no matching function for call to 'length'}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires single argument 'X', but no arguments were provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires single argument 'X', but no arguments were provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires single argument 'X', but no arguments were provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires single argument 'X', but no arguments were provided}}
}
void test_too_many_arg(float2 p0)
{
- return __builtin_hlsl_length(p0, p0);
- // expected-error@-1 {{too many arguments to function call, expected 1, have
2}}
+ return length(p0, p0);
+ // expected-error@-1 {{no matching function for call to 'length'}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires single argument 'X', but 2 arguments were provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires single argument 'X', but 2 arguments were provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function not viable:
requires single argument 'X', but 2 arguments were provided}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not
viable: requires single argument 'X', but 2 arguments were provided}}
+}
+
+float double_to_float_type(double p0) {
+ return length(p0);
+ // expected-error@-1 {{call to 'length' is ambiguous}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function}}
}
-bool builtin_bool_to_float_type_promotion(bool p1)
+
+float bool_to_float_type_promotion(bool p1)
{
- return __builtin_hlsl_length(p1);
- // expected-error@-1 {passing 'bool' to parameter of incompatible type
'float'}}
+ return length(p1);
+ // expected-error@-1 {{call to 'length' is ambiguous}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function}}
}
-bool builtin_length_int_to_float_promotion(int p1)
+float length_int_to_float_promotion(int p1)
{
- return __builtin_hlsl_length(p1);
- // expected-error@-1 {{passing 'int' to parameter of incompatible type
'float'}}
+ return length(p1);
+ // expected-error@-1 {{call to 'length' is ambiguous}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function}}
}
-bool2 builtin_length_int2_to_float2_promotion(int2 p1)
+float2 length_int2_to_float2_promotion(int2 p1)
{
- return __builtin_hlsl_length(p1);
- // expected-error@-1 {{passing 'int2' (aka 'vector<int, 2>') to parameter of
incompatible type '__attribute__((__vector_size__(2 * sizeof(float)))) float'
(vector of 2 'float' values)}}
+ return length(p1);
+ // expected-error@-1 {{call to 'length' is ambiguous}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function}}
+ // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function}}
}
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td
b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index 3b1d1a88e01a8b..ef48af5b42dbf1 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -93,7 +93,6 @@ def int_dx_isinf :
DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i1
def int_dx_lerp : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty,
LLVMMatchType<0>,LLVMMatchType<0>],
[IntrNoMem]>;
-def int_dx_length : DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
def int_dx_imad : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>,
LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_dx_umad : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>,
LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
def int_dx_normalize : DefaultAttrsIntrinsic<[LLVMMatchType<0>],
[llvm_anyfloat_ty], [IntrNoMem]>;
diff --git a/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
b/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
index 51dc3025f0c379..cf142806bb1df6 100644
--- a/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
+++ b/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
@@ -57,7 +57,6 @@ static bool isIntrinsicExpansion(Function &F) {
case Intrinsic::dx_nclamp:
case Intrinsic::dx_degrees:
case Intrinsic::dx_lerp:
- case Intrinsic::dx_length:
case Intrinsic::dx_normalize:
case Intrinsic::dx_fdot:
case Intrinsic::dx_sdot:
@@ -292,32 +291,6 @@ static Value *expandAnyOrAllIntrinsic(CallInst *Orig,
return Result;
}
-static Value *expandLengthIntrinsic(CallInst *Orig) {
- Value *X = Orig->getOperand(0);
- IRBuilder<> Builder(Orig);
- Type *Ty = X->getType();
- Type *EltTy = Ty->getScalarType();
-
- // Though dx.length does work on scalar type, we can optimize it to just emit
- // fabs, in CGBuiltin.cpp. We shouldn't see a scalar type here because
- // CGBuiltin.cpp should have emitted a fabs call.
