[HLSL] Implement the `lit` intrinsic #134171

kmpeng · 2025-04-02T22:58:14Z

Closes #99135.

Tasks completed:

Wrote implementation in hlsl_intrinsics.h/hlsl_intrinsic_helpers.h
Added codegen tests to clang/test/CodeGenHLSL/builtins/lit.hlsl

llvmbot · 2025-04-02T22:58:47Z

@llvm/pr-subscribers-clang

@llvm/pr-subscribers-hlsl

Author: Kaitlin Peng (kmpeng)

Changes

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

Full diff: https://github.com/llvm/llvm-project/pull/134171.diff

8 Files Affected:

(modified) clang/lib/Headers/hlsl.h (+1-1)
(modified) clang/lib/Headers/hlsl/hlsl_compat_overloads.h (+16)
(modified) clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h (+10)
(modified) clang/lib/Headers/hlsl/hlsl_intrinsics.h (+31)
(added) clang/test/CodeGenHLSL/builtins/lit-overloads.hlsl (+93)
(added) clang/test/CodeGenHLSL/builtins/lit.hlsl (+36)
(added) clang/test/SemaHLSL/BuiltIns/lit-errors-16bit.hlsl (+9)
(added) clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl (+41)

diff --git a/clang/lib/Headers/hlsl.h b/clang/lib/Headers/hlsl.h
index b494b4d0f78bb..2bc1973f6eb2b 100644
--- a/clang/lib/Headers/hlsl.h
+++ b/clang/lib/Headers/hlsl.h
@@ -22,10 +22,10 @@
 
 // HLSL standard library function declarations/definitions.
 #include "hlsl/hlsl_alias_intrinsics.h"
+#include "hlsl/hlsl_intrinsics.h"
 #if __HLSL_VERSION <= __HLSL_202x
 #include "hlsl/hlsl_compat_overloads.h"
 #endif
-#include "hlsl/hlsl_intrinsics.h"
 
 #if defined(__clang__)
 #pragma clang diagnostic pop
diff --git a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
index 47ae34adfe541..cbf5364c0b29c 100644
--- a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
+++ b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
@@ -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 <typename T>
+constexpr __detail::enable_if_t<__detail::is_arithmetic<T>::Value &&
+                                    (__detail::is_same<double, T>::value ||
+                                     __detail::is_same<int, T>::value ||
+                                     __detail::is_same<uint, T>::value ||
+                                     __detail::is_same<int64_t, T>::value ||
+                                     __detail::is_same<uint64_t, T>::value),
+                                vector<T, 4>>
+lit(T NDotL, T NDotH, T M) {
+  return lit((float)NDotL, (float)NDotH, (float)M);
+}
+
 //===----------------------------------------------------------------------===//
 // log builtins overloads
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
index 8cdd63d7e07bb..3a41635a89aa4 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
@@ -101,6 +101,16 @@ constexpr vector<T, N> smoothstep_vec_impl(vector<T, N> Min, vector<T, N> Max,
 #endif
 }
 
+template <typename T> constexpr vector<T, 4> lit_impl(T NDotL, T NDotH, T M) {
+  bool DiffuseCond = NDotL < 0;
+  T Diffuse = select<T>(DiffuseCond, 0, NDotL);
+  vector<T, 4> Result = {1, Diffuse, 0, 1};
+  bool SpecularCond = or (DiffuseCond, (NDotH < 0));
+  T SpecularExp = exp(log(NDotH) * M);
+  Result[2] = select<T>(SpecularCond, 0, SpecularExp);
+  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..5d23eb89b889b 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<float, N> X) {
   return __detail::length_vec_impl(X);
 }
 
