[clang] [HLSL] Implement a header only distance intrinsic (PR #117240)
Farzon Lotfi via cfe-commits
cfe-commits at lists.llvm.org
Thu Nov 21 13:05:25 PST 2024
https://github.com/farzonl created https://github.com/llvm/llvm-project/pull/117240
- For the HLSL intrinsic used `const inline` because that seems to be the pattern `countbits`
- For the helper functions the pattern in `hlsl_detail.h` was to do `constexpr` like with `bit_cast` and `enable_if_t` So did the same here.
- Distance in DXC is defined as [Length(X-Y)](https://github.com/microsoft/DirectXShaderCompiler/blob/848b7c42bd8da13693273513412c0a554c7918a1/lib/HLSL/HLOperationLower.cpp#L2188C1-L2198C2), So doing the same here.
- Maybe this also means we need to move length into the header
- This resolves all the DirectX specific parts of https://github.com/llvm/llvm-project/issues/99107
- The codegen will be functionally correct for SPIRV, but will not emit the GLSL Distance opcode.
- There are many potential solutions for that. Atm, we intend to address the GLSL specific ops with Inline SPIRV.
>From 6aaf0c42844409fca63f30fa716b9b65f2127f85 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Thu, 21 Nov 2024 14:46:31 -0500
Subject: [PATCH] [HLSL] Implement a header only distance intrinsic
---
clang/lib/Headers/hlsl/hlsl_detail.h | 8 ++
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 39 ++++++++++
clang/test/CodeGenHLSL/builtins/distance.hlsl | 76 +++++++++++++++++++
.../SemaHLSL/BuiltIns/distance-errors.hlsl | 33 ++++++++
4 files changed, 156 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/distance.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/distance-errors.hlsl
diff --git a/clang/lib/Headers/hlsl/hlsl_detail.h b/clang/lib/Headers/hlsl/hlsl_detail.h
index 8d5fd941331531..99cb3fa4a6d2e8 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> {
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index a484d04155d6b2..93d180618c6657 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -853,6 +853,45 @@ float3 degrees(float3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_degrees)
float4 degrees(float4);
+//===----------------------------------------------------------------------===//
+// distance builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn K distance(T X, T Y)
+/// \brief Returns a distance scalar between two vectors of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+template <typename T>
+constexpr __detail::enable_if_t<
+ __detail::is_same<float, T>::value || __detail::is_same<half, T>::value, T>
+distance_impl(T X, T Y) {
+ return __builtin_elementwise_abs(X - Y);
+}
+
+template <typename T, int N>
+constexpr __detail::enable_if_t<
+ __detail::is_same<float, T>::value || __detail::is_same<half, T>::value, T>
+distance_vec_impl(vector<T, N> X, vector<T, N> Y) {
+ return __builtin_hlsl_length(X - Y);
+}
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline half distance(half X, half Y) { return distance_impl(X, Y); }
+
+template <int N>
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline half distance(vector<half, N> X, vector<half, N> Y) {
+ return distance_vec_impl(X, Y);
+}
+
+const inline float distance(float X, float Y) { return distance_impl(X, Y); }
+
+template <int N>
+const inline float distance(vector<float, N> X, vector<float, N> Y) {
+ return distance_vec_impl(X, Y);
+}
+
//===----------------------------------------------------------------------===//
// dot product builtins
//===----------------------------------------------------------------------===//
diff --git a/clang/test/CodeGenHLSL/builtins/distance.hlsl b/clang/test/CodeGenHLSL/builtins/distance.hlsl
new file mode 100644
index 00000000000000..2ff2947ac49095
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/distance.hlsl
@@ -0,0 +1,76 @@
+// 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
+
+// CHECK-LABEL: define noundef half @_Z18test_distance_halfDhDh(
+// CHECK-SAME: half noundef [[X:%.*]], half noundef [[Y:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[SUB_I:%.*]] = fsub half [[X]], [[Y]]
+// CHECK-NEXT: [[ELT_ABS_I:%.*]] = tail call noundef half @llvm.fabs.f16(half [[SUB_I]])
+// CHECK-NEXT: ret half [[ELT_ABS_I]]
+//
+half test_distance_half(half X, half Y) { return distance(X, Y); }
+
+// CHECK-LABEL: define noundef half @_Z19test_distance_half2Dv2_DhS_(
+// CHECK-SAME: <2 x half> noundef [[X:%.*]], <2 x half> noundef [[Y:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[SUB_I:%.*]] = fsub <2 x half> [[X]], [[Y]]
+// CHECK-NEXT: [[HLSL_LENGTH_I:%.*]] = tail call noundef half @llvm.dx.length.v2f16(<2 x half> [[SUB_I]])
+// CHECK-NEXT: ret half [[HLSL_LENGTH_I]]
+//
+half test_distance_half2(half2 X, half2 Y) { return distance(X, Y); }