- Value *Elt = Builder.CreateExtractElement(X, (uint64_t)0);
- auto *XVec = dyn_cast<FixedVectorType>(Ty);
- unsigned XVecSize = XVec->getNumElements();
- if (!(Ty->isVectorTy() && XVecSize > 1))
- report_fatal_error(Twine("Invalid input type for length intrinsic"),
- /* gen_crash_diag=*/false);
-
- Value *Sum = Builder.CreateFMul(Elt, Elt);
- for (unsigned I = 1; I < XVecSize; I++) {
- Elt = Builder.CreateExtractElement(X, I);
- Value *Mul = Builder.CreateFMul(Elt, Elt);
- Sum = Builder.CreateFAdd(Sum, Mul);
- }
- return Builder.CreateIntrinsic(EltTy, Intrinsic::sqrt, ArrayRef<Value
*>{Sum},
- nullptr, "elt.sqrt");
-}
-
static Value *expandLerpIntrinsic(CallInst *Orig) {
Value *X = Orig->getOperand(0);
Value *Y = Orig->getOperand(1);
@@ -589,9 +562,6 @@ static bool expandIntrinsic(Function &F, CallInst *Orig) {
case Intrinsic::dx_lerp:
Result = expandLerpIntrinsic(Orig);
break;
- case Intrinsic::dx_length:
- Result = expandLengthIntrinsic(Orig);
- break;
case Intrinsic::dx_normalize:
Result = expandNormalizeIntrinsic(Orig);
break;
diff --git a/llvm/test/CodeGen/DirectX/length.ll
b/llvm/test/CodeGen/DirectX/length.ll
deleted file mode 100644
index fc5868a7f6e82c..00000000000000
--- a/llvm/test/CodeGen/DirectX/length.ll
+++ /dev/null
@@ -1,116 +0,0 @@
-; RUN: opt -S -dxil-intrinsic-expansion < %s | FileCheck %s
--check-prefixes=CHECK,EXPCHECK
-; RUN: opt -S -dxil-intrinsic-expansion -dxil-op-lower
-mtriple=dxil-pc-shadermodel6.3-library < %s | FileCheck %s
--check-prefixes=CHECK,DOPCHECK
-
-; Make sure dxil operation function calls for length are generated for
half/float.
-
-declare half @llvm.fabs.f16(half)
-declare half @llvm.dx.length.v2f16(<2 x half>)
-declare half @llvm.dx.length.v3f16(<3 x half>)
-declare half @llvm.dx.length.v4f16(<4 x half>)
-
-declare float @llvm.fabs.f32(float)
-declare float @llvm.dx.length.v2f32(<2 x float>)
-declare float @llvm.dx.length.v3f32(<3 x float>)
-declare float @llvm.dx.length.v4f32(<4 x float>)
-
-define noundef half @test_length_half2(<2 x half> noundef %p0) {
-entry:
- ; CHECK: extractelement <2 x half> %{{.*}}, i64 0
- ; CHECK: fmul half %{{.*}}, %{{.*}}
- ; CHECK: extractelement <2 x half> %{{.*}}, i64 1
- ; CHECK: fmul half %{{.*}}, %{{.*}}
- ; CHECK: fadd half %{{.*}}, %{{.*}}
- ; EXPCHECK: call half @llvm.sqrt.f16(half %{{.*}})
- ; DOPCHECK: call half @dx.op.unary.f16(i32 24, half %{{.*}})
-
- %hlsl.length = call half @llvm.dx.length.v2f16(<2 x half> %p0)
- ret half %hlsl.length
-}
-
-define noundef half @test_length_half3(<3 x half> noundef %p0) {
-entry:
- ; CHECK: extractelement <3 x half> %{{.*}}, i64 0
- ; CHECK: fmul half %{{.*}}, %{{.*}}
- ; CHECK: extractelement <3 x half> %{{.*}}, i64 1
- ; CHECK: fmul half %{{.*}}, %{{.*}}
- ; CHECK: fadd half %{{.*}}, %{{.*}}
- ; CHECK: extractelement <3 x half> %{{.*}}, i64 2