+//===----------------------------------------------------------------------===//
+// lit builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn vector<T, 4> 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 <typename T>
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_arithmetic<T>::Value &&
+                                       __detail::is_same<half, T>::value,
+                                   vector<T, 4>> lit(T NDotL, T NDotH, T M) {
+  return __detail::lit_impl(NDotL, NDotH, M);
+}
+
+template <typename T>
+const inline __detail::enable_if_t<__detail::is_arithmetic<T>::Value &&
+                                       __detail::is_same<float, T>::value,
+                                   vector<T, 4>>
+lit(T NDotL, T NDotH, T M) {
+  return __detail::lit_impl(NDotL, NDotH, M);
+}
+
 //===----------------------------------------------------------------------===//
 // D3DCOLORtoUBYTE4 builtin
 //===----------------------------------------------------------------------===//
diff --git a/clang/test/CodeGenHLSL/builtins/lit-overloads.hlsl b/clang/test/CodeGenHLSL/builtins/lit-overloads.hlsl
new file mode 100644
index 0000000000000..676889ecf16b0
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/lit-overloads.hlsl
@@ -0,0 +1,93 @@
+// RUN: %clang_cc1 -std=hlsl202x -finclude-default-header -x hlsl -triple dxil-pc-shadermodel6.3-library %s \
+// RUN:  -emit-llvm -o - | \
+// RUN:  FileCheck %s --check-prefixes=CHECK
+
+// CHECK-LABEL: test_lit_double
+// CHECK: %conv.i = fptrunc reassoc nnan ninf nsz arcp afn double %{{.*}} to float
+// CHECK: %conv1.i = fptrunc reassoc nnan ninf nsz arcp afn double %{{.*}} to float
+// CHECK: %conv2.i = fptrunc reassoc nnan ninf nsz arcp afn double %{{.*}} to float
+// CHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecinit.i = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float poison>, float %{{.*}}, i32 1
+// CHECK: %cmp4.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.or.i = or i1 %{{.*}}, %cmp4.i
+// CHECK: %elt.log.i = call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float %{{.*}})
+// CHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn float %elt.log.i, %{{.*}}
+// CHECK: %elt.exp.i = call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float %mul.i)
+// CHECK: %hlsl.select7.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecins.i = insertelement <4 x float> %{{.*}}, float %hlsl.select7.i, i32 2
+// CHECK: %conv3.i = fpext reassoc nnan ninf nsz arcp afn <4 x float> %{{.*}} to <4 x double>
+// CHECK: ret <4 x double> %conv3.i
+double4 test_lit_double(double NDotL, double NDotH, double M) { return lit(NDotL, NDotH, M); }
+
+// CHECK-LABEL: test_lit_int
+// CHECK: %conv.i = sitofp i32 %{{.*}} to float
+// CHECK: %conv1.i = sitofp i32 %{{.*}} to float
+// CHECK: %conv2.i = sitofp i32 %{{.*}} to float
+// CHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecinit.i = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float poison>, float %{{.*}}, i32 1
+// CHECK: %cmp4.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.or.i = or i1 %{{.*}}, %cmp4.i
+// CHECK: %elt.log.i = call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float %{{.*}})
+// CHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn float %elt.log.i, %{{.*}}
+// CHECK: %elt.exp.i = call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float %mul.i)
+// CHECK: %hlsl.select7.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecins.i = insertelement <4 x float> %{{.*}}, float %hlsl.select7.i, i32 2
+// CHECK: %conv3.i = fptosi <4 x float> %{{.*}} to <4 x i32>
+// CHECK: ret <4 x i32> %conv3.i
+int4 test_lit_int(int NDotL, int NDotH, int M) { return lit(NDotL, NDotH, M); }
+
+// CHECK-LABEL: test_lit_uint
+// CHECK: %conv.i = uitofp i32 %{{.*}} to float
+// CHECK: %conv1.i = uitofp i32 %{{.*}} to float
+// CHECK: %conv2.i = uitofp i32 %{{.*}} to float
+// CHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecinit.i = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float poison>, float %{{.*}}, i32 1
+// CHECK: %cmp4.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.or.i = or i1 %{{.*}}, %cmp4.i
+// CHECK: %elt.log.i = call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float %{{.*}})
+// CHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn float %elt.log.i, %{{.*}}
+// CHECK: %elt.exp.i = call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float %mul.i)
+// CHECK: %hlsl.select7.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecins.i = insertelement <4 x float> %{{.*}}, float %hlsl.select7.i, i32 2
+// CHECK: %conv3.i = fptoui <4 x float> %{{.*}} to <4 x i32>
+// CHECK: ret <4 x i32> %conv3.i
+uint4 test_lit_uint(uint NDotL, uint NDotH, uint M) { return lit(NDotL, NDotH, M); }
+
+// CHECK-LABEL: test_lit_int64_t
+// CHECK: %conv.i = sitofp i64 %{{.*}} to float
+// CHECK: %conv1.i = sitofp i64 %{{.*}} to float
+// CHECK: %conv2.i = sitofp i64 %{{.*}} to float
+// CHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecinit.i = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float poison>, float %{{.*}}, i32 1
+// CHECK: %cmp4.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.or.i = or i1 %{{.*}}, %cmp4.i
+// CHECK: %elt.log.i = call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float %{{.*}})
+// CHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn float %elt.log.i, %{{.*}}
+// CHECK: %elt.exp.i = call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float %mul.i)
+// CHECK: %hlsl.select7.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecins.i = insertelement <4 x float> %{{.*}}, float %hlsl.select7.i, i32 2
+// CHECK: %conv3.i = fptosi <4 x float> %{{.*}} to <4 x i64>
+// CHECK: ret <4 x i64> %conv3.i
+int64_t4 test_lit_int64_t(int64_t NDotL, int64_t NDotH, int64_t M) { return lit(NDotL, NDotH, M); }
+
+// CHECK-LABEL: test_lit_uint64_t
+// CHECK: %conv.i = uitofp i64 %{{.*}} to float
+// CHECK: %conv1.i = uitofp i64 %{{.*}} to float
+// CHECK: %conv2.i = uitofp i64 %{{.*}} to float
+// CHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecinit.i = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float poison>, float %{{.*}}, i32 1
+// CHECK: %cmp4.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.or.i = or i1 %{{.*}}, %cmp4.i
+// CHECK: %elt.log.i = call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float %{{.*}})
+// CHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn float %elt.log.i, %{{.*}}
+// CHECK: %elt.exp.i = call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float %mul.i)
