[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-21 Thread Sumit Agarwal via cfe-commits 


@@ -2473,11 +2473,9 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned 
BuiltinID, CallExpr *TheCall) {
 break;
   }
   case Builtin::BI__builtin_hlsl_dot2add: {
-// Check number of arguments should be 3
 if (SemaRef.checkArgCount(TheCall, 3))
   return true;
 
-// Check first two arguments are vector of length 2 with half data type
 auto checkHalfVectorOfSize2 = [](clang::QualType PassedType) -> bool {

sumitsays wrote:

Makes sense. I am thinking on it a bit stretched: Now we don't need to do any 
semantics check (just need to set the type) because we are explicitly 
specifying the exact overload.
Isn't it? 

https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-21 Thread Sumit Agarwal via cfe-commits 


@@ -0,0 +1,36 @@
+// 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_too_few_arg() {
+  return __builtin_hlsl_dot2add();
+  // expected-error@-1 {{too few arguments to function call, expected 3, have 
0}}
+}
+
+float test_too_many_arg(half2 p1, half2 p2, float p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3, p1);
+  // expected-error@-1 {{too many arguments to function call, expected 3, have 
4}}
+}
+
+float test_float_arg2_type(half2 p1, float2 p2, float p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3);
+  // expected-error@-1 {{passing 'float2' (aka 'vector') to 
parameter of incompatible type '__attribute__((__vector_size__(2 * 
sizeof(half half' (vector of 2 'half' values)}}
+}
+
+float test_float_arg1_type(float2 p1, half2 p2, float p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3);
+  // expected-error@-1 {{passing 'float2' (aka 'vector') to 
parameter of incompatible type '__attribute__((__vector_size__(2 * 
sizeof(half half' (vector of 2 'half' values)}}
+}
+
+float test_double_arg3_type(half2 p1, half2 p2, double p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3);
+  // expected-error@-1 {{passing 'double' to parameter of incompatible type 
'float'}}
+}
+
+float test_float_arg1_arg2_type(float2 p1, float2 p2, float p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3);
+  // expected-error@-1 {{passing 'float2' (aka 'vector') to 
parameter of incompatible type '__attribute__((__vector_size__(2 * 
sizeof(half half' (vector of 2 'half' values)}}
+}
+
+float test_int16_arg1_arg2_type(int16_t2 p1, int16_t2 p2, float p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3);
+  // expected-error@-1 {{passing 'int16_t2' (aka 'vector') to 
parameter of incompatible type '__attribute__((__vector_size__(2 * 
sizeof(half half' (vector of 2 'half' values)}}
+}

sumitsays wrote:

I tried testing `dot2add` and found that it is not throwing error when we pass 
incorrect type. Then I tried adding `__detail::enable_if` attribute in 
`hlsl_intrinsic.h` but realized that we can only use `__details::enable_if` 
when the intrinsic is templated. 
So, I thought even if we are not testing `dot2add` directly, the end user (when 
use `dot2add` from hlsl) will still get an error because ultimately the 
corresponding builtin will get called.

https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] dot2add working for dxil without sema check (PR #131237)

2025-03-20 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays updated 
https://github.com/llvm/llvm-project/pull/131237

>From 6d5c4053c90975d64e378e52779dab9c3ffb64cd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal 
Date: Thu, 13 Mar 2025 16:02:32 -0700
Subject: [PATCH 1/5] dot2add working for dxil without sema check

---
 clang/include/clang/Basic/Builtins.td  |   6 +
 clang/lib/CodeGen/CGBuiltin.cpp|  10 +
 clang/lib/CodeGen/CGHLSLRuntime.h  |   1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h   | 366 +
 llvm/include/llvm/IR/IntrinsicsDirectX.td  |   4 +
 llvm/include/llvm/IR/IntrinsicsSPIRV.td|   5 +
 llvm/lib/Target/DirectX/DXIL.td|  11 +
 llvm/lib/Target/DirectX/DXILOpLowering.cpp |  33 ++
 8 files changed, 436 insertions(+)