- ; CHECK: fmul half %{{.*}}, %{{.*}}
- ; CHECK: fadd half %{{.*}}, %{{.*}}
- ; EXPCHECK: call half @llvm.sqrt.f16(half %{{.*}})
- ; DOPCHECK: call half @dx.op.unary.f16(i32 24, half %{{.*}})
-
- %hlsl.length = call half @llvm.dx.length.v3f16(<3 x half> %p0)
- ret half %hlsl.length
-}
-
-define noundef half @test_length_half4(<4 x half> noundef %p0) {
-entry:
- ; CHECK: extractelement <4 x half> %{{.*}}, i64 0
- ; CHECK: fmul half %{{.*}}, %{{.*}}
- ; CHECK: extractelement <4 x half> %{{.*}}, i64 1
- ; CHECK: fmul half %{{.*}}, %{{.*}}
- ; CHECK: fadd half %{{.*}}, %{{.*}}
- ; CHECK: extractelement <4 x half> %{{.*}}, i64 2
- ; CHECK: fmul half %{{.*}}, %{{.*}}
- ; CHECK: fadd half %{{.*}}, %{{.*}}
- ; CHECK: extractelement <4 x half> %{{.*}}, i64 3
- ; CHECK: fmul half %{{.*}}, %{{.*}}
- ; CHECK: fadd half %{{.*}}, %{{.*}}
- ; EXPCHECK: call half @llvm.sqrt.f16(half %{{.*}})
- ; DOPCHECK: call half @dx.op.unary.f16(i32 24, half %{{.*}})
-
- %hlsl.length = call half @llvm.dx.length.v4f16(<4 x half> %p0)
- ret half %hlsl.length
-}
-
-define noundef float @test_length_float2(<2 x float> noundef %p0) {
-entry:
- ; CHECK: extractelement <2 x float> %{{.*}}, i64 0
- ; CHECK: fmul float %{{.*}}, %{{.*}}
- ; CHECK: extractelement <2 x float> %{{.*}}, i64 1
- ; CHECK: fmul float %{{.*}}, %{{.*}}
- ; CHECK: fadd float %{{.*}}, %{{.*}}
- ; EXPCHECK: call float @llvm.sqrt.f32(float %{{.*}})
- ; DOPCHECK: call float @dx.op.unary.f32(i32 24, float %{{.*}})
-
- %hlsl.length = call float @llvm.dx.length.v2f32(<2 x float> %p0)
- ret float %hlsl.length
-}
-
-define noundef float @test_length_float3(<3 x float> noundef %p0) {
-entry:
- ; CHECK: extractelement <3 x float> %{{.*}}, i64 0
- ; CHECK: fmul float %{{.*}}, %{{.*}}
- ; CHECK: extractelement <3 x float> %{{.*}}, i64 1
- ; CHECK: fmul float %{{.*}}, %{{.*}}
- ; CHECK: fadd float %{{.*}}, %{{.*}}
- ; CHECK: extractelement <3 x float> %{{.*}}, i64 2
- ; CHECK: fmul float %{{.*}}, %{{.*}}
- ; CHECK: fadd float %{{.*}}, %{{.*}}
- ; EXPCHECK: call float @llvm.sqrt.f32(float %{{.*}})
- ; DOPCHECK: call float @dx.op.unary.f32(i32 24, float %{{.*}})
-
- %hlsl.length = call float @llvm.dx.length.v3f32(<3 x float> %p0)
- ret float %hlsl.length
-}
-
-define noundef float @test_length_float4(<4 x float> noundef %p0) {
-entry:
- ; CHECK: extractelement <4 x float> %{{.*}}, i64 0
- ; CHECK: fmul float %{{.*}}, %{{.*}}
- ; CHECK: extractelement <4 x float> %{{.*}}, i64 1
- ; CHECK: fmul float %{{.*}}, %{{.*}}
- ; CHECK: fadd float %{{.*}}, %{{.*}}
- ; CHECK: extractelement <4 x float> %{{.*}}, i64 2
- ; CHECK: fmul float %{{.*}}, %{{.*}}
- ; CHECK: fadd float %{{.*}}, %{{.*}}
- ; CHECK: extractelement <4 x float> %{{.*}}, i64 3
- ; CHECK: fmul float %{{.*}}, %{{.*}}
- ; CHECK: fadd float %{{.*}}, %{{.*}}
- ; EXPCHECK: call float @llvm.sqrt.f32(float %{{.*}})