+// CHECK: %hlsl.select7.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecins.i = insertelement <4 x float> %{{.*}}, float %hlsl.select7.i, i32 2
+// CHECK: %conv3.i = fptoui <4 x float> %{{.*}} to <4 x i64>
+// CHECK: ret <4 x i64> %conv3.i
+uint64_t4 test_lit_uint64_t(uint64_t NDotL, uint64_t NDotH, uint64_t M) { return lit(NDotL, NDotH, M); }
diff --git a/clang/test/CodeGenHLSL/builtins/lit.hlsl b/clang/test/CodeGenHLSL/builtins/lit.hlsl
new file mode 100644
index 0000000000000..1737a460e020f
--- /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: 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]], 0xH0000
+// CHECK-NEXT:    [[HLSL_SELECT_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn half @llvm.maxnum.f16(half [[NDOTL]], half 0xH0000)
+// CHECK-NEXT:    [[VECINIT2_I:%.*]] = insertelement <4 x half> <half 0xH3C00, half poison, half poison, half 0xH3C00>, half [[HLSL_SELECT_I]], i64 1
+// CHECK-NEXT:    [[CMP4_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt half [[NDOTH]], 0xH0000
+// 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 0xH0000, 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]]
+//
+half4 test_lit_half(half NDotL, half NDotH, half M) { return lit(NDotL, NDotH, M); }
+
+// CHECK-LABEL: test_lit_float
+// CHECK-SAME: float noundef nofpclass(nan inf) [[NDOTL:%.*]], float noundef nofpclass(nan inf) [[NDOTH:%.*]], float noundef nofpclass(nan inf) [[M:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[CMP_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt float [[NDOTL]], 0.000000e+00
+// CHECK-NEXT:    [[HLSL_SELECT_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn float @llvm.maxnum.f32(float [[NDOTL]], float 0.000000e+00)
+// CHECK-NEXT:    [[VECINIT2_I:%.*]] = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float 1.000000e+00>, float [[HLSL_SELECT_I]], i64 1
+// CHECK-NEXT:    [[CMP4_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt float [[NDOTH]], 0.000000e+00
+// CHECK-NEXT:    [[HLSL_OR_I:%.*]] = or i1 [[CMP_I]], [[CMP4_I]]
+// CHECK-NEXT:    [[ELT_LOG_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float [[NDOTH]])
+// CHECK-NEXT:    [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn float [[ELT_LOG_I]], [[M]]
+// CHECK-NEXT:    [[ELT_EXP_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float [[MUL_I]])
+// CHECK-NEXT:    [[HLSL_SELECT7_I:%.*]] = select reassoc nnan ninf nsz arcp afn i1 [[HLSL_OR_I]], float 0.000000e+00, float [[ELT_EXP_I]]
+// CHECK-NEXT:    [[VECINS_I:%.*]] = insertelement <4 x float> [[VECINIT2_I]], float [[HLSL_SELECT7_I]], i64 2
+// CHECK-NEXT:    ret <4 x float> [[VECINS_I]]
+//
+float4 test_lit_float(float NDotL, float NDotH, float M) { return lit(NDotL, NDotH, M); }
diff --git a/clang/test/SemaHLSL/BuiltIns/lit-errors-16bit.hlsl b/clang/test/SemaHLSL/BuiltIns/lit-errors-16bit.hlsl
new file mode 100644
index 0000000000000..311bad9a0ef79
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/lit-errors-16bit.hlsl
@@ -0,0 +1,9 @@
+// RUN: not %clang_dxc -enable-16bit-types -T cs_6_0 -HV 202x %s 2>&1  | FileCheck %s -DTEST_TYPE=half
+// RUN: not %clang_dxc -enable-16bit-types -T cs_6_0 -HV 202x %s 2>&1  | FileCheck %s -DTEST_TYPE=int16_t
+// RUN: not %clang_dxc -enable-16bit-types -T cs_6_0 -HV 202x %s 2>&1  | FileCheck %s -DTEST_TYPE=uint16_t
+
+// check we error on 16 bit type if shader model is too old
+// CHECK: '-enable-16bit-types' option requires target HLSL Version >= 2018 and shader model >= 6.2, but HLSL Version is 'hlsl202x' and shader model is '6.0'
+vector<TEST_TYPE,4> test_error(TEST_TYPE p0) {
+  return lit(p0, p0, p0);
+}
diff --git a/clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl
new file mode 100644
index 0000000000000..799fb62a9560d
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl
@@ -0,0 +1,41 @@
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only -disable-llvm-passes -verify
+
+float4 test_no_second_arg(float p0) {
+  return lit(p0);
+  // expected-error@-1 {{no matching function for call to 'lit'}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 1 was provided}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 1 was provided}}
+  // expected-note@hlsl/hlsl_compat_overloads.h:* {{candidate function template not viable: requires 3 arguments, but 1 was provided}}
+}
+
+float4 test_no_third_arg(float p0) {
+  return lit(p0, p0);
+  // expected-error@-1 {{no matching function for call to 'lit'}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 2 were provided}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 2 were provided}}
+  // expected-note@hlsl/hlsl_compat_overloads.h:* {{candidate function template not viable: requires 3 arguments, but 2 were provided}}
+}
+
+float4 test_too_many_arg(float p0) {
+  return lit(p0, p0, p0, p0);
+  // expected-error@-1 {{no matching function for call to 'lit'}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 4 were provided}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 4 were provided}}
+  // expected-note@hlsl/hlsl_compat_overloads.h:* {{candidate function template not viable: requires 3 arguments, but 4 were provided}}
+}
+
+float4 test_vec_inputs(float2 p0, float2 p1, float2 p2) {
+  return lit(p0, p1, p2);
+  // expected-error@-1  {{no matching function for call to 'lit'}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate template ignored: substitution failure [with T = float2]: invalid vector element type 'vector<float, 2>' (vector of 2 'float' values)}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate template ignored: substitution failure [with T = float2]: invalid vector element type 'vector<float, 2>' (vector of 2 'float' values)}}
+  // expected-note@hlsl/hlsl_compat_overloads.h:* {{candidate template ignored: substitution failure [with T = float2]: invalid vector element type 'vector<float, 2>' (vector of 2 'float' values)}}
+}
+
+float4 test_vec1_inputs(float1 p0, float1 p1, float1 p2) {
+  return lit(p0, p1, p2);
+  // expected-error@-1  {{no matching function for call to 'lit'}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate template ignored: substitution failure [with T = float1]: invalid vector element type 'vector<float, 1>' (vector of 1 'float' value)}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate template ignored: substitution failure [with T = float1]: invalid vector element type 'vector<float, 1>' (vector of 1 'float' value)}}
+  // expected-note@hlsl/hlsl_compat_overloads.h:* {{candidate template ignored: substitution failure [with T = float1]: invalid vector element type 'vector<float, 1>' (vector of 1 'float' value)}}
+}