+
+// CHECK-LABEL: define noundef half @_Z19test_distance_half3Dv3_DhS_(
+// CHECK-SAME: <3 x half> noundef [[X:%.*]], <3 x half> noundef [[Y:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[SUB_I:%.*]] = fsub <3 x half> [[X]], [[Y]]
+// CHECK-NEXT: [[HLSL_LENGTH_I:%.*]] = tail call noundef half @llvm.dx.length.v3f16(<3 x half> [[SUB_I]])
+// CHECK-NEXT: ret half [[HLSL_LENGTH_I]]
+//
+half test_distance_half3(half3 X, half3 Y) { return distance(X, Y); }
+
+// CHECK-LABEL: define noundef half @_Z19test_distance_half4Dv4_DhS_(
+// CHECK-SAME: <4 x half> noundef [[X:%.*]], <4 x half> noundef [[Y:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[SUB_I:%.*]] = fsub <4 x half> [[X]], [[Y]]
+// CHECK-NEXT: [[HLSL_LENGTH_I:%.*]] = tail call noundef half @llvm.dx.length.v4f16(<4 x half> [[SUB_I]])
+// CHECK-NEXT: ret half [[HLSL_LENGTH_I]]
+//
+half test_distance_half4(half4 X, half4 Y) { return distance(X, Y); }
+
+// CHECK-LABEL: define noundef float @_Z19test_distance_floatff(
+// CHECK-SAME: float noundef [[X:%.*]], float noundef [[Y:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[SUB_I:%.*]] = fsub float [[X]], [[Y]]
+// CHECK-NEXT: [[ELT_ABS_I:%.*]] = tail call noundef float @llvm.fabs.f32(float [[SUB_I]])
+// CHECK-NEXT: ret float [[ELT_ABS_I]]
+//
+float test_distance_float(float X, float Y) { return distance(X, Y); }
+
+// CHECK-LABEL: define noundef float @_Z20test_distance_float2Dv2_fS_(
+// CHECK-SAME: <2 x float> noundef [[X:%.*]], <2 x float> noundef [[Y:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[SUB_I:%.*]] = fsub <2 x float> [[X]], [[Y]]
+// CHECK-NEXT: [[HLSL_LENGTH_I:%.*]] = tail call noundef float @llvm.dx.length.v2f32(<2 x float> [[SUB_I]])
+// CHECK-NEXT: ret float [[HLSL_LENGTH_I]]
+//
+float test_distance_float2(float2 X, float2 Y) { return distance(X, Y); }
+
+// CHECK-LABEL: define noundef float @_Z20test_distance_float3Dv3_fS_(
+// CHECK-SAME: <3 x float> noundef [[X:%.*]], <3 x float> noundef [[Y:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[SUB_I:%.*]] = fsub <3 x float> [[X]], [[Y]]
+// CHECK-NEXT: [[HLSL_LENGTH_I:%.*]] = tail call noundef float @llvm.dx.length.v3f32(<3 x float> [[SUB_I]])
+// CHECK-NEXT: ret float [[HLSL_LENGTH_I]]
+//
+float test_distance_float3(float3 X, float3 Y) { return distance(X, Y); }
+
+// CHECK-LABEL: define noundef float @_Z20test_distance_float4Dv4_fS_(
+// CHECK-SAME: <4 x float> noundef [[X:%.*]], <4 x float> noundef [[Y:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[SUB_I:%.*]] = fsub <4 x float> [[X]], [[Y]]
+// CHECK-NEXT: [[HLSL_LENGTH_I:%.*]] = tail call noundef float @llvm.dx.length.v4f32(<4 x float> [[SUB_I]])
+// CHECK-NEXT: ret float [[HLSL_LENGTH_I]]
+//
+float test_distance_float4(float4 X, float4 Y) { return distance(X, Y); }
diff --git a/clang/test/SemaHLSL/BuiltIns/distance-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/distance-errors.hlsl
new file mode 100644
index 00000000000000..e996bf5d2cb7c5
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/distance-errors.hlsl
@@ -0,0 +1,33 @@
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only -disable-llvm-passes -verify
+
+float test_no_second_arg(float2 p0) {
+ return distance(p0);
+ // expected-error at -1 {{no matching function for call to 'distance'}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function not viable: requires 2 arguments, but 1 was provided}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function not viable: requires 2 arguments, but 1 was provided}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 2 arguments, but 1 was provided}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 2 arguments, but 1 was provided}}
+}
+
+float test_too_many_arg(float2 p0) {
+ return distance(p0, p0, p0);
+ // expected-error at -1 {{no matching function for call to 'distance'}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function not viable: requires 2 arguments, but 3 were provided}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function not viable: requires 2 arguments, but 3 were provided}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 2 arguments, but 3 were provided}}
+ // expected-note at 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 distance(p0, p1);
+ // expected-error at -1 {{call to 'distance' is ambiguous}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function}}
+}
+
+float test_int_inputs(int p0, int p1) {
+ return distance(p0, p1);
+ // expected-error at -1 {{call to 'distance' is ambiguous}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function}}
+ // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function}}
+}
More information about the cfe-commits
mailing list