- ; DOPCHECK: call float @dx.op.unary.f32(i32 24, float %{{.*}})
-
- %hlsl.length = call float @llvm.dx.length.v4f32(<4 x float> %p0)
- ret float %hlsl.length
-}
diff --git a/llvm/test/CodeGen/DirectX/length_error.ll
b/llvm/test/CodeGen/DirectX/length_error.ll
deleted file mode 100644
index 143b41fc506e1d..00000000000000
--- a/llvm/test/CodeGen/DirectX/length_error.ll
+++ /dev/null
@@ -1,10 +0,0 @@
-; RUN: not opt -S -dxil-intrinsic-expansion -dxil-op-lower
-mtriple=dxil-pc-shadermodel6.3-library %s 2>&1 | FileCheck %s
-
-; DXIL operation length does not support double overload type
-; CHECK: Cannot create Sqrt operation: Invalid overload type
-
-define noundef double @test_length_double2(<2 x double> noundef %p0) {
-entry:
- %hlsl.length = call double @llvm.dx.length.v2f32(<2 x double> %p0)
- ret double %hlsl.length
-}
diff --git a/llvm/test/CodeGen/DirectX/length_invalid_intrinsic_error.ll
b/llvm/test/CodeGen/DirectX/length_invalid_intrinsic_error.ll
deleted file mode 100644
index f722de2f9029e7..00000000000000
--- a/llvm/test/CodeGen/DirectX/length_invalid_intrinsic_error.ll
+++ /dev/null
@@ -1,10 +0,0 @@
-; RUN: not opt -S -dxil-intrinsic-expansion -dxil-op-lower
-mtriple=dxil-pc-shadermodel6.3-library %s 2>&1 | FileCheck %s
-
-; DXIL operation length does not support 1-element vector types.
-; CHECK: LLVM ERROR: Invalid input type for length intrinsic
-
-define noundef float @test_length_float(<1 x float> noundef %p0) {
-entry:
- %hlsl.length = call float @llvm.dx.length.v1f32(<1 x float> %p0)
- ret float %hlsl.length
-}
diff --git a/llvm/test/CodeGen/DirectX/length_invalid_intrinsic_error_scalar.ll
b/llvm/test/CodeGen/DirectX/length_invalid_intrinsic_error_scalar.ll
deleted file mode 100644
index ac3a0513eb6b27..00000000000000
--- a/llvm/test/CodeGen/DirectX/length_invalid_intrinsic_error_scalar.ll
+++ /dev/null
@@ -1,10 +0,0 @@
-; RUN: not opt -S -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-library %s
2>&1 | FileCheck %s
-
-; DXIL operation length does not support scalar types
-; CHECK: error: invalid intrinsic signature
-
-define noundef float @test_length_float(float noundef %p0) {
-entry:
- %hlsl.length = call float @llvm.dx.length.f32(float %p0)
- ret float %hlsl.length
-}
>From 7173a486d82a3959fc6157a58488b10ece492821 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlo...@microsoft.com>
Date: Thu, 9 Jan 2025 13:55:33 -0500
Subject: [PATCH 2/3] [HLSL] Add 6.2 SM on half availability for length
intrinsic - attribute keeps getting deleted on bad rebases
---
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index cf287e598f76ba..f4470d5de61234 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -1297,10 +1297,13 @@ float4 lerp(float4, float4, float4);
///
/// Length is based on the following formula: sqrt(x[0]^2 + x[1]^2 + ...).