diff --git a/clang/include/clang/Basic/Builtins.td 
b/clang/include/clang/Basic/Builtins.td
index 72a5e495c4059..8a0237171f7df 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4891,6 +4891,12 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c126f88b9e3a5..8c0f42e0c85f7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19681,6 +19681,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned 
BuiltinID,
 getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
 ArrayRef{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+Value *A = EmitScalarExpr(E->getArg(0));
+Value *B = EmitScalarExpr(E->getArg(1));
+Value *C = EmitScalarExpr(E->getArg(2));
+
+Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+return Builder.CreateIntrinsic(
+/*ReturnType=*/C->getType(), ID, ArrayRef{A, B, C}, nullptr,
+"hlsl.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
 Value *A = EmitScalarExpr(E->getArg(0));
 Value *B = EmitScalarExpr(E->getArg(1));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h 
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 23f187b24284f..5ba69819e5431 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,6 +99,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h 
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..3849e018d 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,6 +118,372 @@ const inline float 
distance(__detail::HLSL_FIXED_VECTOR X,
 }
 
 
//===--===//
+<<< Updated upstream
+===
+// dot product builtins
+//===--===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3);

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-20 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays updated 
https://github.com/llvm/llvm-project/pull/131237

>From 6d5c4053c90975d64e378e52779dab9c3ffb64cd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal 
Date: Thu, 13 Mar 2025 16:02:32 -0700
Subject: [PATCH 1/7] dot2add working for dxil without sema check

---
 clang/include/clang/Basic/Builtins.td  |   6 +
 clang/lib/CodeGen/CGBuiltin.cpp|  10 +
 clang/lib/CodeGen/CGHLSLRuntime.h  |   1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h   | 366 +
 llvm/include/llvm/IR/IntrinsicsDirectX.td  |   4 +
 llvm/include/llvm/IR/IntrinsicsSPIRV.td|   5 +
 llvm/lib/Target/DirectX/DXIL.td|  11 +
 llvm/lib/Target/DirectX/DXILOpLowering.cpp |  33 ++
 8 files changed, 436 insertions(+)

diff --git a/clang/include/clang/Basic/Builtins.td 
b/clang/include/clang/Basic/Builtins.td
index 72a5e495c4059..8a0237171f7df 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4891,6 +4891,12 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c126f88b9e3a5..8c0f42e0c85f7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19681,6 +19681,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned 
BuiltinID,
 getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
 ArrayRef{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+Value *A = EmitScalarExpr(E->getArg(0));
+Value *B = EmitScalarExpr(E->getArg(1));
+Value *C = EmitScalarExpr(E->getArg(2));
+
+Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+return Builder.CreateIntrinsic(
+/*ReturnType=*/C->getType(), ID, ArrayRef{A, B, C}, nullptr,
+"hlsl.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
 Value *A = EmitScalarExpr(E->getArg(0));
 Value *B = EmitScalarExpr(E->getArg(1));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h 
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 23f187b24284f..5ba69819e5431 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,6 +99,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h 
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..3849e018d 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,6 +118,372 @@ const inline float 
distance(__detail::HLSL_FIXED_VECTOR X,
 }
 
 
//===--===//
+<<< Updated upstream
+===
+// dot product builtins
+//===--===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3);

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-20 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] dot2add working for dxil without sema check (PR #131237)

2025-03-20 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays updated 
https://github.com/llvm/llvm-project/pull/131237

>From 6d5c4053c90975d64e378e52779dab9c3ffb64cd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal 
Date: Thu, 13 Mar 2025 16:02:32 -0700
Subject: [PATCH 1/6] dot2add working for dxil without sema check

---
 clang/include/clang/Basic/Builtins.td  |   6 +
 clang/lib/CodeGen/CGBuiltin.cpp|  10 +
 clang/lib/CodeGen/CGHLSLRuntime.h  |   1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h   | 366 +
 llvm/include/llvm/IR/IntrinsicsDirectX.td  |   4 +
 llvm/include/llvm/IR/IntrinsicsSPIRV.td|   5 +
 llvm/lib/Target/DirectX/DXIL.td|  11 +
 llvm/lib/Target/DirectX/DXILOpLowering.cpp |  33 ++
 8 files changed, 436 insertions(+)