llvmbot · 2025-04-02T22:58:48Z

@llvm/pr-subscribers-backend-x86

Author: Kaitlin Peng (kmpeng)

Changes

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

Full diff: https://github.com/llvm/llvm-project/pull/134171.diff

8 Files Affected:

(modified) clang/lib/Headers/hlsl.h (+1-1)
(modified) clang/lib/Headers/hlsl/hlsl_compat_overloads.h (+16)
(modified) clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h (+10)
(modified) clang/lib/Headers/hlsl/hlsl_intrinsics.h (+31)
(added) clang/test/CodeGenHLSL/builtins/lit-overloads.hlsl (+93)
(added) clang/test/CodeGenHLSL/builtins/lit.hlsl (+36)
(added) clang/test/SemaHLSL/BuiltIns/lit-errors-16bit.hlsl (+9)
(added) clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl (+41)

diff --git a/clang/lib/Headers/hlsl.h b/clang/lib/Headers/hlsl.h
index b494b4d0f78bb..2bc1973f6eb2b 100644
--- a/clang/lib/Headers/hlsl.h
+++ b/clang/lib/Headers/hlsl.h
@@ -22,10 +22,10 @@
 
 // HLSL standard library function declarations/definitions.
 #include "hlsl/hlsl_alias_intrinsics.h"
+#include "hlsl/hlsl_intrinsics.h"
 #if __HLSL_VERSION <= __HLSL_202x
 #include "hlsl/hlsl_compat_overloads.h"
 #endif
-#include "hlsl/hlsl_intrinsics.h"
 
 #if defined(__clang__)
 #pragma clang diagnostic pop
diff --git a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
index 47ae34adfe541..cbf5364c0b29c 100644
--- a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
+++ b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
@@ -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 <typename T>
+constexpr __detail::enable_if_t<__detail::is_arithmetic<T>::Value &&
+                                    (__detail::is_same<double, T>::value ||
+                                     __detail::is_same<int, T>::value ||
+                                     __detail::is_same<uint, T>::value ||
+                                     __detail::is_same<int64_t, T>::value ||
+                                     __detail::is_same<uint64_t, T>::value),
+                                vector<T, 4>>
+lit(T NDotL, T NDotH, T M) {
+  return lit((float)NDotL, (float)NDotH, (float)M);
+}
+
 //===----------------------------------------------------------------------===//
 // log builtins overloads
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
index 8cdd63d7e07bb..3a41635a89aa4 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
@@ -101,6 +101,16 @@ constexpr vector<T, N> smoothstep_vec_impl(vector<T, N> Min, vector<T, N> Max,
 #endif
 }
 
+template <typename T> constexpr vector<T, 4> lit_impl(T NDotL, T NDotH, T M) {
+  bool DiffuseCond = NDotL < 0;
+  T Diffuse = select<T>(DiffuseCond, 0, NDotL);
+  vector<T, 4> Result = {1, Diffuse, 0, 1};
+  bool SpecularCond = or (DiffuseCond, (NDotH < 0));
+  T SpecularExp = exp(log(NDotH) * M);
+  Result[2] = select<T>(SpecularCond, 0, SpecularExp);
+  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..5d23eb89b889b 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<float, N> X) {
   return __detail::length_vec_impl(X);
 }
 