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
const inline half length(half X) { return __detail::length_impl(X); }
const inline float length(float X) { return __detail::length_impl(X); }
-template <int N> const inline half length(vector<half, N> X) {
+template <int N>
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline half length(vector<half, N> X) {
return __detail::length_vec_impl(X);
}
>From 15620e54d634dfca5909bee749236ca05a5a7ded Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlo...@microsoft.com>
Date: Fri, 10 Jan 2025 12:32:08 -0500
Subject: [PATCH 3/3] replace loop unroll length algo with one that doesn't
have to unroll to fix test break.
---
clang/lib/Headers/hlsl/hlsl_detail.h | 7 +-
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 2 +-
clang/test/CodeGenHLSL/builtins/length.hlsl | 93 +++++++++------------
3 files changed, 42 insertions(+), 60 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h
b/clang/lib/Headers/hlsl/hlsl_detail.h
index 392075d276b188..33d394f7883a6e 100644
--- a/clang/lib/Headers/hlsl/hlsl_detail.h
+++ b/clang/lib/Headers/hlsl/hlsl_detail.h
@@ -48,12 +48,9 @@ length_impl(T X) {
}
template <typename T, int N>
-enable_if_t<is_same<float, T>::value || is_same<half, T>::value, T>
+constexpr enable_if_t<is_same<float, T>::value || is_same<half, T>::value, T>
length_vec_impl(vector<T, N> X) {
- vector<T, N> XSquared = X * X;
- T XSquaredSum = XSquared[0];
- [unroll] for (int i = 1; i < N; ++i) XSquaredSum += XSquared[i];
- return __builtin_elementwise_sqrt(XSquaredSum);
+ return __builtin_elementwise_sqrt(__builtin_hlsl_dot(X, X));
}
} // namespace __detail
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index f4470d5de61234..e5ff7531d25d91 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -1301,7 +1301,7 @@ _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
const inline half length(half X) { return __detail::length_impl(X); }
const inline float length(float X) { return __detail::length_impl(X); }
-template <int N>
+template <int N>
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
const inline half length(vector<half, N> X) {
return __detail::length_vec_impl(X);
diff --git a/clang/test/CodeGenHLSL/builtins/length.hlsl
b/clang/test/CodeGenHLSL/builtins/length.hlsl
index ecf2edcf0b05f1..fcf3ee76ba5bbd 100644
--- a/clang/test/CodeGenHLSL/builtins/length.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/length.hlsl
@@ -1,44 +1,48 @@
-// 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 -O1 -o - | FileCheck %s
+// RUN: -emit-llvm -O1 -o - | FileCheck %s --check-prefixes=CHECK,DXCHECK \
+// RUN: -DTARGET=dx
-// CHECK-LABEL: define noundef nofpclass(nan inf) half @_Z16test_length_halfDh(
+// RUN: %clang_cc1 -finclude-default-header -triple \
+// RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \
+// RUN: -emit-llvm -O1 -o - | FileCheck %s --check-prefixes=CHECK,SPVCHECK \
+// RUN: -DTARGET=spv
+
+
+// SPVCHECK-LABEL: define spir_func noundef nofpclass(nan inf) half
@_Z16test_length_halfDh(
+// DXCHECK-LABEL: define noundef nofpclass(nan inf) half
@_Z16test_length_halfDh(
// CHECK-SAME: half noundef nofpclass(nan inf) [[P0:%.*]]) local_unnamed_addr
#[[ATTR0:[0-9]+]] {
// CHECK-NEXT: [[ENTRY:.*:]]
// CHECK-NEXT: [[ELT_ABS_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
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 nofpclass(nan inf) half
@_Z17test_length_half2Dv2_Dh(
+// SPVCHECK-LABEL: define spir_func noundef nofpclass(nan inf) half