diff --git a/clang/include/clang/Basic/Builtins.td 
b/clang/include/clang/Basic/Builtins.td
index 72a5e495c4059..8a0237171f7df 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4891,6 +4891,12 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c126f88b9e3a5..8c0f42e0c85f7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19681,6 +19681,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned 
BuiltinID,
 getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
 ArrayRef{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+Value *A = EmitScalarExpr(E->getArg(0));
+Value *B = EmitScalarExpr(E->getArg(1));
+Value *C = EmitScalarExpr(E->getArg(2));
+
+Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+return Builder.CreateIntrinsic(
+/*ReturnType=*/C->getType(), ID, ArrayRef{A, B, C}, nullptr,
+"hlsl.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
 Value *A = EmitScalarExpr(E->getArg(0));
 Value *B = EmitScalarExpr(E->getArg(1));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h 
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 23f187b24284f..5ba69819e5431 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,6 +99,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h 
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..3849e018d 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,6 +118,372 @@ const inline float 
distance(__detail::HLSL_FIXED_VECTOR X,
 }
 
 
//===--===//
+<<< Updated upstream
+===
+// dot product builtins
+//===--===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3);

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-20 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays ready_for_review 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-20 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-20 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-20 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-20 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-21 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-21 Thread Sumit Agarwal via cfe-commits 


@@ -0,0 +1,8 @@
+; RUN: opt -S -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-compute %s | 
FileCheck %s
+
+define noundef float @dot2add_simple(<2 x half> noundef %a, <2 x half> noundef 
%b, float %c) {
+entry:
+; CHECK: call float @dx.op.dot2AddHalf(i32 162, float %c, half %0, half %1, 
half %2, half %3)
+  %ret = call float @llvm.dx.dot2add(<2 x half> %a, <2 x half> %b, float %c)

sumitsays wrote:

Thank you @V-FEXrt for pointing the official documentation. My one cent is also 
towards not having half in the name as there is only 1 possible overload. At 
the same time, I am fine with either names.

https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-21 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays updated 
https://github.com/llvm/llvm-project/pull/131237

>From 6d5c4053c90975d64e378e52779dab9c3ffb64cd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal 
Date: Thu, 13 Mar 2025 16:02:32 -0700
Subject: [PATCH 1/9] dot2add working for dxil without sema check

---
 clang/include/clang/Basic/Builtins.td  |   6 +
 clang/lib/CodeGen/CGBuiltin.cpp|  10 +
 clang/lib/CodeGen/CGHLSLRuntime.h  |   1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h   | 366 +
 llvm/include/llvm/IR/IntrinsicsDirectX.td  |   4 +
 llvm/include/llvm/IR/IntrinsicsSPIRV.td|   5 +
 llvm/lib/Target/DirectX/DXIL.td|  11 +
 llvm/lib/Target/DirectX/DXILOpLowering.cpp |  33 ++
 8 files changed, 436 insertions(+)

diff --git a/clang/include/clang/Basic/Builtins.td 
b/clang/include/clang/Basic/Builtins.td
index 72a5e495c4059..8a0237171f7df 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4891,6 +4891,12 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c126f88b9e3a5..8c0f42e0c85f7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19681,6 +19681,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned 
BuiltinID,
 getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
 ArrayRef{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+Value *A = EmitScalarExpr(E->getArg(0));
+Value *B = EmitScalarExpr(E->getArg(1));
+Value *C = EmitScalarExpr(E->getArg(2));
+
+Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+return Builder.CreateIntrinsic(
+/*ReturnType=*/C->getType(), ID, ArrayRef{A, B, C}, nullptr,
+"hlsl.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
 Value *A = EmitScalarExpr(E->getArg(0));
 Value *B = EmitScalarExpr(E->getArg(1));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h 
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 23f187b24284f..5ba69819e5431 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,6 +99,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h 
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..3849e018d 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,6 +118,372 @@ const inline float 
distance(__detail::HLSL_FIXED_VECTOR X,
 }
 
 
//===--===//
+<<< Updated upstream
+===
+// dot product builtins
+//===--===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3);

[clang] [llvm] dot2add working for dxil without sema check (PR #131237)

2025-03-22 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays updated 
https://github.com/llvm/llvm-project/pull/131237

>From 6d5c4053c90975d64e378e52779dab9c3ffb64cd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal 
Date: Thu, 13 Mar 2025 16:02:32 -0700
Subject: [PATCH 1/4] dot2add working for dxil without sema check