+//===----------------------------------------------------------------------===//
+// lit builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn vector<T, 4> 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 <typename T>
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_arithmetic<T>::Value &&
+                                       __detail::is_same<half, T>::value,
+                                   vector<T, 4>> lit(T NDotL, T NDotH, T M) {
+  return __detail::lit_impl(NDotL, NDotH, M);
+}
+
+template <typename T>
+const inline __detail::enable_if_t<__detail::is_arithmetic<T>::Value &&
+                                       __detail::is_same<float, T>::value,
+                                   vector<T, 4>>
+lit(T NDotL, T NDotH, T M) {
+  return __detail::lit_impl(NDotL, NDotH, M);
+}
+
 //===----------------------------------------------------------------------===//
 // D3DCOLORtoUBYTE4 builtin
 //===----------------------------------------------------------------------===//
diff --git a/clang/test/CodeGenHLSL/builtins/lit-overloads.hlsl b/clang/test/CodeGenHLSL/builtins/lit-overloads.hlsl
new file mode 100644
index 0000000000000..676889ecf16b0
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/lit-overloads.hlsl
@@ -0,0 +1,93 @@
+// RUN: %clang_cc1 -std=hlsl202x -finclude-default-header -x hlsl -triple dxil-pc-shadermodel6.3-library %s \
+// RUN:  -emit-llvm -o - | \
+// RUN:  FileCheck %s --check-prefixes=CHECK
+
+// CHECK-LABEL: test_lit_double
+// CHECK: %conv.i = fptrunc reassoc nnan ninf nsz arcp afn double %{{.*}} to float
+// CHECK: %conv1.i = fptrunc reassoc nnan ninf nsz arcp afn double %{{.*}} to float
+// CHECK: %conv2.i = fptrunc reassoc nnan ninf nsz arcp afn double %{{.*}} to float
+// CHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecinit.i = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float poison>, float %{{.*}}, i32 1
+// CHECK: %cmp4.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.or.i = or i1 %{{.*}}, %cmp4.i
+// CHECK: %elt.log.i = call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float %{{.*}})
+// CHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn float %elt.log.i, %{{.*}}
+// CHECK: %elt.exp.i = call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float %mul.i)
+// CHECK: %hlsl.select7.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecins.i = insertelement <4 x float> %{{.*}}, float %hlsl.select7.i, i32 2
+// CHECK: %conv3.i = fpext reassoc nnan ninf nsz arcp afn <4 x float> %{{.*}} to <4 x double>
+// CHECK: ret <4 x double> %conv3.i
+double4 test_lit_double(double NDotL, double NDotH, double M) { return lit(NDotL, NDotH, M); }
+
+// CHECK-LABEL: test_lit_int
+// CHECK: %conv.i = sitofp i32 %{{.*}} to float
+// CHECK: %conv1.i = sitofp i32 %{{.*}} to float
+// CHECK: %conv2.i = sitofp i32 %{{.*}} to float
+// CHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecinit.i = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float poison>, float %{{.*}}, i32 1
+// CHECK: %cmp4.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.or.i = or i1 %{{.*}}, %cmp4.i
+// CHECK: %elt.log.i = call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float %{{.*}})
+// CHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn float %elt.log.i, %{{.*}}
+// CHECK: %elt.exp.i = call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float %mul.i)
+// CHECK: %hlsl.select7.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecins.i = insertelement <4 x float> %{{.*}}, float %hlsl.select7.i, i32 2
+// CHECK: %conv3.i = fptosi <4 x float> %{{.*}} to <4 x i32>
+// CHECK: ret <4 x i32> %conv3.i
+int4 test_lit_int(int NDotL, int NDotH, int M) { return lit(NDotL, NDotH, M); }
+
+// CHECK-LABEL: test_lit_uint
+// CHECK: %conv.i = uitofp i32 %{{.*}} to float
+// CHECK: %conv1.i = uitofp i32 %{{.*}} to float
+// CHECK: %conv2.i = uitofp i32 %{{.*}} to float
+// CHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecinit.i = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float poison>, float %{{.*}}, i32 1
+// CHECK: %cmp4.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.or.i = or i1 %{{.*}}, %cmp4.i
+// CHECK: %elt.log.i = call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float %{{.*}})
+// CHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn float %elt.log.i, %{{.*}}
+// CHECK: %elt.exp.i = call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float %mul.i)
+// CHECK: %hlsl.select7.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecins.i = insertelement <4 x float> %{{.*}}, float %hlsl.select7.i, i32 2
+// CHECK: %conv3.i = fptoui <4 x float> %{{.*}} to <4 x i32>
+// CHECK: ret <4 x i32> %conv3.i
+uint4 test_lit_uint(uint NDotL, uint NDotH, uint M) { return lit(NDotL, NDotH, M); }
+
+// CHECK-LABEL: test_lit_int64_t
+// CHECK: %conv.i = sitofp i64 %{{.*}} to float
+// CHECK: %conv1.i = sitofp i64 %{{.*}} to float
+// CHECK: %conv2.i = sitofp i64 %{{.*}} to float
+// CHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecinit.i = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float poison>, float %{{.*}}, i32 1
+// CHECK: %cmp4.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.or.i = or i1 %{{.*}}, %cmp4.i
+// CHECK: %elt.log.i = call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float %{{.*}})
+// CHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn float %elt.log.i, %{{.*}}
+// CHECK: %elt.exp.i = call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float %mul.i)
+// CHECK: %hlsl.select7.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecins.i = insertelement <4 x float> %{{.*}}, float %hlsl.select7.i, i32 2
+// CHECK: %conv3.i = fptosi <4 x float> %{{.*}} to <4 x i64>
+// CHECK: ret <4 x i64> %conv3.i
+int64_t4 test_lit_int64_t(int64_t NDotL, int64_t NDotH, int64_t M) { return lit(NDotL, NDotH, M); }
+
+// CHECK-LABEL: test_lit_uint64_t
+// CHECK: %conv.i = uitofp i64 %{{.*}} to float
+// CHECK: %conv1.i = uitofp i64 %{{.*}} to float
+// CHECK: %conv2.i = uitofp i64 %{{.*}} to float
+// CHECK: %cmp.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.select.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecinit.i = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float poison>, float %{{.*}}, i32 1
+// CHECK: %cmp4.i = fcmp reassoc nnan ninf nsz arcp afn olt float %{{.*}}, 0.000000e+00
+// CHECK: %hlsl.or.i = or i1 %{{.*}}, %cmp4.i
+// CHECK: %elt.log.i = call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float %{{.*}})
+// CHECK: %mul.i = fmul reassoc nnan ninf nsz arcp afn float %elt.log.i, %{{.*}}
+// CHECK: %elt.exp.i = call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float %mul.i)
+// CHECK: %hlsl.select7.i = select reassoc nnan ninf nsz arcp afn i1 %{{.*}}, float 0.000000e+00, float %{{.*}}
+// CHECK: %vecins.i = insertelement <4 x float> %{{.*}}, float %hlsl.select7.i, i32 2
+// CHECK: %conv3.i = fptoui <4 x float> %{{.*}} to <4 x i64>
+// CHECK: ret <4 x i64> %conv3.i
+uint64_t4 test_lit_uint64_t(uint64_t NDotL, uint64_t NDotH, uint64_t M) { return lit(NDotL, NDotH, M); }
diff --git a/clang/test/CodeGenHLSL/builtins/lit.hlsl b/clang/test/CodeGenHLSL/builtins/lit.hlsl
new file mode 100644
index 0000000000000..1737a460e020f
--- /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: 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]], 0xH0000
+// CHECK-NEXT:    [[HLSL_SELECT_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn half @llvm.maxnum.f16(half [[NDOTL]], half 0xH0000)
+// CHECK-NEXT:    [[VECINIT2_I:%.*]] = insertelement <4 x half> <half 0xH3C00, half poison, half poison, half 0xH3C00>, half [[HLSL_SELECT_I]], i64 1
+// CHECK-NEXT:    [[CMP4_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt half [[NDOTH]], 0xH0000
+// 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 0xH0000, 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]]
+//
+half4 test_lit_half(half NDotL, half NDotH, half M) { return lit(NDotL, NDotH, M); }
+
+// CHECK-LABEL: test_lit_float
+// CHECK-SAME: float noundef nofpclass(nan inf) [[NDOTL:%.*]], float noundef nofpclass(nan inf) [[NDOTH:%.*]], float noundef nofpclass(nan inf) [[M:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[CMP_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt float [[NDOTL]], 0.000000e+00
+// CHECK-NEXT:    [[HLSL_SELECT_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn float @llvm.maxnum.f32(float [[NDOTL]], float 0.000000e+00)
+// CHECK-NEXT:    [[VECINIT2_I:%.*]] = insertelement <4 x float> <float 1.000000e+00, float poison, float poison, float 1.000000e+00>, float [[HLSL_SELECT_I]], i64 1
+// CHECK-NEXT:    [[CMP4_I:%.*]] = fcmp reassoc nnan ninf nsz arcp afn olt float [[NDOTH]], 0.000000e+00
+// CHECK-NEXT:    [[HLSL_OR_I:%.*]] = or i1 [[CMP_I]], [[CMP4_I]]
+// CHECK-NEXT:    [[ELT_LOG_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn float @llvm.log.f32(float [[NDOTH]])
+// CHECK-NEXT:    [[MUL_I:%.*]] = fmul reassoc nnan ninf nsz arcp afn float [[ELT_LOG_I]], [[M]]
+// CHECK-NEXT:    [[ELT_EXP_I:%.*]] = tail call reassoc nnan ninf nsz arcp afn float @llvm.exp.f32(float [[MUL_I]])
+// CHECK-NEXT:    [[HLSL_SELECT7_I:%.*]] = select reassoc nnan ninf nsz arcp afn i1 [[HLSL_OR_I]], float 0.000000e+00, float [[ELT_EXP_I]]
+// CHECK-NEXT:    [[VECINS_I:%.*]] = insertelement <4 x float> [[VECINIT2_I]], float [[HLSL_SELECT7_I]], i64 2
+// CHECK-NEXT:    ret <4 x float> [[VECINS_I]]
+//
+float4 test_lit_float(float NDotL, float NDotH, float M) { return lit(NDotL, NDotH, M); }
diff --git a/clang/test/SemaHLSL/BuiltIns/lit-errors-16bit.hlsl b/clang/test/SemaHLSL/BuiltIns/lit-errors-16bit.hlsl
new file mode 100644
index 0000000000000..311bad9a0ef79
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/lit-errors-16bit.hlsl
@@ -0,0 +1,9 @@
+// RUN: not %clang_dxc -enable-16bit-types -T cs_6_0 -HV 202x %s 2>&1  | FileCheck %s -DTEST_TYPE=half
+// RUN: not %clang_dxc -enable-16bit-types -T cs_6_0 -HV 202x %s 2>&1  | FileCheck %s -DTEST_TYPE=int16_t
+// RUN: not %clang_dxc -enable-16bit-types -T cs_6_0 -HV 202x %s 2>&1  | FileCheck %s -DTEST_TYPE=uint16_t
+
+// check we error on 16 bit type if shader model is too old
+// CHECK: '-enable-16bit-types' option requires target HLSL Version >= 2018 and shader model >= 6.2, but HLSL Version is 'hlsl202x' and shader model is '6.0'
+vector<TEST_TYPE,4> test_error(TEST_TYPE p0) {
+  return lit(p0, p0, p0);
+}
diff --git a/clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl
new file mode 100644
index 0000000000000..799fb62a9560d
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl
@@ -0,0 +1,41 @@
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only -disable-llvm-passes -verify
+
+float4 test_no_second_arg(float p0) {
+  return lit(p0);
+  // expected-error@-1 {{no matching function for call to 'lit'}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 1 was provided}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 1 was provided}}
+  // expected-note@hlsl/hlsl_compat_overloads.h:* {{candidate function template not viable: requires 3 arguments, but 1 was provided}}
+}
+
+float4 test_no_third_arg(float p0) {
+  return lit(p0, p0);
+  // expected-error@-1 {{no matching function for call to 'lit'}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 2 were provided}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 2 were provided}}
+  // expected-note@hlsl/hlsl_compat_overloads.h:* {{candidate function template not viable: requires 3 arguments, but 2 were provided}}
+}
+
+float4 test_too_many_arg(float p0) {
+  return lit(p0, p0, p0, p0);
+  // expected-error@-1 {{no matching function for call to 'lit'}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 4 were provided}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate function template not viable: requires 3 arguments, but 4 were provided}}
+  // expected-note@hlsl/hlsl_compat_overloads.h:* {{candidate function template not viable: requires 3 arguments, but 4 were provided}}
+}
+
+float4 test_vec_inputs(float2 p0, float2 p1, float2 p2) {
+  return lit(p0, p1, p2);
+  // expected-error@-1  {{no matching function for call to 'lit'}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate template ignored: substitution failure [with T = float2]: invalid vector element type 'vector<float, 2>' (vector of 2 'float' values)}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate template ignored: substitution failure [with T = float2]: invalid vector element type 'vector<float, 2>' (vector of 2 'float' values)}}
+  // expected-note@hlsl/hlsl_compat_overloads.h:* {{candidate template ignored: substitution failure [with T = float2]: invalid vector element type 'vector<float, 2>' (vector of 2 'float' values)}}
+}
+
+float4 test_vec1_inputs(float1 p0, float1 p1, float1 p2) {
+  return lit(p0, p1, p2);
+  // expected-error@-1  {{no matching function for call to 'lit'}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate template ignored: substitution failure [with T = float1]: invalid vector element type 'vector<float, 1>' (vector of 1 'float' value)}}
+  // expected-note@hlsl/hlsl_intrinsics.h:* {{candidate template ignored: substitution failure [with T = float1]: invalid vector element type 'vector<float, 1>' (vector of 1 'float' value)}}
+  // expected-note@hlsl/hlsl_compat_overloads.h:* {{candidate template ignored: substitution failure [with T = float1]: invalid vector element type 'vector<float, 1>' (vector of 1 'float' value)}}
+}

clang/lib/Headers/hlsl/hlsl_intrinsics.h

clang/test/CodeGenHLSL/builtins/lit.hlsl

clang/lib/Headers/hlsl/hlsl_compat_overloads.h

clang/lib/Headers/hlsl/hlsl_intrinsics.h

clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h

clang/lib/Headers/hlsl/hlsl_intrinsics.h

clang/lib/Headers/hlsl.h

github-actions · 2025-04-03T22:12:05Z

✅ With the latest revision this PR passed the C/C++ code formatter.

farzonl · 2025-04-07T14:51:01Z

I'm concerned that you had to remove your change in clang/lib/Headers/hlsl.h. I think you did it because you needed a template definition before you could specialize it. My concern is that now the generic template is gated by #if __HLSL_VERSION <= __HLSL_202x which mean you should see an error at shader compile time if the HLSL version were 202y.

clang/test/CodeGenHLSL/builtins/lit-overloads.hlsl

clang/test/SemaHLSL/BuiltIns/lit-errors-hlsl202y.hlsl

Icohedron · 2025-04-07T21:03:58Z

In my opinion, we should just define lit without templates, thereby following Clang's overload resolution rules and allowing implicit vector truncation. Overload resolution and implicit vector truncation are behaviors that should be consistent across the language. To me, it feels wrong to have special-case behavior on language-defining features implemented/overridden on a per-function basis.

Implicit vector truncation is also an area that needs addressing as a problem with the language, rather than being a special-case behavior handled by each function's implementation. Again, a compiler warning would go a long way to addressing problems where implicit truncation may not be intended by the programmer.

warning: implicit truncation of vector type [-Wconversion]

Or maybe implicit truncation should be removed from the language entirely (for 202y perhaps)?

If you can make the behavior consistent with DXC under all cases with some special implementation in hlsl_compat_overloads for 202x, then you can do that, but I don't think we should be writing an implementation that matches DXC now and then rewrite it entirely again when we swap to 202y that makes the language more consistent (which is one of our goals for HLSL's future, right?)

llvm-beanz · 2025-04-07T22:36:16Z

We should have half and float overloads for lit and let the rest be ambiguous.

Because lit is used primarily for graphics, it is extremely unlikely anyone is intentionally using it for double or integer values.

clang/test/SemaHLSL/BuiltIns/lit-errors-16bit.hlsl

clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl

clang/lib/Headers/hlsl/hlsl_compat_overloads.h

farzonl · 2025-04-07T22:51:00Z

We should have half and float overloads for lit and let the rest be ambiguous.

Because lit is used primarily for graphics, it is extremely unlikely anyone is intentionally using it for double or integer values.

I think we are all in agreement about not resolving lit cases outside of half and float. The question is how to do it.

Deric and I were of the opinion that we shouldn't do templates and instead just define a function overload for half and float.
#134171 (comment)
#134171 (comment)

Kaitlin astutely pointed out that lit in dxc doesn't support implicit vector truncation. So we have been trying to find a way to get the same restrictions to just half and float using templates. Thats how we landed on the is_arithmetic approach for HLSL202X and deleting the generic template for 202Y.

Given your feedback here it sounds like you want us to drop the template and allow for implicit vector truncation so that the error will be ambiguous instead of call to deleted function 'lit'?

llvm-beanz · 2025-04-08T17:23:42Z

Given your feedback here it sounds like you want us to drop the template and allow for implicit vector truncation so that the error will be ambiguous instead of call to deleted function 'lit'?

Yes, let's not add compatibility overloads unless we have driving reasons for them.

…`hlsl_intrinsics.h`

farzonl · 2025-04-08T18:44:43Z

clang/test/CodeGenHLSL/builtins/lit.hlsl

@@ -0,0 +1,34 @@
+// 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


Don't do -01 here. Drop the check-next and just check for the instructions you are expecting like
select, or, exp, log, and select again

Just for future reference, when do we want to use -O1 and/or -disable-llvm-passes in tests? I haven't been able to figure out a consistent rule looking through the other codegen tests.

Generally you should prefer -disable-llvm-passes since it makes the test run faster

clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl

Closes llvm#99135. Tasks completed: - Wrote implementation in `hlsl_intrinsics.h`/`hlsl_intrinsic_helpers.h` - Added codegen tests to `clang/test/CodeGenHLSL/builtins/lit.hlsl`

llvmbot added clang Clang issues not falling into any other category backend:X86 clang:headers Headers provided by Clang, e.g. for intrinsics HLSL HLSL Language Support labels Apr 2, 2025

github-project-automation bot added this to HLSL Support Apr 2, 2025

spall reviewed Apr 2, 2025

View reviewed changes

clang/lib/Headers/hlsl/hlsl_intrinsics.h Outdated Show resolved Hide resolved

clang/test/CodeGenHLSL/builtins/lit.hlsl Outdated Show resolved Hide resolved

farzonl reviewed Apr 2, 2025

View reviewed changes

clang/lib/Headers/hlsl/hlsl_compat_overloads.h Outdated Show resolved Hide resolved

farzonl reviewed Apr 3, 2025

View reviewed changes

clang/lib/Headers/hlsl/hlsl_intrinsics.h Outdated Show resolved Hide resolved

Icohedron reviewed Apr 3, 2025

View reviewed changes

clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h Outdated Show resolved Hide resolved

clang/lib/Headers/hlsl/hlsl_intrinsics.h Outdated Show resolved Hide resolved

clang/lib/Headers/hlsl.h Outdated Show resolved Hide resolved

Icohedron reviewed Apr 7, 2025

View reviewed changes

clang/test/CodeGenHLSL/builtins/lit-overloads.hlsl Outdated Show resolved Hide resolved

Icohedron reviewed Apr 7, 2025

View reviewed changes

clang/test/SemaHLSL/BuiltIns/lit-errors-hlsl202y.hlsl Outdated Show resolved Hide resolved

llvm-beanz reviewed Apr 7, 2025

View reviewed changes

clang/test/SemaHLSL/BuiltIns/lit-errors-16bit.hlsl Outdated Show resolved Hide resolved

clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl Outdated Show resolved Hide resolved

llvm-beanz reviewed Apr 7, 2025

View reviewed changes

clang/lib/Headers/hlsl/hlsl_compat_overloads.h Outdated Show resolved Hide resolved

kmpeng added 9 commits April 8, 2025 11:13

finished lit implementation, added codegen and sema tests

d878048

rename variables

282dd49

WIP accept double/int inputs and downcast to floats

9905d25

WIP double/int overloads

d9d1731

WIP fix double/int overloads

9064f10

finished double/int overloads and overload tests

6026d8f

address PR comments - formatting changes

0f6d9bf

address PR comments - turn clang-format off for or line in lit_impl

409fd4c

allow bool overload, add bool test

199e9b1

kmpeng added 3 commits April 8, 2025 11:13

template specialization

3f8e6a5

add >= hlsl202y primary template definition, add hlsl202y sema tests

321e54e

remove compat overloads and corresponding tests, remove templates in …

df05a3d

…`hlsl_intrinsics.h`

kmpeng force-pushed the lit-intrinsic branch from 02c6726 to df05a3d Compare April 8, 2025 18:13

Icohedron approved these changes Apr 8, 2025

View reviewed changes

farzonl reviewed Apr 8, 2025

View reviewed changes

clang/test/SemaHLSL/BuiltIns/lit-errors.hlsl Outdated Show resolved Hide resolved

remove -O1 from lit.hlsl run command, remove lit-errors.hlsl

584df9e

farzonl approved these changes Apr 9, 2025

View reviewed changes

kmpeng merged commit 2ab2276 into llvm:main Apr 9, 2025
11 checks passed

damyanp moved this to Closed in HLSL Support Apr 25, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[HLSL] Implement the `lit` intrinsic #134171

[HLSL] Implement the `lit` intrinsic #134171

kmpeng commented Apr 2, 2025 •

edited

Loading

llvmbot commented Apr 2, 2025 •

edited

Loading

llvmbot commented Apr 2, 2025

github-actions bot commented Apr 3, 2025 •

edited

Loading

farzonl commented Apr 7, 2025

Icohedron commented Apr 7, 2025

llvm-beanz commented Apr 7, 2025

farzonl commented Apr 7, 2025 •

edited

Loading

llvm-beanz commented Apr 8, 2025

farzonl Apr 8, 2025

kmpeng Apr 8, 2025

Icohedron Apr 9, 2025

		@@ -0,0 +1,34 @@
		// 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

[HLSL] Implement the lit intrinsic #134171

[HLSL] Implement the lit intrinsic #134171

Conversation

kmpeng commented Apr 2, 2025 • edited Loading

llvmbot commented Apr 2, 2025 • edited Loading

llvmbot commented Apr 2, 2025

github-actions bot commented Apr 3, 2025 • edited Loading

farzonl commented Apr 7, 2025

Icohedron commented Apr 7, 2025

llvm-beanz commented Apr 7, 2025

farzonl commented Apr 7, 2025 • edited Loading

llvm-beanz commented Apr 8, 2025

farzonl Apr 8, 2025

Choose a reason for hiding this comment

kmpeng Apr 8, 2025

Choose a reason for hiding this comment

Icohedron Apr 9, 2025

Choose a reason for hiding this comment

[HLSL] Implement the `lit` intrinsic #134171

[HLSL] Implement the `lit` intrinsic #134171

kmpeng commented Apr 2, 2025 •

edited

Loading

llvmbot commented Apr 2, 2025 •

edited

Loading

github-actions bot commented Apr 3, 2025 •

edited

Loading

farzonl commented Apr 7, 2025 •

edited

Loading