@_Z17test_length_half2Dv2_Dh(
+// DXCHECK-LABEL: define noundef nofpclass(nan inf) half
@_Z17test_length_half2Dv2_Dh(
// CHECK-SAME: <2 x half> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn half
[[VECEXT_I]], [[VECEXT1_I]]
-// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef half @llvm.sqrt.f16(half [[ADD_I]])
+// CHECK-NEXT: [[HLSL_DOT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn half @llvm.[[TARGET]].fdot.v2f16(<2 x half> [[P0]], <2 x half> [[P0]])
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef half @llvm.sqrt.f16(half [[HLSL_DOT_I]])
// CHECK-NEXT: ret half [[TMP0]]
//
+
+
half test_length_half2(half2 p0)
{
return length(p0);
}
-// CHECK-LABEL: define noundef nofpclass(nan inf) half
@_Z17test_length_half3Dv3_Dh(
+// SPVCHECK-LABEL: define spir_func noundef nofpclass(nan inf) half
@_Z17test_length_half3Dv3_Dh(
+// DXCHECK-LABEL: define noundef nofpclass(nan inf) half
@_Z17test_length_half3Dv3_Dh(
// CHECK-SAME: <3 x half> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn half
[[VECEXT1_I]], [[VECEXT_I]]
-// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <3 x half> [[MUL_I]],
i64 2
-// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd reassoc nnan ninf nsz arcp afn half
[[VECEXT1_I_1]], [[ADD_I]]
-// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef half @llvm.sqrt.f16(half [[ADD_I_1]])
+// CHECK-NEXT: [[HLSL_DOT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn half @llvm.[[TARGET]].fdot.v3f16(<3 x half> [[P0]], <3 x half> [[P0]])
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef half @llvm.sqrt.f16(half [[HLSL_DOT_I]])
// CHECK-NEXT: ret half [[TMP0]]
//
half test_length_half3(half3 p0)
@@ -46,18 +50,12 @@ half test_length_half3(half3 p0)
return length(p0);
}
-// CHECK-LABEL: define noundef nofpclass(nan inf) half
@_Z17test_length_half4Dv4_Dh(
+// SPVCHECK-LABEL: define spir_func noundef nofpclass(nan inf) half
@_Z17test_length_half4Dv4_Dh(
+// DXCHECK-LABEL: define noundef nofpclass(nan inf) half
@_Z17test_length_half4Dv4_Dh(
// CHECK-SAME: <4 x half> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn half
[[VECEXT1_I]], [[VECEXT_I]]
-// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <4 x half> [[MUL_I]],
i64 2
-// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd reassoc nnan ninf nsz arcp afn half
[[VECEXT1_I_1]], [[ADD_I]]
-// CHECK-NEXT: [[VECEXT1_I_2:%.*]] = extractelement <4 x half> [[MUL_I]],
i64 3
-// CHECK-NEXT: [[ADD_I_2:%.*]] = fadd reassoc nnan ninf nsz arcp afn half
[[VECEXT1_I_2]], [[ADD_I_1]]
-// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef half @llvm.sqrt.f16(half [[ADD_I_2]])
+// CHECK-NEXT: [[HLSL_DOT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn half @llvm.[[TARGET]].fdot.v4f16(<4 x half> [[P0]], <4 x half> [[P0]])
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef half @llvm.sqrt.f16(half [[HLSL_DOT_I]])
// CHECK-NEXT: ret half [[TMP0]]
//
half test_length_half4(half4 p0)
@@ -65,8 +63,8 @@ half test_length_half4(half4 p0)
return length(p0);
}
-
-// CHECK-LABEL: define noundef nofpclass(nan inf) float
@_Z17test_length_floatf(
+// SPVCHECK-LABEL: define spir_func noundef nofpclass(nan inf) float
@_Z17test_length_floatf(
+// DXCHECK-LABEL: define noundef nofpclass(nan inf) float
@_Z17test_length_floatf(
// CHECK-SAME: float noundef nofpclass(nan inf) [[P0:%.*]]) local_unnamed_addr
#[[ATTR0]] {
// CHECK-NEXT: [[ENTRY:.*:]]
// CHECK-NEXT: [[ELT_ABS_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef float @llvm.fabs.f32(float [[P0]])
@@ -77,14 +75,12 @@ float test_length_float(float p0)
return length(p0);
}
-// CHECK-LABEL: define noundef nofpclass(nan inf) float
@_Z18test_length_float2Dv2_f(
+// SPVCHECK-LABEL: define spir_func noundef nofpclass(nan inf) float
@_Z18test_length_float2Dv2_f(
+// DXCHECK-LABEL: define noundef nofpclass(nan inf) float
@_Z18test_length_float2Dv2_f(
// CHECK-SAME: <2 x float> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn float
[[VECEXT_I]], [[VECEXT1_I]]
-// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef float @llvm.sqrt.f32(float [[ADD_I]])
+// CHECK-NEXT: [[HLSL_DOT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn float @llvm.[[TARGET]].fdot.v2f32(<2 x float> [[P0]], <2 x float> [[P0]])
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef float @llvm.sqrt.f32(float [[HLSL_DOT_I]])
// CHECK-NEXT: ret float [[TMP0]]
//
float test_length_float2(float2 p0)
@@ -92,16 +88,12 @@ float test_length_float2(float2 p0)
return length(p0);
}
-// CHECK-LABEL: define noundef nofpclass(nan inf) float
@_Z18test_length_float3Dv3_f(
+// SPVCHECK-LABEL: define spir_func noundef nofpclass(nan inf) float
@_Z18test_length_float3Dv3_f(
+// DXCHECK-LABEL: define noundef nofpclass(nan inf) float
@_Z18test_length_float3Dv3_f(
// CHECK-SAME: <3 x float> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn float
[[VECEXT1_I]], [[VECEXT_I]]
-// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <3 x float> [[MUL_I]],
i64 2
-// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd reassoc nnan ninf nsz arcp afn float
[[VECEXT1_I_1]], [[ADD_I]]
-// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef float @llvm.sqrt.f32(float [[ADD_I_1]])
+// CHECK-NEXT: [[HLSL_DOT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn float @llvm.[[TARGET]].fdot.v3f32(<3 x float> [[P0]], <3 x float> [[P0]])
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef float @llvm.sqrt.f32(float [[HLSL_DOT_I]])
// CHECK-NEXT: ret float [[TMP0]]
//
float test_length_float3(float3 p0)
@@ -109,19 +101,12 @@ float test_length_float3(float3 p0)
return length(p0);
}
-
-// CHECK-LABEL: define noundef nofpclass(nan inf) float
@_Z18test_length_float4Dv4_f(
+// SPVCHECK-LABEL: define spir_func noundef nofpclass(nan inf) float
@_Z18test_length_float4Dv4_f(
+// DXCHECK-LABEL: define noundef nofpclass(nan inf) float
@_Z18test_length_float4Dv4_f(
// CHECK-SAME: <4 x float> noundef nofpclass(nan inf) [[P0:%.*]])
local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn <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 reassoc nnan ninf nsz arcp afn float
[[VECEXT1_I]], [[VECEXT_I]]
-// CHECK-NEXT: [[VECEXT1_I_1:%.*]] = extractelement <4 x float> [[MUL_I]],
i64 2
-// CHECK-NEXT: [[ADD_I_1:%.*]] = fadd reassoc nnan ninf nsz arcp afn float
[[VECEXT1_I_1]], [[ADD_I]]
-// CHECK-NEXT: [[VECEXT1_I_2:%.*]] = extractelement <4 x float> [[MUL_I]],
i64 3
-// CHECK-NEXT: [[ADD_I_2:%.*]] = fadd reassoc nnan ninf nsz arcp afn float
[[VECEXT1_I_2]], [[ADD_I_1]]
-// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef float @llvm.sqrt.f32(float [[ADD_I_2]])
+// CHECK-NEXT: [[HLSL_DOT_I:%.*]] = tail call reassoc nnan ninf nsz arcp
afn float @llvm.[[TARGET]].fdot.v4f32(<4 x float> [[P0]], <4 x float> [[P0]])
+// CHECK-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn
noundef float @llvm.sqrt.f32(float [[HLSL_DOT_I]])
// CHECK-NEXT: ret float [[TMP0]]
//
float test_length_float4(float4 p0)
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