---
 clang/include/clang/Basic/Builtins.td  |   6 +
 clang/lib/CodeGen/CGBuiltin.cpp|  10 +
 clang/lib/CodeGen/CGHLSLRuntime.h  |   1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h   | 366 +
 llvm/include/llvm/IR/IntrinsicsDirectX.td  |   4 +
 llvm/include/llvm/IR/IntrinsicsSPIRV.td|   5 +
 llvm/lib/Target/DirectX/DXIL.td|  11 +
 llvm/lib/Target/DirectX/DXILOpLowering.cpp |  33 ++
 8 files changed, 436 insertions(+)

diff --git a/clang/include/clang/Basic/Builtins.td 
b/clang/include/clang/Basic/Builtins.td
index 72a5e495c4059..8a0237171f7df 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4891,6 +4891,12 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c126f88b9e3a5..8c0f42e0c85f7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19681,6 +19681,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned 
BuiltinID,
 getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
 ArrayRef{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+Value *A = EmitScalarExpr(E->getArg(0));
+Value *B = EmitScalarExpr(E->getArg(1));
+Value *C = EmitScalarExpr(E->getArg(2));
+
+Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+return Builder.CreateIntrinsic(
+/*ReturnType=*/C->getType(), ID, ArrayRef{A, B, C}, nullptr,
+"hlsl.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
 Value *A = EmitScalarExpr(E->getArg(0));
 Value *B = EmitScalarExpr(E->getArg(1));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h 
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 23f187b24284f..5ba69819e5431 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,6 +99,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h 
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..3849e018d 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,6 +118,372 @@ const inline float 
distance(__detail::HLSL_FIXED_VECTOR X,
 }
 
 
//===--===//
+<<< Updated upstream
+===
+// dot product builtins
+//===--===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3);

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-24 Thread Sumit Agarwal via cfe-commits 


@@ -2473,11 +2473,9 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned 
BuiltinID, CallExpr *TheCall) {
 break;
   }
   case Builtin::BI__builtin_hlsl_dot2add: {
-// Check number of arguments should be 3
 if (SemaRef.checkArgCount(TheCall, 3))
   return true;
 
-// Check first two arguments are vector of length 2 with half data type
 auto checkHalfVectorOfSize2 = [](clang::QualType PassedType) -> bool {

sumitsays wrote:

Surprisingly removing these checks does not throw semantic errors, which it 
should have.

https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-24 Thread Sumit Agarwal via cfe-commits 


@@ -4894,7 +4894,7 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
 def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot2add"];
   let Attributes = [NoThrow, Const, CustomTypeChecking];

sumitsays wrote:

`clang.exe` is not able to compile 
`clang/test/CodeGenHLSL/builtins/dot2add.hlsl` even after passing the flag 
`-fnative-half-type` saying 

`error: passing 'vector<_Float16, 2>' (vector of 2 '_Float16' values) to 
parameter of incompatible type '__attribute__((__vector_size__(2 * 
sizeof(half half' (vector of 2 'half' values)`.

https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-04-04 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-04-05 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-04-05 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-04-05 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-04-05 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-04-05 Thread Sumit Agarwal via cfe-commits 


@@ -4894,7 +4894,7 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
 def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot2add"];
   let Attributes = [NoThrow, Const, CustomTypeChecking];

sumitsays wrote:

Removing this makes Codegen unhappy. It is not able to implicitly convert 
`half` to `_Float16_t`. I think we need to find a way to use `half` here. 

https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-04-04 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays edited 
https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-27 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays updated 
https://github.com/llvm/llvm-project/pull/131237

>From 6d5c4053c90975d64e378e52779dab9c3ffb64cd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal 
Date: Thu, 13 Mar 2025 16:02:32 -0700
Subject: [PATCH 01/12] dot2add working for dxil without sema check

---
 clang/include/clang/Basic/Builtins.td  |   6 +
 clang/lib/CodeGen/CGBuiltin.cpp|  10 +
 clang/lib/CodeGen/CGHLSLRuntime.h  |   1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h   | 366 +
 llvm/include/llvm/IR/IntrinsicsDirectX.td  |   4 +
 llvm/include/llvm/IR/IntrinsicsSPIRV.td|   5 +
 llvm/lib/Target/DirectX/DXIL.td|  11 +
 llvm/lib/Target/DirectX/DXILOpLowering.cpp |  33 ++
 8 files changed, 436 insertions(+)

diff --git a/clang/include/clang/Basic/Builtins.td 
b/clang/include/clang/Basic/Builtins.td
index 72a5e495c4059..8a0237171f7df 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4891,6 +4891,12 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c126f88b9e3a5..8c0f42e0c85f7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19681,6 +19681,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned 
BuiltinID,
 getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
 ArrayRef{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+Value *A = EmitScalarExpr(E->getArg(0));
+Value *B = EmitScalarExpr(E->getArg(1));
+Value *C = EmitScalarExpr(E->getArg(2));
+
+Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+return Builder.CreateIntrinsic(
+/*ReturnType=*/C->getType(), ID, ArrayRef{A, B, C}, nullptr,
+"hlsl.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
 Value *A = EmitScalarExpr(E->getArg(0));
 Value *B = EmitScalarExpr(E->getArg(1));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h 
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 23f187b24284f..5ba69819e5431 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,6 +99,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h 
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..3849e018d 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,6 +118,372 @@ const inline float 
distance(__detail::HLSL_FIXED_VECTOR X,
 }
 
 
//===--===//
+<<< Updated upstream
+===
+// dot product builtins
+//===--===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-27 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays updated 
https://github.com/llvm/llvm-project/pull/131237

>From 6d5c4053c90975d64e378e52779dab9c3ffb64cd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal 
Date: Thu, 13 Mar 2025 16:02:32 -0700
Subject: [PATCH 01/13] dot2add working for dxil without sema check

---
 clang/include/clang/Basic/Builtins.td  |   6 +
 clang/lib/CodeGen/CGBuiltin.cpp|  10 +
 clang/lib/CodeGen/CGHLSLRuntime.h  |   1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h   | 366 +
 llvm/include/llvm/IR/IntrinsicsDirectX.td  |   4 +
 llvm/include/llvm/IR/IntrinsicsSPIRV.td|   5 +
 llvm/lib/Target/DirectX/DXIL.td|  11 +
 llvm/lib/Target/DirectX/DXILOpLowering.cpp |  33 ++
 8 files changed, 436 insertions(+)

diff --git a/clang/include/clang/Basic/Builtins.td 
b/clang/include/clang/Basic/Builtins.td
index 72a5e495c4059..8a0237171f7df 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4891,6 +4891,12 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c126f88b9e3a5..8c0f42e0c85f7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19681,6 +19681,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned 
BuiltinID,
 getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
 ArrayRef{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+Value *A = EmitScalarExpr(E->getArg(0));
+Value *B = EmitScalarExpr(E->getArg(1));
+Value *C = EmitScalarExpr(E->getArg(2));
+
+Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+return Builder.CreateIntrinsic(
+/*ReturnType=*/C->getType(), ID, ArrayRef{A, B, C}, nullptr,
+"hlsl.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
 Value *A = EmitScalarExpr(E->getArg(0));
 Value *B = EmitScalarExpr(E->getArg(1));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h 
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 23f187b24284f..5ba69819e5431 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,6 +99,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h 
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..3849e018d 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,6 +118,372 @@ const inline float 
distance(__detail::HLSL_FIXED_VECTOR X,
 }
 
 
//===--===//
+<<< Updated upstream
+===
+// dot product builtins
+//===--===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-04-03 Thread Sumit Agarwal via cfe-commits 


@@ -175,6 +175,21 @@ const inline float 
distance(__detail::HLSL_FIXED_VECTOR X,
   return __detail::distance_vec_impl(X, Y);
 }
 
+//===--===//
+// dot2add builtins

sumitsays wrote:

The general pattern is to use `builtins` in this file for all other intrinsics. 
Is there any specific reason you think it should be `builtin` just for this 
intrinsic?

https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-26 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays updated 
https://github.com/llvm/llvm-project/pull/131237

>From 6d5c4053c90975d64e378e52779dab9c3ffb64cd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal 
Date: Thu, 13 Mar 2025 16:02:32 -0700
Subject: [PATCH 01/12] dot2add working for dxil without sema check

---
 clang/include/clang/Basic/Builtins.td  |   6 +
 clang/lib/CodeGen/CGBuiltin.cpp|  10 +
 clang/lib/CodeGen/CGHLSLRuntime.h  |   1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h   | 366 +
 llvm/include/llvm/IR/IntrinsicsDirectX.td  |   4 +
 llvm/include/llvm/IR/IntrinsicsSPIRV.td|   5 +
 llvm/lib/Target/DirectX/DXIL.td|  11 +
 llvm/lib/Target/DirectX/DXILOpLowering.cpp |  33 ++
 8 files changed, 436 insertions(+)

diff --git a/clang/include/clang/Basic/Builtins.td 
b/clang/include/clang/Basic/Builtins.td
index 72a5e495c4059..8a0237171f7df 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4891,6 +4891,12 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c126f88b9e3a5..8c0f42e0c85f7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19681,6 +19681,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned 
BuiltinID,
 getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
 ArrayRef{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+Value *A = EmitScalarExpr(E->getArg(0));
+Value *B = EmitScalarExpr(E->getArg(1));
+Value *C = EmitScalarExpr(E->getArg(2));
+
+Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+return Builder.CreateIntrinsic(
+/*ReturnType=*/C->getType(), ID, ArrayRef{A, B, C}, nullptr,
+"hlsl.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
 Value *A = EmitScalarExpr(E->getArg(0));
 Value *B = EmitScalarExpr(E->getArg(1));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h 
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 23f187b24284f..5ba69819e5431 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,6 +99,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h 
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..3849e018d 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,6 +118,372 @@ const inline float 
distance(__detail::HLSL_FIXED_VECTOR X,
 }
 
 
//===--===//
+<<< Updated upstream
+===
+// dot product builtins
+//===--===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-03-25 Thread Sumit Agarwal via cfe-commits 

https://github.com/sumitsays updated 
https://github.com/llvm/llvm-project/pull/131237

>From 6d5c4053c90975d64e378e52779dab9c3ffb64cd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal 
Date: Thu, 13 Mar 2025 16:02:32 -0700
Subject: [PATCH 1/8] dot2add working for dxil without sema check

---
 clang/include/clang/Basic/Builtins.td  |   6 +
 clang/lib/CodeGen/CGBuiltin.cpp|  10 +
 clang/lib/CodeGen/CGHLSLRuntime.h  |   1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h   | 366 +
 llvm/include/llvm/IR/IntrinsicsDirectX.td  |   4 +
 llvm/include/llvm/IR/IntrinsicsSPIRV.td|   5 +
 llvm/lib/Target/DirectX/DXIL.td|  11 +
 llvm/lib/Target/DirectX/DXILOpLowering.cpp |  33 ++
 8 files changed, 436 insertions(+)

diff --git a/clang/include/clang/Basic/Builtins.td 
b/clang/include/clang/Basic/Builtins.td
index 72a5e495c4059..8a0237171f7df 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4891,6 +4891,12 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c126f88b9e3a5..8c0f42e0c85f7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19681,6 +19681,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned 
BuiltinID,
 getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
 ArrayRef{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+Value *A = EmitScalarExpr(E->getArg(0));
+Value *B = EmitScalarExpr(E->getArg(1));
+Value *C = EmitScalarExpr(E->getArg(2));
+
+Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+return Builder.CreateIntrinsic(
+/*ReturnType=*/C->getType(), ID, ArrayRef{A, B, C}, nullptr,
+"hlsl.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
 Value *A = EmitScalarExpr(E->getArg(0));
 Value *B = EmitScalarExpr(E->getArg(1));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h 
b/clang/lib/CodeGen/CGHLSLRuntime.h
index 23f187b24284f..5ba69819e5431 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,6 +99,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h 
b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..3849e018d 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,6 +118,372 @@ const inline float 
distance(__detail::HLSL_FIXED_VECTOR X,
 }
 
 
//===--===//
+<<< Updated upstream
+===
+// dot product builtins
+//===--===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3);

[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

2025-04-04 Thread Sumit Agarwal via cfe-commits 

sumitsays wrote:

> Please run clang format

Ran it and it didn't modify any files.

![image](https://github.com/user-attachments/assets/8ccb4429-2ce6-4dae-8da2-65381cd6efde)


https://github.com/llvm/llvm-project/pull/131237
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits