[clang] [llvm] [HLSL] Implement `WaveReadLaneAt` intrinsic (PR #111010)

Finn Plummer via cfe-commits cfe-commits at lists.llvm.org
Fri Oct 11 15:53:41 PDT 2024


https://github.com/inbelic updated https://github.com/llvm/llvm-project/pull/111010

>From 70089645ec5cf62b491a56df96ec46f4328fbc11 Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Thu, 3 Oct 2024 11:43:51 -0700
Subject: [PATCH 01/14] [HLSL] Implement `WaveReadLaneAt` intrinsic

    - create a clang built-in in Builtins.td
    - add semantic checking in SemaHLSL.cpp
    - link the WaveReadLaneAt api in hlsl_intrinsics.h
    - add lowering to spirv backend op GroupNonUniformShuffle
      with Scope = 2 (Group) in SPIRVInstructionSelector.cpp

    - add tests for HLSL intrinsic lowering to spirv intrinsic in
      WaveReadLaneAt.hlsl
    - add tests for sema checks in WaveReadLaneAt-errors.hlsl
    - add spir-v backend tests in WaveReadLaneAt.ll
---
 clang/include/clang/Basic/Builtins.td         |  6 +++
 clang/lib/CodeGen/CGBuiltin.cpp               | 16 ++++++++
 clang/lib/CodeGen/CGHLSLRuntime.h             |  1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h      |  7 ++++
 clang/lib/Sema/SemaHLSL.cpp                   | 20 ++++++++++
 .../CodeGenHLSL/builtins/WaveReadLaneAt.hlsl  | 40 +++++++++++++++++++
 .../BuiltIns/WaveReadLaneAt-errors.hlsl       | 21 ++++++++++
 llvm/include/llvm/IR/IntrinsicsSPIRV.td       |  1 +
 .../Target/SPIRV/SPIRVInstructionSelector.cpp | 15 +++++++
 .../SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll   | 28 +++++++++++++
 10 files changed, 155 insertions(+)
 create mode 100644 clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
 create mode 100644 clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
 create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 8090119e512fbb..eec9acd4d27d7d 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4703,6 +4703,12 @@ def HLSLWaveIsFirstLane : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "bool()";
 }
 
+def HLSLWaveReadLaneAt : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_wave_read_lane_at"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLClamp : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_elementwise_clamp"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index da3eca73bfb575..dff56af9282e9d 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18835,6 +18835,22 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
     Intrinsic::ID ID = CGM.getHLSLRuntime().getWaveIsFirstLaneIntrinsic();
     return EmitRuntimeCall(Intrinsic::getDeclaration(&CGM.getModule(), ID));
   }
+  case Builtin::BI__builtin_hlsl_wave_read_lane_at: {
+    // Due to the use of variadic arguments we must explicitly retreive them and
+    // create our function type.
+    Value *OpExpr = EmitScalarExpr(E->getArg(0));
+    Value *OpIndex = EmitScalarExpr(E->getArg(1));
+    llvm::FunctionType *FT = llvm::FunctionType::get(
+        OpExpr->getType(), ArrayRef{OpExpr->getType(), OpIndex->getType()},
+        false);
+
+    // Get overloaded name
+    std::string name =
+        Intrinsic::getName(CGM.getHLSLRuntime().getWaveReadLaneAtIntrinsic(),
+                           ArrayRef{OpExpr->getType()}, &CGM.getModule());
+    return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, name, {}, false, true),
+                           ArrayRef{OpExpr, OpIndex}, "hlsl.wave.read.lane.at");
+  }
   case Builtin::BI__builtin_hlsl_elementwise_sign: {
     Value *Op0 = EmitScalarExpr(E->getArg(0));
     llvm::Type *Xty = Op0->getType();
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index a8aabca7348ffb..a639ce2d784f4a 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -87,6 +87,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveIsFirstLane, wave_is_first_lane)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(WaveReadLaneAt, wave_read_lane_at)
 
   //===----------------------------------------------------------------------===//
   // End of reserved area for HLSL intrinsic getters.
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 810a16d75f0228..a7bdc353ae71bf 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2015,6 +2015,13 @@ _HLSL_AVAILABILITY(shadermodel, 6.0)
 _HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_is_first_lane)
 __attribute__((convergent)) bool WaveIsFirstLane();
 
+// \brief Returns the value of the expression for the given lane index within
+// the specified wave.
+template <typename T>
+_HLSL_AVAILABILITY(shadermodel, 6.0)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+    __attribute__((convergent)) T WaveReadLaneAt(T, int32_t);
+
 //===----------------------------------------------------------------------===//
 // sign builtins
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 43cc6c81ae5cb0..d54da3fd8375ed 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1956,6 +1956,26 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
       return true;
     break;
   }
+  case Builtin::BI__builtin_hlsl_wave_read_lane_at: {
+    if (SemaRef.checkArgCount(TheCall, 2))
+      return true;
+
+    // Ensure index parameter type can be interpreted as a uint
+    ExprResult Index = TheCall->getArg(1);
+    QualType ArgTyIndex = Index.get()->getType();
+    if (!ArgTyIndex->hasIntegerRepresentation()) {
+      SemaRef.Diag(TheCall->getArg(1)->getBeginLoc(),
+                   diag::err_typecheck_convert_incompatible)
+          << ArgTyIndex << SemaRef.Context.UnsignedIntTy << 1 << 0 << 0;
+      return true;
+    }
+
+    // Ensure return type is the same as the input expr type
+    ExprResult Expr = TheCall->getArg(0);
+    QualType ArgTyExpr = Expr.get()->getType();
+    TheCall->setType(ArgTyExpr);
+    break;
+  }
   case Builtin::BI__builtin_elementwise_acos:
   case Builtin::BI__builtin_elementwise_asin:
   case Builtin::BI__builtin_elementwise_atan:
diff --git a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
new file mode 100644
index 00000000000000..62319ebc04e2db
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
@@ -0,0 +1,40 @@
+// RUN: %clang_cc1 -std=hlsl2021 -finclude-default-header -x hlsl -triple   \
+// RUN:   dxil-pc-shadermodel6.3-compute %s -emit-llvm -disable-llvm-passes -o - | \
+// RUN:   FileCheck %s --check-prefixes=CHECK,CHECK-DXIL
+// RUN: %clang_cc1 -std=hlsl2021 -finclude-default-header -x hlsl -triple   \
+// RUN:   spirv-pc-vulkan-compute %s -emit-llvm -disable-llvm-passes -o - | \
+// RUN:   FileCheck %s --check-prefixes=CHECK,CHECK-SPIRV
+
+// Test basic lowering to runtime function call.
+
+// CHECK-LABEL: test_int
+int test_int(int expr, uint idx) {
+  // CHECK-SPIRV: %[[#entry_tok:]] = call token @llvm.experimental.convergence.entry()
+
+  // CHECK-SPIRV:  %[[RET:.*]] = call [[TY:.*]] @llvm.spv.wave.read.lane.at.i32([[TY]] %[[#]], i32 %[[#]])
+  // CHECK-DXIL:  %[[RET:.*]] = call [[TY:.*]] @llvm.dx.wave.read.lane.at.i32([[TY]] %[[#]], i32 %[[#]])
+
+  // CHECK:  ret [[TY]] %[[RET]]
+  return WaveReadLaneAt(expr, idx);
+}
+
+// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.read.lane.at.i32([[TY]], i32) #[[#attr:]]
+// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.read.lane.at.i32([[TY]], i32) #[[#attr:]]
+
+// Test basic lowering to runtime function call with array and float value.
+
+// CHECK-LABEL: test_floatv4
+float4 test_floatv4(float4 expr, uint idx) {
+  // CHECK-SPIRV: %[[#entry_tok1:]] = call token @llvm.experimental.convergence.entry()
+
+  // CHECK-SPIRV:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.spv.wave.read.lane.at.v4f32([[TY1]] %[[#]], i32 %[[#]])
+  // CHECK-DXIL:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.dx.wave.read.lane.at.v4f32([[TY1]] %[[#]], i32 %[[#]])
+
+  // CHECK:  ret [[TY1]] %[[RET1]]
+  return WaveReadLaneAt(expr, idx);
+}
+
+// CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.read.lane.at.v4f32([[TY1]], i32) #[[#attr]]
+// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.read.lane.at.v4f32([[TY1]], i32) #[[#attr]]
+
+// CHECK: attributes #[[#attr]] = {{{.*}} convergent {{.*}}}
diff --git a/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
new file mode 100644
index 00000000000000..451f2d3a563287
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
@@ -0,0 +1,21 @@
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only -disable-llvm-passes -verify -verify-ignore-unexpected
+
+bool test_too_few_arg() {
+  return __builtin_hlsl_wave_read_lane_at();
+  // expected-error at -1 {{too few arguments to function call, expected 2, have 0}}
+}
+
+float2 test_too_few_arg_1(float2 p0) {
+  return __builtin_hlsl_wave_read_lane_at(p0);
+  // expected-error at -1 {{too few arguments to function call, expected 2, have 1}}
+}
+
+float2 test_too_many_arg(float2 p0) {
+  return __builtin_hlsl_wave_read_lane_at(p0, p0, p0);
+  // expected-error at -1 {{too many arguments to function call, expected 2, have 3}}
+}
+
+float3 test_index_type_check(float3 p0, double idx) {
+  return __builtin_hlsl_wave_read_lane_at(p0, idx);
+  // expected-error at -1 {{passing 'double' to parameter of incompatible type 'unsigned int'}}
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 7ff3d58690ba75..b6ea9ce9b1411e 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -82,5 +82,6 @@ let TargetPrefix = "spv" in {
     [llvm_anyint_ty, LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
     [IntrNoMem, Commutative] >;
   def int_spv_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
+  def int_spv_wave_read_lane_at : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent]>;
   def int_spv_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty]>;
 }
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 7a565249a342d1..a7279193764fa4 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -2653,6 +2653,21 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg,
         .addUse(GR.getSPIRVTypeID(ResType))
         .addUse(GR.getOrCreateConstInt(3, I, IntTy, TII));
   }
+  case Intrinsic::spv_wave_read_lane_at: {
+    assert(I.getNumOperands() == 4);
+    assert(I.getOperand(2).isReg());
+    assert(I.getOperand(3).isReg());
+
+    // Defines the execution scope currently 2 for group, see scope table
+    SPIRVType *IntTy = GR.getOrCreateSPIRVIntegerType(32, I, TII);
+    return BuildMI(BB, I, I.getDebugLoc(),
+                   TII.get(SPIRV::OpGroupNonUniformShuffle))
+        .addDef(ResVReg)
+        .addUse(GR.getSPIRVTypeID(ResType))
+        .addUse(I.getOperand(2).getReg())
+        .addUse(I.getOperand(3).getReg())
+        .addUse(GR.getOrCreateConstInt(2, I, IntTy, TII));
+  }
   case Intrinsic::spv_step:
     return selectStep(ResVReg, ResType, I);
   // Discard intrinsics which we do not expect to actually represent code after
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
new file mode 100644
index 00000000000000..e02a2907ee28a0
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
@@ -0,0 +1,28 @@
+; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv32v1.3-vulkan-unknown %s -o - | FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-vulkan-unknown %s -o - -filetype=obj | spirv-val %}
+
+; Test lowering to spir-v backend
+
+; CHECK-DAG:   %[[#uint:]] = OpTypeInt 32 0
+; CHECK-DAG:   %[[#scope:]] = OpConstant %[[#uint]] 2
+; CHECK-DAG:   %[[#f32:]] = OpTypeFloat 32
+; CHECK-DAG:   %[[#expr:]] = OpFunctionParameter %[[#f32]]
+; CHECK-DAG:   %[[#idx:]] = OpFunctionParameter %[[#uint]]
+
+define spir_func void @test_1(float %expr, i32 %idx) #0 {
+entry:
+  %0 = call token @llvm.experimental.convergence.entry()
+; CHECK:   %[[#ret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#expr]] %[[#idx]] %[[#scope]]
+  %1 = call float @llvm.spv.wave.read.lane.at(float %expr, i32 %idx) [ "convergencectrl"(token %0) ]
+  ret void
+}
+
+declare i32 @__hlsl_wave_get_lane_index() #1
+
+attributes #0 = { convergent norecurse "hlsl.numthreads"="1,1,1" "hlsl.shader"="compute" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #1 = { convergent }
+
+!llvm.module.flags = !{!0, !1}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 4, !"dx.disable_optimizations", i32 1}

>From e8c5dba828f25c2894980631ae019936a3d9e9f6 Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Thu, 3 Oct 2024 12:45:50 -0700
Subject: [PATCH 02/14] [DXIL] add WaveReadLaneAt intrinsic for scalars

    - add WaveReadLaneAt intrinsic to IntrinsicsDirectX.td and mapping
      to DXIL.td

    - add test to show scalar functionality

    - note that this doesn't include support for the scalarizer to
      handle this function will be added in a future pr
---
 llvm/include/llvm/IR/IntrinsicsDirectX.td   |  1 +
 llvm/lib/Target/DirectX/DXIL.td             | 10 ++++
 llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll | 56 +++++++++++++++++++++
 3 files changed, 67 insertions(+)
 create mode 100644 llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll

diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index 555877e7aaf0e5..a12526539e7efe 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -83,6 +83,7 @@ def int_dx_umad : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLV
 def int_dx_normalize : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty], [IntrNoMem]>;
 def int_dx_rsqrt  : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
 def int_dx_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
+def int_dx_wave_read_lane_at : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent]>;
 def int_dx_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
 def int_dx_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
 }
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index 9aa0af3e3a6b17..7283d069988123 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -801,3 +801,13 @@ def WaveIsFirstLane :  DXILOp<110, waveIsFirstLane> {
   let stages = [Stages<DXIL1_0, [all_stages]>];
   let attributes = [Attributes<DXIL1_0, [ReadNone]>];
 }
+
+def WaveReadLaneAt:  DXILOp<117, waveIsFirstLane> {
+  let Doc = "returns the value from the specified lane";
+  let LLVMIntrinsic = int_dx_wave_read_lane_at;
+  let arguments = [OverloadTy, Int32Ty];
+  let result = OverloadTy;
+  let overloads = [Overloads<DXIL1_0, [HalfTy, FloatTy, DoubleTy, Int1Ty, Int16Ty, Int32Ty]>];
+  let stages = [Stages<DXIL1_0, [all_stages]>];
+  let attributes = [Attributes<DXIL1_0, [ReadNone]>];
+}
diff --git a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
new file mode 100644
index 00000000000000..7f524923136349
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
@@ -0,0 +1,56 @@
+; RUN: opt -S -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-compute %s | FileCheck %s
+
+; Test that for scalar values, WaveReadLaneAt maps down to the DirectX op
+
+define noundef half @wave_rla_half(half noundef %expr, i32 noundef %idx) #0 {
+entry:
+; CHECK: call half @dx.op.waveReadLaneAt.f16(i32 117, half %expr, i32 %idx)
+  %ret = call half @llvm.dx.wave.read.lane.at.f16(half %expr, i32 %idx)
+  ret half %ret
+}
+
+define noundef float @wave_rla_float(float noundef %expr, i32 noundef %idx) #0 {
+entry:
+; CHECK: call float @dx.op.waveReadLaneAt.f32(i32 117, float %expr, i32 %idx)
+  %ret = call float @llvm.dx.wave.read.lane.at(float %expr, i32 %idx)
+  ret float %ret
+}
+
+define noundef double @wave_rla_double(double noundef %expr, i32 noundef %idx) #0 {
+entry:
+; CHECK: call double @dx.op.waveReadLaneAt.f64(i32 117, double %expr, i32 %idx)
+  %ret = call double @llvm.dx.wave.read.lane.at(double %expr, i32 %idx)
+  ret double %ret
+}
+
+define noundef i1 @wave_rla_i1(i1 noundef %expr, i32 noundef %idx) #0 {
+entry:
+; CHECK: call i1 @dx.op.waveReadLaneAt.i1(i32 117, i1 %expr, i32 %idx)
+  %ret = call i1 @llvm.dx.wave.read.lane.at.i1(i1 %expr, i32 %idx)
+  ret i1 %ret
+}
+
+define noundef i16 @wave_rla_i16(i16 noundef %expr, i32 noundef %idx) #0 {
+entry:
+; CHECK: call i16 @dx.op.waveReadLaneAt.i16(i32 117, i16 %expr, i32 %idx)
+  %ret = call i16 @llvm.dx.wave.read.lane.at.i16(i16 %expr, i32 %idx)
+  ret i16 %ret
+}
+
+define noundef i32 @wave_rla_i32(i32 noundef %expr, i32 noundef %idx) #0 {
+entry:
+; CHECK: call i32 @dx.op.waveReadLaneAt.i32(i32 117, i32 %expr, i32 %idx)
+  %ret = call i32 @llvm.dx.wave.read.lane.at.i32(i32 %expr, i32 %idx)
+  ret i32 %ret
+}
+
+declare half @llvm.dx.wave.read.lane.at.f16(half, i32) #1
+declare float @llvm.dx.wave.read.lane.at.f32(float, i32) #1
+declare double @llvm.dx.wave.read.lane.at.f64(double, i32) #1
+
+declare i1 @llvm.dx.wave.read.lane.at.i1(i1, i32) #1
+declare i16 @llvm.dx.wave.read.lane.at.i16(i16, i32) #1
+declare i32 @llvm.dx.wave.read.lane.at.i32(i32, i32) #1
+
+attributes #0 = { norecurse "hlsl.numthreads"="1,1,1" "hlsl.shader"="compute" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #1 = { nocallback nofree nosync nounwind willreturn }

>From 1b0746f152b4eaa4f6e661787503fc6f7d1fdc72 Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Fri, 4 Oct 2024 08:05:38 -0700
Subject: [PATCH 03/14] review comments

    - add check for "convergencectrl" token in hlsl -> spirv intrinsic
    - correct the execution scope of the spirv instruction, add
      description
    - fix typo
---
 clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl       | 4 ++--
 llvm/lib/Target/DirectX/DXIL.td                           | 2 +-
 llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp        | 5 +++--
 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll | 2 +-
 4 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
index 62319ebc04e2db..ce2b0fce3c32a3 100644
--- a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
@@ -11,7 +11,7 @@
 int test_int(int expr, uint idx) {
   // CHECK-SPIRV: %[[#entry_tok:]] = call token @llvm.experimental.convergence.entry()
 
-  // CHECK-SPIRV:  %[[RET:.*]] = call [[TY:.*]] @llvm.spv.wave.read.lane.at.i32([[TY]] %[[#]], i32 %[[#]])
+  // CHECK-SPIRV:  %[[RET:.*]] = call [[TY:.*]] @llvm.spv.wave.read.lane.at.i32([[TY]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok]]) ]
   // CHECK-DXIL:  %[[RET:.*]] = call [[TY:.*]] @llvm.dx.wave.read.lane.at.i32([[TY]] %[[#]], i32 %[[#]])
 
   // CHECK:  ret [[TY]] %[[RET]]
@@ -27,7 +27,7 @@ int test_int(int expr, uint idx) {
 float4 test_floatv4(float4 expr, uint idx) {
   // CHECK-SPIRV: %[[#entry_tok1:]] = call token @llvm.experimental.convergence.entry()
 
-  // CHECK-SPIRV:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.spv.wave.read.lane.at.v4f32([[TY1]] %[[#]], i32 %[[#]])
+  // CHECK-SPIRV:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.spv.wave.read.lane.at.v4f32([[TY1]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok1]]) ]
   // CHECK-DXIL:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.dx.wave.read.lane.at.v4f32([[TY1]] %[[#]], i32 %[[#]])
 
   // CHECK:  ret [[TY1]] %[[RET1]]
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index 7283d069988123..fbb9cd41bdcc1e 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -802,7 +802,7 @@ def WaveIsFirstLane :  DXILOp<110, waveIsFirstLane> {
   let attributes = [Attributes<DXIL1_0, [ReadNone]>];
 }
 
-def WaveReadLaneAt:  DXILOp<117, waveIsFirstLane> {
+def WaveReadLaneAt:  DXILOp<117, waveReadLaneAt> {
   let Doc = "returns the value from the specified lane";
   let LLVMIntrinsic = int_dx_wave_read_lane_at;
   let arguments = [OverloadTy, Int32Ty];
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index a7279193764fa4..680965c22400d1 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -2658,7 +2658,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg,
     assert(I.getOperand(2).isReg());
     assert(I.getOperand(3).isReg());
 
-    // Defines the execution scope currently 2 for group, see scope table
+    // IntTy is used to define the execution scope, set to 3 to denote a
+    // cross-lane interaction equivalent to a SPIR-V subgroup.
     SPIRVType *IntTy = GR.getOrCreateSPIRVIntegerType(32, I, TII);
     return BuildMI(BB, I, I.getDebugLoc(),
                    TII.get(SPIRV::OpGroupNonUniformShuffle))
@@ -2666,7 +2667,7 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg,
         .addUse(GR.getSPIRVTypeID(ResType))
         .addUse(I.getOperand(2).getReg())
         .addUse(I.getOperand(3).getReg())
-        .addUse(GR.getOrCreateConstInt(2, I, IntTy, TII));
+        .addUse(GR.getOrCreateConstInt(3, I, IntTy, TII));
   }
   case Intrinsic::spv_step:
     return selectStep(ResVReg, ResType, I);
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
index e02a2907ee28a0..df6643c605d129 100644
--- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
@@ -4,7 +4,7 @@
 ; Test lowering to spir-v backend
 
 ; CHECK-DAG:   %[[#uint:]] = OpTypeInt 32 0
-; CHECK-DAG:   %[[#scope:]] = OpConstant %[[#uint]] 2
+; CHECK-DAG:   %[[#scope:]] = OpConstant %[[#uint]] 3
 ; CHECK-DAG:   %[[#f32:]] = OpTypeFloat 32
 ; CHECK-DAG:   %[[#expr:]] = OpFunctionParameter %[[#f32]]
 ; CHECK-DAG:   %[[#idx:]] = OpFunctionParameter %[[#uint]]

>From 473150df115944655bbe5d23bbfe8b5e1106c604 Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Fri, 4 Oct 2024 09:53:06 -0700
Subject: [PATCH 04/14] review comments

    - extend spirv testcase to include i32 and vector of bools
---
 .../SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll   | 20 ++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
index df6643c605d129..c40fed4f860038 100644
--- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
@@ -6,18 +6,28 @@
 ; CHECK-DAG:   %[[#uint:]] = OpTypeInt 32 0
 ; CHECK-DAG:   %[[#scope:]] = OpConstant %[[#uint]] 3
 ; CHECK-DAG:   %[[#f32:]] = OpTypeFloat 32
-; CHECK-DAG:   %[[#expr:]] = OpFunctionParameter %[[#f32]]
+; CHECK-DAG:   %[[#bool:]] = OpTypeBool
+; CHECK-DAG:   %[[#v4_bool:]] = OpTypeVector %[[#bool]] 4
+; CHECK-DAG:   %[[#fexpr:]] = OpFunctionParameter %[[#f32]]
+; CHECK-DAG:   %[[#iexpr:]] = OpFunctionParameter %[[#uint]]
 ; CHECK-DAG:   %[[#idx:]] = OpFunctionParameter %[[#uint]]
+; CHECK-DAG:   %[[#vbexpr:]] = OpFunctionParameter %[[#v4_bool]]
 
-define spir_func void @test_1(float %expr, i32 %idx) #0 {
+define spir_func void @test_1(float %fexpr, i32 %iexpr, <4 x i1> %vbexpr, i32 %idx) #0 {
 entry:
   %0 = call token @llvm.experimental.convergence.entry()
-; CHECK:   %[[#ret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#expr]] %[[#idx]] %[[#scope]]
-  %1 = call float @llvm.spv.wave.read.lane.at(float %expr, i32 %idx) [ "convergencectrl"(token %0) ]
+; CHECK:   %[[#fret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#fexpr]] %[[#idx]] %[[#scope]]
+  %1 = call float @llvm.spv.wave.read.lane.at.f32(float %fexpr, i32 %idx) [ "convergencectrl"(token %0) ]
+; CHECK:   %[[#iret:]] = OpGroupNonUniformShuffle %[[#uint]] %[[#iexpr]] %[[#idx]] %[[#scope]]
+  %2 = call i32 @llvm.spv.wave.read.lane.at.i32(i32 %iexpr, i32 %idx) [ "convergencectrl"(token %0) ]
+; CHECK:   %[[#vbret:]] = OpGroupNonUniformShuffle %[[#v4_bool]] %[[#vbexpr]] %[[#idx]] %[[#scope]]
+  %3 = call <4 x i1> @llvm.spv.wave.read.lane.at.v4i1(<4 x i1> %vbexpr, i32 %idx) [ "convergencectrl"(token %0) ]
   ret void
 }
 
-declare i32 @__hlsl_wave_get_lane_index() #1
+declare float @__hlsl_wave_read_lane_at.f32(float, i32) #1
+declare i32 @__hlsl_wave_read_lane_at.i32(i32, i32) #1
+declare <4 x i1> @__hlsl_wave_read_lane_at.v4i1(<4 x i1>, i32) #1
 
 attributes #0 = { convergent norecurse "hlsl.numthreads"="1,1,1" "hlsl.shader"="compute" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
 attributes #1 = { convergent }

>From 484b2089ae9f0a0fae09b6d68b70171f1007bade Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Fri, 4 Oct 2024 15:04:36 -0700
Subject: [PATCH 05/14] review comments

    - remove unneeded attributes from test functions
    - provide context to parameters in CGBuiltin
    - update intrinsic naming for clarity
---
 clang/lib/CodeGen/CGBuiltin.cpp               |  6 ++-
 .../CodeGenHLSL/builtins/WaveReadLaneAt.hlsl  |  4 --
 .../Target/SPIRV/SPIRVInstructionSelector.cpp | 41 +++++++++++--------
 llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll   | 27 ++++++------
 .../SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll   | 23 ++++-------
 5 files changed, 48 insertions(+), 53 deletions(-)

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index dff56af9282e9d..fef9b1667241ee 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18848,8 +18848,10 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
     std::string name =
         Intrinsic::getName(CGM.getHLSLRuntime().getWaveReadLaneAtIntrinsic(),
                            ArrayRef{OpExpr->getType()}, &CGM.getModule());
-    return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, name, {}, false, true),
-                           ArrayRef{OpExpr, OpIndex}, "hlsl.wave.read.lane.at");
+    return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, name, {},
+                                                     /*Local=*/false,
+                                                     /*AssumeConvergent=*/true),
+                           ArrayRef{OpExpr, OpIndex}, "hlsl.waveReadLaneAt");
   }
   case Builtin::BI__builtin_hlsl_elementwise_sign: {
     Value *Op0 = EmitScalarExpr(E->getArg(0));
diff --git a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
index ce2b0fce3c32a3..e0c0e48bab7941 100644
--- a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
@@ -10,10 +10,8 @@
 // CHECK-LABEL: test_int
 int test_int(int expr, uint idx) {
   // CHECK-SPIRV: %[[#entry_tok:]] = call token @llvm.experimental.convergence.entry()
-
   // CHECK-SPIRV:  %[[RET:.*]] = call [[TY:.*]] @llvm.spv.wave.read.lane.at.i32([[TY]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok]]) ]
   // CHECK-DXIL:  %[[RET:.*]] = call [[TY:.*]] @llvm.dx.wave.read.lane.at.i32([[TY]] %[[#]], i32 %[[#]])
-
   // CHECK:  ret [[TY]] %[[RET]]
   return WaveReadLaneAt(expr, idx);
 }
@@ -26,10 +24,8 @@ int test_int(int expr, uint idx) {
 // CHECK-LABEL: test_floatv4
 float4 test_floatv4(float4 expr, uint idx) {
   // CHECK-SPIRV: %[[#entry_tok1:]] = call token @llvm.experimental.convergence.entry()
-
   // CHECK-SPIRV:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.spv.wave.read.lane.at.v4f32([[TY1]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok1]]) ]
   // CHECK-DXIL:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.dx.wave.read.lane.at.v4f32([[TY1]] %[[#]], i32 %[[#]])
-
   // CHECK:  ret [[TY1]] %[[RET1]]
   return WaveReadLaneAt(expr, idx);
 }
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 680965c22400d1..d6a2717f557d9f 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -244,6 +244,9 @@ class SPIRVInstructionSelector : public InstructionSelector {
   bool selectSpvThreadId(Register ResVReg, const SPIRVType *ResType,
                          MachineInstr &I) const;
 
+  bool selectWaveReadLaneAt(Register ResVReg, const SPIRVType *ResType,
+                            MachineInstr &I) const;
+
   bool selectStep(Register ResVReg, const SPIRVType *ResType,
                   MachineInstr &I) const;
 
@@ -1853,6 +1856,26 @@ bool SPIRVInstructionSelector::selectSign(Register ResVReg,
   return Result;
 }
 
+bool SPIRVInstructionSelector::selectWaveReadLaneAt(Register ResVReg,
+                                                    const SPIRVType *ResType,
+                                                    MachineInstr &I) const {
+  assert(I.getNumOperands() == 4);
+  assert(I.getOperand(2).isReg());
+  assert(I.getOperand(3).isReg());
+  MachineBasicBlock &BB = *I.getParent();
+
+  // IntTy is used to define the execution scope, set to 3 to denote a
+  // cross-lane interaction equivalent to a SPIR-V subgroup.
+  SPIRVType *IntTy = GR.getOrCreateSPIRVIntegerType(32, I, TII);
+  return BuildMI(BB, I, I.getDebugLoc(),
+                 TII.get(SPIRV::OpGroupNonUniformShuffle))
+      .addDef(ResVReg)
+      .addUse(GR.getSPIRVTypeID(ResType))
+      .addUse(I.getOperand(2).getReg())
+      .addUse(I.getOperand(3).getReg())
+      .addUse(GR.getOrCreateConstInt(3, I, IntTy, TII));
+}
+
 bool SPIRVInstructionSelector::selectStep(Register ResVReg,
                                           const SPIRVType *ResType,
                                           MachineInstr &I) const {
@@ -2653,22 +2676,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg,
         .addUse(GR.getSPIRVTypeID(ResType))
         .addUse(GR.getOrCreateConstInt(3, I, IntTy, TII));
   }
-  case Intrinsic::spv_wave_read_lane_at: {
-    assert(I.getNumOperands() == 4);
-    assert(I.getOperand(2).isReg());
-    assert(I.getOperand(3).isReg());
-
-    // IntTy is used to define the execution scope, set to 3 to denote a
-    // cross-lane interaction equivalent to a SPIR-V subgroup.
-    SPIRVType *IntTy = GR.getOrCreateSPIRVIntegerType(32, I, TII);
-    return BuildMI(BB, I, I.getDebugLoc(),
-                   TII.get(SPIRV::OpGroupNonUniformShuffle))
-        .addDef(ResVReg)
-        .addUse(GR.getSPIRVTypeID(ResType))
-        .addUse(I.getOperand(2).getReg())
-        .addUse(I.getOperand(3).getReg())
-        .addUse(GR.getOrCreateConstInt(3, I, IntTy, TII));
-  }
+  case Intrinsic::spv_wave_read_lane_at:
+    return selectWaveReadLaneAt(ResVReg, ResType, I);
   case Intrinsic::spv_step:
     return selectStep(ResVReg, ResType, I);
   // Discard intrinsics which we do not expect to actually represent code after
diff --git a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
index 7f524923136349..4d2d72e064e61e 100644
--- a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
@@ -2,55 +2,52 @@
 
 ; Test that for scalar values, WaveReadLaneAt maps down to the DirectX op
 
-define noundef half @wave_rla_half(half noundef %expr, i32 noundef %idx) #0 {
+define noundef half @wave_rla_half(half noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call half @dx.op.waveReadLaneAt.f16(i32 117, half %expr, i32 %idx)
   %ret = call half @llvm.dx.wave.read.lane.at.f16(half %expr, i32 %idx)
   ret half %ret
 }
 
-define noundef float @wave_rla_float(float noundef %expr, i32 noundef %idx) #0 {
+define noundef float @wave_rla_float(float noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call float @dx.op.waveReadLaneAt.f32(i32 117, float %expr, i32 %idx)
   %ret = call float @llvm.dx.wave.read.lane.at(float %expr, i32 %idx)
   ret float %ret
 }
 
-define noundef double @wave_rla_double(double noundef %expr, i32 noundef %idx) #0 {
+define noundef double @wave_rla_double(double noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call double @dx.op.waveReadLaneAt.f64(i32 117, double %expr, i32 %idx)
   %ret = call double @llvm.dx.wave.read.lane.at(double %expr, i32 %idx)
   ret double %ret
 }
 
-define noundef i1 @wave_rla_i1(i1 noundef %expr, i32 noundef %idx) #0 {
+define noundef i1 @wave_rla_i1(i1 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i1 @dx.op.waveReadLaneAt.i1(i32 117, i1 %expr, i32 %idx)
   %ret = call i1 @llvm.dx.wave.read.lane.at.i1(i1 %expr, i32 %idx)
   ret i1 %ret
 }
 
-define noundef i16 @wave_rla_i16(i16 noundef %expr, i32 noundef %idx) #0 {
+define noundef i16 @wave_rla_i16(i16 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i16 @dx.op.waveReadLaneAt.i16(i32 117, i16 %expr, i32 %idx)
   %ret = call i16 @llvm.dx.wave.read.lane.at.i16(i16 %expr, i32 %idx)
   ret i16 %ret
 }
 
-define noundef i32 @wave_rla_i32(i32 noundef %expr, i32 noundef %idx) #0 {
+define noundef i32 @wave_rla_i32(i32 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i32 @dx.op.waveReadLaneAt.i32(i32 117, i32 %expr, i32 %idx)
   %ret = call i32 @llvm.dx.wave.read.lane.at.i32(i32 %expr, i32 %idx)
   ret i32 %ret
 }
 
-declare half @llvm.dx.wave.read.lane.at.f16(half, i32) #1
-declare float @llvm.dx.wave.read.lane.at.f32(float, i32) #1
-declare double @llvm.dx.wave.read.lane.at.f64(double, i32) #1
+declare half @llvm.dx.wave.read.lane.at.f16(half, i32)
+declare float @llvm.dx.wave.read.lane.at.f32(float, i32)
+declare double @llvm.dx.wave.read.lane.at.f64(double, i32)
 
-declare i1 @llvm.dx.wave.read.lane.at.i1(i1, i32) #1
-declare i16 @llvm.dx.wave.read.lane.at.i16(i16, i32) #1
-declare i32 @llvm.dx.wave.read.lane.at.i32(i32, i32) #1
-
-attributes #0 = { norecurse "hlsl.numthreads"="1,1,1" "hlsl.shader"="compute" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #1 = { nocallback nofree nosync nounwind willreturn }
+declare i1 @llvm.dx.wave.read.lane.at.i1(i1, i32)
+declare i16 @llvm.dx.wave.read.lane.at.i16(i16, i32)
+declare i32 @llvm.dx.wave.read.lane.at.i32(i32, i32)
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
index c40fed4f860038..506df2e0572c5d 100644
--- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
@@ -13,26 +13,17 @@
 ; CHECK-DAG:   %[[#idx:]] = OpFunctionParameter %[[#uint]]
 ; CHECK-DAG:   %[[#vbexpr:]] = OpFunctionParameter %[[#v4_bool]]
 
-define spir_func void @test_1(float %fexpr, i32 %iexpr, <4 x i1> %vbexpr, i32 %idx) #0 {
+define spir_func void @test_1(float %fexpr, i32 %iexpr, <4 x i1> %vbexpr, i32 %idx) {
 entry:
-  %0 = call token @llvm.experimental.convergence.entry()
 ; CHECK:   %[[#fret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#fexpr]] %[[#idx]] %[[#scope]]
-  %1 = call float @llvm.spv.wave.read.lane.at.f32(float %fexpr, i32 %idx) [ "convergencectrl"(token %0) ]
+  %0 = call float @llvm.spv.wave.read.lane.at.f32(float %fexpr, i32 %idx)
 ; CHECK:   %[[#iret:]] = OpGroupNonUniformShuffle %[[#uint]] %[[#iexpr]] %[[#idx]] %[[#scope]]
-  %2 = call i32 @llvm.spv.wave.read.lane.at.i32(i32 %iexpr, i32 %idx) [ "convergencectrl"(token %0) ]
+  %1 = call i32 @llvm.spv.wave.read.lane.at.i32(i32 %iexpr, i32 %idx)
 ; CHECK:   %[[#vbret:]] = OpGroupNonUniformShuffle %[[#v4_bool]] %[[#vbexpr]] %[[#idx]] %[[#scope]]
-  %3 = call <4 x i1> @llvm.spv.wave.read.lane.at.v4i1(<4 x i1> %vbexpr, i32 %idx) [ "convergencectrl"(token %0) ]
+  %2 = call <4 x i1> @llvm.spv.wave.read.lane.at.v4i1(<4 x i1> %vbexpr, i32 %idx)
   ret void
 }
 
-declare float @__hlsl_wave_read_lane_at.f32(float, i32) #1
-declare i32 @__hlsl_wave_read_lane_at.i32(i32, i32) #1
-declare <4 x i1> @__hlsl_wave_read_lane_at.v4i1(<4 x i1>, i32) #1
-
-attributes #0 = { convergent norecurse "hlsl.numthreads"="1,1,1" "hlsl.shader"="compute" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #1 = { convergent }
-
-!llvm.module.flags = !{!0, !1}
-
-!0 = !{i32 1, !"wchar_size", i32 4}
-!1 = !{i32 4, !"dx.disable_optimizations", i32 1}
+declare float @__hlsl_wave_read_lane_at.f32(float, i32)
+declare i32 @__hlsl_wave_read_lane_at.i32(i32, i32)
+declare <4 x i1> @__hlsl_wave_read_lane_at.v4i1(<4 x i1>, i32)

>From dc3fc0bd141735c262fa6fcf72aaa5243d70f459 Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Fri, 4 Oct 2024 15:35:56 -0700
Subject: [PATCH 06/14] review comments:

    - change spv/dx intrinsics naming from wave_read_lane_at to
      waveReadLaneAt
---
 clang/lib/CodeGen/CGHLSLRuntime.h             |  2 +-
 .../CodeGenHLSL/builtins/WaveReadLaneAt.hlsl  | 16 ++++++-------
 llvm/include/llvm/IR/IntrinsicsDirectX.td     |  2 +-
 llvm/include/llvm/IR/IntrinsicsSPIRV.td       |  2 +-
 llvm/lib/Target/DirectX/DXIL.td               |  2 +-
 .../Target/SPIRV/SPIRVInstructionSelector.cpp |  2 +-
 llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll   | 24 +++++++++----------
 .../SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll   | 12 +++++-----
 8 files changed, 31 insertions(+), 31 deletions(-)

diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index a639ce2d784f4a..1fa89d417729f5 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -87,7 +87,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveIsFirstLane, wave_is_first_lane)
-  GENERATE_HLSL_INTRINSIC_FUNCTION(WaveReadLaneAt, wave_read_lane_at)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(WaveReadLaneAt, waveReadLaneAt)
 
   //===----------------------------------------------------------------------===//
   // End of reserved area for HLSL intrinsic getters.
diff --git a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
index e0c0e48bab7941..0162c0bb874ca9 100644
--- a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
@@ -10,27 +10,27 @@
 // CHECK-LABEL: test_int
 int test_int(int expr, uint idx) {
   // CHECK-SPIRV: %[[#entry_tok:]] = call token @llvm.experimental.convergence.entry()
-  // CHECK-SPIRV:  %[[RET:.*]] = call [[TY:.*]] @llvm.spv.wave.read.lane.at.i32([[TY]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok]]) ]
-  // CHECK-DXIL:  %[[RET:.*]] = call [[TY:.*]] @llvm.dx.wave.read.lane.at.i32([[TY]] %[[#]], i32 %[[#]])
+  // CHECK-SPIRV:  %[[RET:.*]] = call [[TY:.*]] @llvm.spv.waveReadLaneAt.i32([[TY]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok]]) ]
+  // CHECK-DXIL:  %[[RET:.*]] = call [[TY:.*]] @llvm.dx.waveReadLaneAt.i32([[TY]] %[[#]], i32 %[[#]])
   // CHECK:  ret [[TY]] %[[RET]]
   return WaveReadLaneAt(expr, idx);
 }
 
-// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.read.lane.at.i32([[TY]], i32) #[[#attr:]]
-// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.read.lane.at.i32([[TY]], i32) #[[#attr:]]
+// CHECK-DXIL: declare [[TY]] @llvm.dx.waveReadLaneAt.i32([[TY]], i32) #[[#attr:]]
+// CHECK-SPIRV: declare [[TY]] @llvm.spv.waveReadLaneAt.i32([[TY]], i32) #[[#attr:]]
 
 // Test basic lowering to runtime function call with array and float value.
 
 // CHECK-LABEL: test_floatv4
 float4 test_floatv4(float4 expr, uint idx) {
   // CHECK-SPIRV: %[[#entry_tok1:]] = call token @llvm.experimental.convergence.entry()
-  // CHECK-SPIRV:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.spv.wave.read.lane.at.v4f32([[TY1]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok1]]) ]
-  // CHECK-DXIL:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.dx.wave.read.lane.at.v4f32([[TY1]] %[[#]], i32 %[[#]])
+  // CHECK-SPIRV:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.spv.waveReadLaneAt.v4f32([[TY1]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok1]]) ]
+  // CHECK-DXIL:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.dx.waveReadLaneAt.v4f32([[TY1]] %[[#]], i32 %[[#]])
   // CHECK:  ret [[TY1]] %[[RET1]]
   return WaveReadLaneAt(expr, idx);
 }
 
-// CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.read.lane.at.v4f32([[TY1]], i32) #[[#attr]]
-// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.read.lane.at.v4f32([[TY1]], i32) #[[#attr]]
+// CHECK-DXIL: declare [[TY1]] @llvm.dx.waveReadLaneAt.v4f32([[TY1]], i32) #[[#attr]]
+// CHECK-SPIRV: declare [[TY1]] @llvm.spv.waveReadLaneAt.v4f32([[TY1]], i32) #[[#attr]]
 
 // CHECK: attributes #[[#attr]] = {{{.*}} convergent {{.*}}}
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index a12526539e7efe..36dcdc8992087e 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -83,7 +83,7 @@ def int_dx_umad : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLV
 def int_dx_normalize : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty], [IntrNoMem]>;
 def int_dx_rsqrt  : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
 def int_dx_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
-def int_dx_wave_read_lane_at : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent]>;
+def int_dx_waveReadLaneAt : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent]>;
 def int_dx_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
 def int_dx_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
 }
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index b6ea9ce9b1411e..19ed7f4a3970b5 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -82,6 +82,6 @@ let TargetPrefix = "spv" in {
     [llvm_anyint_ty, LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
     [IntrNoMem, Commutative] >;
   def int_spv_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
-  def int_spv_wave_read_lane_at : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent]>;
+  def int_spv_waveReadLaneAt : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent]>;
   def int_spv_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty]>;
 }
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index fbb9cd41bdcc1e..2e307a0ae16da5 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -804,7 +804,7 @@ def WaveIsFirstLane :  DXILOp<110, waveIsFirstLane> {
 
 def WaveReadLaneAt:  DXILOp<117, waveReadLaneAt> {
   let Doc = "returns the value from the specified lane";
-  let LLVMIntrinsic = int_dx_wave_read_lane_at;
+  let LLVMIntrinsic = int_dx_waveReadLaneAt;
   let arguments = [OverloadTy, Int32Ty];
   let result = OverloadTy;
   let overloads = [Overloads<DXIL1_0, [HalfTy, FloatTy, DoubleTy, Int1Ty, Int16Ty, Int32Ty]>];
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index d6a2717f557d9f..229d57e47626bf 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -2676,7 +2676,7 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg,
         .addUse(GR.getSPIRVTypeID(ResType))
         .addUse(GR.getOrCreateConstInt(3, I, IntTy, TII));
   }
-  case Intrinsic::spv_wave_read_lane_at:
+  case Intrinsic::spv_waveReadLaneAt:
     return selectWaveReadLaneAt(ResVReg, ResType, I);
   case Intrinsic::spv_step:
     return selectStep(ResVReg, ResType, I);
diff --git a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
index 4d2d72e064e61e..1a1c426a4e5557 100644
--- a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
@@ -5,49 +5,49 @@
 define noundef half @wave_rla_half(half noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call half @dx.op.waveReadLaneAt.f16(i32 117, half %expr, i32 %idx)
-  %ret = call half @llvm.dx.wave.read.lane.at.f16(half %expr, i32 %idx)
+  %ret = call half @llvm.dx.waveReadLaneAt.f16(half %expr, i32 %idx)
   ret half %ret
 }
 
 define noundef float @wave_rla_float(float noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call float @dx.op.waveReadLaneAt.f32(i32 117, float %expr, i32 %idx)
-  %ret = call float @llvm.dx.wave.read.lane.at(float %expr, i32 %idx)
+  %ret = call float @llvm.dx.waveReadLaneAt(float %expr, i32 %idx)
   ret float %ret
 }
 
 define noundef double @wave_rla_double(double noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call double @dx.op.waveReadLaneAt.f64(i32 117, double %expr, i32 %idx)
-  %ret = call double @llvm.dx.wave.read.lane.at(double %expr, i32 %idx)
+  %ret = call double @llvm.dx.waveReadLaneAt(double %expr, i32 %idx)
   ret double %ret
 }
 
 define noundef i1 @wave_rla_i1(i1 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i1 @dx.op.waveReadLaneAt.i1(i32 117, i1 %expr, i32 %idx)
-  %ret = call i1 @llvm.dx.wave.read.lane.at.i1(i1 %expr, i32 %idx)
+  %ret = call i1 @llvm.dx.waveReadLaneAt.i1(i1 %expr, i32 %idx)
   ret i1 %ret
 }
 
 define noundef i16 @wave_rla_i16(i16 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i16 @dx.op.waveReadLaneAt.i16(i32 117, i16 %expr, i32 %idx)
-  %ret = call i16 @llvm.dx.wave.read.lane.at.i16(i16 %expr, i32 %idx)
+  %ret = call i16 @llvm.dx.waveReadLaneAt.i16(i16 %expr, i32 %idx)
   ret i16 %ret
 }
 
 define noundef i32 @wave_rla_i32(i32 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i32 @dx.op.waveReadLaneAt.i32(i32 117, i32 %expr, i32 %idx)
-  %ret = call i32 @llvm.dx.wave.read.lane.at.i32(i32 %expr, i32 %idx)
+  %ret = call i32 @llvm.dx.waveReadLaneAt.i32(i32 %expr, i32 %idx)
   ret i32 %ret
 }
 
-declare half @llvm.dx.wave.read.lane.at.f16(half, i32)
-declare float @llvm.dx.wave.read.lane.at.f32(float, i32)
-declare double @llvm.dx.wave.read.lane.at.f64(double, i32)
+declare half @llvm.dx.waveReadLaneAt.f16(half, i32)
+declare float @llvm.dx.waveReadLaneAt.f32(float, i32)
+declare double @llvm.dx.waveReadLaneAt.f64(double, i32)
 
-declare i1 @llvm.dx.wave.read.lane.at.i1(i1, i32)
-declare i16 @llvm.dx.wave.read.lane.at.i16(i16, i32)
-declare i32 @llvm.dx.wave.read.lane.at.i32(i32, i32)
+declare i1 @llvm.dx.waveReadLaneAt.i1(i1, i32)
+declare i16 @llvm.dx.waveReadLaneAt.i16(i16, i32)
+declare i32 @llvm.dx.waveReadLaneAt.i32(i32, i32)
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
index 506df2e0572c5d..88f3ddc9f5bc55 100644
--- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
@@ -16,14 +16,14 @@
 define spir_func void @test_1(float %fexpr, i32 %iexpr, <4 x i1> %vbexpr, i32 %idx) {
 entry:
 ; CHECK:   %[[#fret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#fexpr]] %[[#idx]] %[[#scope]]
-  %0 = call float @llvm.spv.wave.read.lane.at.f32(float %fexpr, i32 %idx)
+  %0 = call float @llvm.spv.waveReadLaneAt.f32(float %fexpr, i32 %idx)
 ; CHECK:   %[[#iret:]] = OpGroupNonUniformShuffle %[[#uint]] %[[#iexpr]] %[[#idx]] %[[#scope]]
-  %1 = call i32 @llvm.spv.wave.read.lane.at.i32(i32 %iexpr, i32 %idx)
+  %1 = call i32 @llvm.spv.waveReadLaneAt.i32(i32 %iexpr, i32 %idx)
 ; CHECK:   %[[#vbret:]] = OpGroupNonUniformShuffle %[[#v4_bool]] %[[#vbexpr]] %[[#idx]] %[[#scope]]
-  %2 = call <4 x i1> @llvm.spv.wave.read.lane.at.v4i1(<4 x i1> %vbexpr, i32 %idx)
+  %2 = call <4 x i1> @llvm.spv.waveReadLaneAt.v4i1(<4 x i1> %vbexpr, i32 %idx)
   ret void
 }
 
-declare float @__hlsl_wave_read_lane_at.f32(float, i32)
-declare i32 @__hlsl_wave_read_lane_at.i32(i32, i32)
-declare <4 x i1> @__hlsl_wave_read_lane_at.v4i1(<4 x i1>, i32)
+declare float @__hlsl_waveReadLaneAt.f32(float, i32)
+declare i32 @__hlsl_waveReadLaneAt.i32(i32, i32)
+declare <4 x i1> @__hlsl_waveReadLaneAt.v4i1(<4 x i1>, i32)

>From 5a2c5945ec293cd0d217e827b8e155b70e0af02b Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Mon, 7 Oct 2024 08:53:13 -0700
Subject: [PATCH 07/14] review comments:

    - add `IntrNoMem` to intrinsics to denote no memory access
---
 llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +-
 llvm/include/llvm/IR/IntrinsicsSPIRV.td   | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index 36dcdc8992087e..ab0e2f187bc23f 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -83,7 +83,7 @@ def int_dx_umad : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLV
 def int_dx_normalize : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty], [IntrNoMem]>;
 def int_dx_rsqrt  : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
 def int_dx_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
-def int_dx_waveReadLaneAt : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent]>;
+def int_dx_waveReadLaneAt : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
 def int_dx_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
 def int_dx_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
 }
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 19ed7f4a3970b5..0b674d1569f8f9 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -82,6 +82,6 @@ let TargetPrefix = "spv" in {
     [llvm_anyint_ty, LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
     [IntrNoMem, Commutative] >;
   def int_spv_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
-  def int_spv_waveReadLaneAt : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent]>;
+  def int_spv_waveReadLaneAt : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
   def int_spv_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty]>;
 }

>From 32f10151be4a026672d7d04879c9907594a7c6a1 Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Mon, 7 Oct 2024 17:47:59 -0700
Subject: [PATCH 08/14] review comments:

    - remove use of arbitrary type template to match the
      `gen_intrin_main` definition of any scalar or vector with a valid
      component type of bool and numeric types

    - correct type check to not allow a vector of ints
---
 clang/lib/Headers/hlsl/hlsl_intrinsics.h      | 77 ++++++++++++++++++-
 clang/lib/Sema/SemaHLSL.cpp                   |  2 +-
 .../BuiltIns/WaveReadLaneAt-errors.hlsl       |  5 ++
 .../SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll   |  6 +-
 4 files changed, 84 insertions(+), 6 deletions(-)

diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index a7bdc353ae71bf..ae37a1aa71e9c1 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -2015,12 +2015,85 @@ _HLSL_AVAILABILITY(shadermodel, 6.0)
 _HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_is_first_lane)
 __attribute__((convergent)) bool WaveIsFirstLane();
 
+//===----------------------------------------------------------------------===//
+// WaveReadLaneAt builtins
+//===----------------------------------------------------------------------===//
+
 // \brief Returns the value of the expression for the given lane index within
 // the specified wave.
-template <typename T>
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) bool WaveReadLaneAt(bool, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) bool2 WaveReadLaneAt(bool2, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) bool3 WaveReadLaneAt(bool3, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) bool4 WaveReadLaneAt(bool4, int32_t);
+
+#ifdef __HLSL_ENABLE_16_BIT
 _HLSL_AVAILABILITY(shadermodel, 6.0)
 _HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
-    __attribute__((convergent)) T WaveReadLaneAt(T, int32_t);
+__attribute__((convergent)) int16_t WaveReadLaneAt(int16_t, int32_t);
+_HLSL_AVAILABILITY(shadermodel, 6.0)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int16_t2 WaveReadLaneAt(int16_t2, int32_t);
+_HLSL_AVAILABILITY(shadermodel, 6.0)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int16_t3 WaveReadLaneAt(int16_t3, int32_t);
+_HLSL_AVAILABILITY(shadermodel, 6.0)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int16_t4 WaveReadLaneAt(int16_t4, int32_t);
+#endif
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.0)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) half WaveReadLaneAt(half, int32_t);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.0)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) half2 WaveReadLaneAt(half2, int32_t);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.0)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) half3 WaveReadLaneAt(half3, int32_t);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.0)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) half4 WaveReadLaneAt(half4, int32_t);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int WaveReadLaneAt(int, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int2 WaveReadLaneAt(int2, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int3 WaveReadLaneAt(int3, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int4 WaveReadLaneAt(int4, int32_t);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) float WaveReadLaneAt(float, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) float2 WaveReadLaneAt(float2, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) float3 WaveReadLaneAt(float3, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) float4 WaveReadLaneAt(float4, int32_t);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int64_t WaveReadLaneAt(int64_t, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int64_t2 WaveReadLaneAt(int64_t2, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int64_t3 WaveReadLaneAt(int64_t3, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) int64_t4 WaveReadLaneAt(int64_t4, int32_t);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) double WaveReadLaneAt(double, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) double2 WaveReadLaneAt(double2, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) double3 WaveReadLaneAt(double3, int32_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_read_lane_at)
+__attribute__((convergent)) double4 WaveReadLaneAt(double4, int32_t);
 
 //===----------------------------------------------------------------------===//
 // sign builtins
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index d54da3fd8375ed..0be2c220d68369 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1963,7 +1963,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
     // Ensure index parameter type can be interpreted as a uint
     ExprResult Index = TheCall->getArg(1);
     QualType ArgTyIndex = Index.get()->getType();
-    if (!ArgTyIndex->hasIntegerRepresentation()) {
+    if (!ArgTyIndex->isIntegerType()) {
       SemaRef.Diag(TheCall->getArg(1)->getBeginLoc(),
                    diag::err_typecheck_convert_incompatible)
           << ArgTyIndex << SemaRef.Context.UnsignedIntTy << 1 << 0 << 0;
diff --git a/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
index 451f2d3a563287..817f61af54b30b 100644
--- a/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
@@ -19,3 +19,8 @@ float3 test_index_type_check(float3 p0, double idx) {
   return __builtin_hlsl_wave_read_lane_at(p0, idx);
   // expected-error at -1 {{passing 'double' to parameter of incompatible type 'unsigned int'}}
 }
+
+float3 test_index_type_check(float3 p0, int3 idxs) {
+  return __builtin_hlsl_wave_read_lane_at(p0, idxs);
+  // expected-error at -1 {{passing 'int3' (aka 'vector<int, 3>') to parameter of incompatible type 'unsigned int'}}
+}
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
index 88f3ddc9f5bc55..a4660404e8b8af 100644
--- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
@@ -24,6 +24,6 @@ entry:
   ret void
 }
 
-declare float @__hlsl_waveReadLaneAt.f32(float, i32)
-declare i32 @__hlsl_waveReadLaneAt.i32(i32, i32)
-declare <4 x i1> @__hlsl_waveReadLaneAt.v4i1(<4 x i1>, i32)
+declare float @llvm.spv.waveReadLaneAt.f32(float, i32)
+declare i32 @llvm.spv.waveReadLaneAt.i32(i32, i32)
+declare <4 x i1> @llvm.spv.waveReadLaneAt.v4i1(<4 x i1>, i32)

>From 906f105f16080616ea84958e6e27761cd398e7b9 Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Tue, 8 Oct 2024 08:53:22 -0700
Subject: [PATCH 09/14] self review:

    - add support for selecting a non-side effect convergent intrinsic
    - update test case to not be optimized away
---
 .../Target/SPIRV/SPIRVInstructionSelector.cpp |  1 +
 .../SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll   | 40 ++++++++++++-------
 2 files changed, 27 insertions(+), 14 deletions(-)

diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 229d57e47626bf..fb1732847f1a1a 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -431,6 +431,7 @@ bool SPIRVInstructionSelector::spvSelect(Register ResVReg,
 
   case TargetOpcode::G_INTRINSIC:
   case TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS:
+  case TargetOpcode::G_INTRINSIC_CONVERGENT:
   case TargetOpcode::G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS:
     return selectIntrinsic(ResVReg, ResType, I);
   case TargetOpcode::G_BITREVERSE:
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
index a4660404e8b8af..59d40d04acd6e1 100644
--- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
@@ -1,27 +1,39 @@
 ; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv32v1.3-vulkan-unknown %s -o - | FileCheck %s
 ; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-vulkan-unknown %s -o - -filetype=obj | spirv-val %}
 
-; Test lowering to spir-v backend
+; Test lowering to spir-v backend for various types and scalar/vector
 
-; CHECK-DAG:   %[[#uint:]] = OpTypeInt 32 0
-; CHECK-DAG:   %[[#scope:]] = OpConstant %[[#uint]] 3
 ; CHECK-DAG:   %[[#f32:]] = OpTypeFloat 32
+; CHECK-DAG:   %[[#uint:]] = OpTypeInt 32 0
 ; CHECK-DAG:   %[[#bool:]] = OpTypeBool
 ; CHECK-DAG:   %[[#v4_bool:]] = OpTypeVector %[[#bool]] 4
-; CHECK-DAG:   %[[#fexpr:]] = OpFunctionParameter %[[#f32]]
-; CHECK-DAG:   %[[#iexpr:]] = OpFunctionParameter %[[#uint]]
-; CHECK-DAG:   %[[#idx:]] = OpFunctionParameter %[[#uint]]
-; CHECK-DAG:   %[[#vbexpr:]] = OpFunctionParameter %[[#v4_bool]]
+; CHECK-DAG:   %[[#scope:]] = OpConstant %[[#uint]] 3
 
-define spir_func void @test_1(float %fexpr, i32 %iexpr, <4 x i1> %vbexpr, i32 %idx) {
+; CHECK:   %[[#fexpr:]] = OpFunctionParameter %[[#f32]]
+; CHECK:   %[[#idx1:]] = OpFunctionParameter %[[#uint]]
+define float @test_1(float %fexpr, i32 %idx) {
 entry:
-; CHECK:   %[[#fret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#fexpr]] %[[#idx]] %[[#scope]]
+; CHECK:   %[[#fret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#fexpr]] %[[#idx1]] %[[#scope]]
   %0 = call float @llvm.spv.waveReadLaneAt.f32(float %fexpr, i32 %idx)
-; CHECK:   %[[#iret:]] = OpGroupNonUniformShuffle %[[#uint]] %[[#iexpr]] %[[#idx]] %[[#scope]]
-  %1 = call i32 @llvm.spv.waveReadLaneAt.i32(i32 %iexpr, i32 %idx)
-; CHECK:   %[[#vbret:]] = OpGroupNonUniformShuffle %[[#v4_bool]] %[[#vbexpr]] %[[#idx]] %[[#scope]]
-  %2 = call <4 x i1> @llvm.spv.waveReadLaneAt.v4i1(<4 x i1> %vbexpr, i32 %idx)
-  ret void
+  ret float %0
+}
+
+; CHECK:   %[[#iexpr:]] = OpFunctionParameter %[[#uint]]
+; CHECK:   %[[#idx2:]] = OpFunctionParameter %[[#uint]]
+define i32 @test_2(i32 %iexpr, i32 %idx) {
+entry:
+; CHECK:   %[[#iret:]] = OpGroupNonUniformShuffle %[[#uint]] %[[#iexpr]] %[[#idx2]] %[[#scope]]
+  %0 = call i32 @llvm.spv.waveReadLaneAt.i32(i32 %iexpr, i32 %idx)
+  ret i32 %0
+}
+
+; CHECK:   %[[#vbexpr:]] = OpFunctionParameter %[[#v4_bool]]
+; CHECK:   %[[#idx3:]] = OpFunctionParameter %[[#uint]]
+define <4 x i1> @test_3(<4 x i1> %vbexpr, i32 %idx) {
+entry:
+; CHECK:   %[[#vbret:]] = OpGroupNonUniformShuffle %[[#v4_bool]] %[[#vbexpr]] %[[#idx3]] %[[#scope]]
+  %0 = call <4 x i1> @llvm.spv.waveReadLaneAt.v4i1(<4 x i1> %vbexpr, i32 %idx)
+  ret <4 x i1> %0
 }
 
 declare float @llvm.spv.waveReadLaneAt.f32(float, i32)

>From 1fc05b6e910976d2e2f13f0690c3e227f161bc21 Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Tue, 8 Oct 2024 17:23:32 -0700
Subject: [PATCH 10/14] review comments:

    - fix variable name to be uppercase
    - remove unneeded flags from SemaHLSL testcase
---
 clang/lib/CodeGen/CGBuiltin.cpp                         | 4 ++--
 clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index fef9b1667241ee..d54b39e5fc69a7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18845,10 +18845,10 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
         false);
 
     // Get overloaded name
-    std::string name =
+    std::string Name =
         Intrinsic::getName(CGM.getHLSLRuntime().getWaveReadLaneAtIntrinsic(),
                            ArrayRef{OpExpr->getType()}, &CGM.getModule());
-    return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, name, {},
+    return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
                                                      /*Local=*/false,
                                                      /*AssumeConvergent=*/true),
                            ArrayRef{OpExpr, OpIndex}, "hlsl.waveReadLaneAt");
diff --git a/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
index 817f61af54b30b..e2de250c1bfa11 100644
--- a/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only -disable-llvm-passes -verify -verify-ignore-unexpected
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -emit-llvm-only -disable-llvm-passes -verify
 
 bool test_too_few_arg() {
   return __builtin_hlsl_wave_read_lane_at();

>From f181e2734298d96f92ecf00b817296115e4e7db0 Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Wed, 9 Oct 2024 09:56:45 -0700
Subject: [PATCH 11/14] review comments:

    - switch to wave_readlaneat to follow llvm naming conventions
---
 clang/lib/CodeGen/CGBuiltin.cpp               |  2 +-
 clang/lib/CodeGen/CGHLSLRuntime.h             |  2 +-
 .../CodeGenHLSL/builtins/WaveReadLaneAt.hlsl  | 16 ++++++-------
 llvm/include/llvm/IR/IntrinsicsDirectX.td     |  2 +-
 llvm/include/llvm/IR/IntrinsicsSPIRV.td       |  2 +-
 llvm/lib/Target/DirectX/DXIL.td               |  2 +-
 .../Target/SPIRV/SPIRVInstructionSelector.cpp |  2 +-
 llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll   | 24 +++++++++----------
 .../SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll   | 12 +++++-----
 9 files changed, 32 insertions(+), 32 deletions(-)

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index d54b39e5fc69a7..bdf2043ed77788 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18851,7 +18851,7 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
     return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
                                                      /*Local=*/false,
                                                      /*AssumeConvergent=*/true),
-                           ArrayRef{OpExpr, OpIndex}, "hlsl.waveReadLaneAt");
+                           ArrayRef{OpExpr, OpIndex}, "hlsl.wave.readlaneAt");
   }
   case Builtin::BI__builtin_hlsl_elementwise_sign: {
     Value *Op0 = EmitScalarExpr(E->getArg(0));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index 1fa89d417729f5..bb9e27b99be299 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -87,7 +87,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveIsFirstLane, wave_is_first_lane)
-  GENERATE_HLSL_INTRINSIC_FUNCTION(WaveReadLaneAt, waveReadLaneAt)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(WaveReadLaneAt, wave_readlaneat)
 
   //===----------------------------------------------------------------------===//
   // End of reserved area for HLSL intrinsic getters.
diff --git a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
index 0162c0bb874ca9..b2d572672161a5 100644
--- a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
@@ -10,27 +10,27 @@
 // CHECK-LABEL: test_int
 int test_int(int expr, uint idx) {
   // CHECK-SPIRV: %[[#entry_tok:]] = call token @llvm.experimental.convergence.entry()
-  // CHECK-SPIRV:  %[[RET:.*]] = call [[TY:.*]] @llvm.spv.waveReadLaneAt.i32([[TY]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok]]) ]
-  // CHECK-DXIL:  %[[RET:.*]] = call [[TY:.*]] @llvm.dx.waveReadLaneAt.i32([[TY]] %[[#]], i32 %[[#]])
+  // CHECK-SPIRV:  %[[RET:.*]] = call [[TY:.*]] @llvm.spv.wave.readlaneat.i32([[TY]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok]]) ]
+  // CHECK-DXIL:  %[[RET:.*]] = call [[TY:.*]] @llvm.dx.wave.readlaneat.i32([[TY]] %[[#]], i32 %[[#]])
   // CHECK:  ret [[TY]] %[[RET]]
   return WaveReadLaneAt(expr, idx);
 }
 
-// CHECK-DXIL: declare [[TY]] @llvm.dx.waveReadLaneAt.i32([[TY]], i32) #[[#attr:]]
-// CHECK-SPIRV: declare [[TY]] @llvm.spv.waveReadLaneAt.i32([[TY]], i32) #[[#attr:]]
+// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.readlaneat.i32([[TY]], i32) #[[#attr:]]
+// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.readlaneat.i32([[TY]], i32) #[[#attr:]]
 
 // Test basic lowering to runtime function call with array and float value.
 
 // CHECK-LABEL: test_floatv4
 float4 test_floatv4(float4 expr, uint idx) {
   // CHECK-SPIRV: %[[#entry_tok1:]] = call token @llvm.experimental.convergence.entry()
-  // CHECK-SPIRV:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.spv.waveReadLaneAt.v4f32([[TY1]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok1]]) ]
-  // CHECK-DXIL:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.dx.waveReadLaneAt.v4f32([[TY1]] %[[#]], i32 %[[#]])
+  // CHECK-SPIRV:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.spv.wave.readlaneat.v4f32([[TY1]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok1]]) ]
+  // CHECK-DXIL:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.dx.wave.readlaneat.v4f32([[TY1]] %[[#]], i32 %[[#]])
   // CHECK:  ret [[TY1]] %[[RET1]]
   return WaveReadLaneAt(expr, idx);
 }
 
-// CHECK-DXIL: declare [[TY1]] @llvm.dx.waveReadLaneAt.v4f32([[TY1]], i32) #[[#attr]]
-// CHECK-SPIRV: declare [[TY1]] @llvm.spv.waveReadLaneAt.v4f32([[TY1]], i32) #[[#attr]]
+// CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.readlaneat.v4f32([[TY1]], i32) #[[#attr]]
+// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.readlaneat.v4f32([[TY1]], i32) #[[#attr]]
 
 // CHECK: attributes #[[#attr]] = {{{.*}} convergent {{.*}}}
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index ab0e2f187bc23f..be2eca3d9ac6da 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -83,7 +83,7 @@ def int_dx_umad : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLV
 def int_dx_normalize : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty], [IntrNoMem]>;
 def int_dx_rsqrt  : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
 def int_dx_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
-def int_dx_waveReadLaneAt : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
+def int_dx_wave_readlaneat : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
 def int_dx_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
 def int_dx_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
 }
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 0b674d1569f8f9..552eebb1932a7d 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -82,6 +82,6 @@ let TargetPrefix = "spv" in {
     [llvm_anyint_ty, LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
     [IntrNoMem, Commutative] >;
   def int_spv_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
-  def int_spv_waveReadLaneAt : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
+  def int_spv_wave_readlaneat : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
   def int_spv_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty]>;
 }
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index 2e307a0ae16da5..a3a5ebd36b79e6 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -804,7 +804,7 @@ def WaveIsFirstLane :  DXILOp<110, waveIsFirstLane> {
 
 def WaveReadLaneAt:  DXILOp<117, waveReadLaneAt> {
   let Doc = "returns the value from the specified lane";
-  let LLVMIntrinsic = int_dx_waveReadLaneAt;
+  let LLVMIntrinsic = int_dx_wave_readlaneat;
   let arguments = [OverloadTy, Int32Ty];
   let result = OverloadTy;
   let overloads = [Overloads<DXIL1_0, [HalfTy, FloatTy, DoubleTy, Int1Ty, Int16Ty, Int32Ty]>];
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index fb1732847f1a1a..b470b71c969271 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -2677,7 +2677,7 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg,
         .addUse(GR.getSPIRVTypeID(ResType))
         .addUse(GR.getOrCreateConstInt(3, I, IntTy, TII));
   }
-  case Intrinsic::spv_waveReadLaneAt:
+  case Intrinsic::spv_wave_readlaneat:
     return selectWaveReadLaneAt(ResVReg, ResType, I);
   case Intrinsic::spv_step:
     return selectStep(ResVReg, ResType, I);
diff --git a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
index 1a1c426a4e5557..c390804ffdeae4 100644
--- a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
@@ -5,49 +5,49 @@
 define noundef half @wave_rla_half(half noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call half @dx.op.waveReadLaneAt.f16(i32 117, half %expr, i32 %idx)
-  %ret = call half @llvm.dx.waveReadLaneAt.f16(half %expr, i32 %idx)
+  %ret = call half @llvm.dx.wave.readlaneat.f16(half %expr, i32 %idx)
   ret half %ret
 }
 
 define noundef float @wave_rla_float(float noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call float @dx.op.waveReadLaneAt.f32(i32 117, float %expr, i32 %idx)
-  %ret = call float @llvm.dx.waveReadLaneAt(float %expr, i32 %idx)
+  %ret = call float @llvm.dx.wave.readlaneat(float %expr, i32 %idx)
   ret float %ret
 }
 
 define noundef double @wave_rla_double(double noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call double @dx.op.waveReadLaneAt.f64(i32 117, double %expr, i32 %idx)
-  %ret = call double @llvm.dx.waveReadLaneAt(double %expr, i32 %idx)
+  %ret = call double @llvm.dx.wave.readlaneat(double %expr, i32 %idx)
   ret double %ret
 }
 
 define noundef i1 @wave_rla_i1(i1 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i1 @dx.op.waveReadLaneAt.i1(i32 117, i1 %expr, i32 %idx)
-  %ret = call i1 @llvm.dx.waveReadLaneAt.i1(i1 %expr, i32 %idx)
+  %ret = call i1 @llvm.dx.wave.readlaneat.i1(i1 %expr, i32 %idx)
   ret i1 %ret
 }
 
 define noundef i16 @wave_rla_i16(i16 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i16 @dx.op.waveReadLaneAt.i16(i32 117, i16 %expr, i32 %idx)
-  %ret = call i16 @llvm.dx.waveReadLaneAt.i16(i16 %expr, i32 %idx)
+  %ret = call i16 @llvm.dx.wave.readlaneat.i16(i16 %expr, i32 %idx)
   ret i16 %ret
 }
 
 define noundef i32 @wave_rla_i32(i32 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i32 @dx.op.waveReadLaneAt.i32(i32 117, i32 %expr, i32 %idx)
-  %ret = call i32 @llvm.dx.waveReadLaneAt.i32(i32 %expr, i32 %idx)
+  %ret = call i32 @llvm.dx.wave.readlaneat.i32(i32 %expr, i32 %idx)
   ret i32 %ret
 }
 
-declare half @llvm.dx.waveReadLaneAt.f16(half, i32)
-declare float @llvm.dx.waveReadLaneAt.f32(float, i32)
-declare double @llvm.dx.waveReadLaneAt.f64(double, i32)
+declare half @llvm.dx.wave.readlaneat.f16(half, i32)
+declare float @llvm.dx.wave.readlaneat.f32(float, i32)
+declare double @llvm.dx.wave.readlaneat.f64(double, i32)
 
-declare i1 @llvm.dx.waveReadLaneAt.i1(i1, i32)
-declare i16 @llvm.dx.waveReadLaneAt.i16(i16, i32)
-declare i32 @llvm.dx.waveReadLaneAt.i32(i32, i32)
+declare i1 @llvm.dx.wave.readlaneat.i1(i1, i32)
+declare i16 @llvm.dx.wave.readlaneat.i16(i16, i32)
+declare i32 @llvm.dx.wave.readlaneat.i32(i32, i32)
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
index 59d40d04acd6e1..2810eca8907c20 100644
--- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
@@ -14,7 +14,7 @@
 define float @test_1(float %fexpr, i32 %idx) {
 entry:
 ; CHECK:   %[[#fret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#fexpr]] %[[#idx1]] %[[#scope]]
-  %0 = call float @llvm.spv.waveReadLaneAt.f32(float %fexpr, i32 %idx)
+  %0 = call float @llvm.spv.wave.readlaneat.f32(float %fexpr, i32 %idx)
   ret float %0
 }
 
@@ -23,7 +23,7 @@ entry:
 define i32 @test_2(i32 %iexpr, i32 %idx) {
 entry:
 ; CHECK:   %[[#iret:]] = OpGroupNonUniformShuffle %[[#uint]] %[[#iexpr]] %[[#idx2]] %[[#scope]]
-  %0 = call i32 @llvm.spv.waveReadLaneAt.i32(i32 %iexpr, i32 %idx)
+  %0 = call i32 @llvm.spv.wave.readlaneat.i32(i32 %iexpr, i32 %idx)
   ret i32 %0
 }
 
@@ -32,10 +32,10 @@ entry:
 define <4 x i1> @test_3(<4 x i1> %vbexpr, i32 %idx) {
 entry:
 ; CHECK:   %[[#vbret:]] = OpGroupNonUniformShuffle %[[#v4_bool]] %[[#vbexpr]] %[[#idx3]] %[[#scope]]
-  %0 = call <4 x i1> @llvm.spv.waveReadLaneAt.v4i1(<4 x i1> %vbexpr, i32 %idx)
+  %0 = call <4 x i1> @llvm.spv.wave.readlaneat.v4i1(<4 x i1> %vbexpr, i32 %idx)
   ret <4 x i1> %0
 }
 
-declare float @llvm.spv.waveReadLaneAt.f32(float, i32)
-declare i32 @llvm.spv.waveReadLaneAt.i32(i32, i32)
-declare <4 x i1> @llvm.spv.waveReadLaneAt.v4i1(<4 x i1>, i32)
+declare float @llvm.spv.wave.readlaneat.f32(float, i32)
+declare i32 @llvm.spv.wave.readlaneat.i32(i32, i32)
+declare <4 x i1> @llvm.spv.wave.readlaneat.v4i1(<4 x i1>, i32)

>From dd73a1e9aa0d05f261ff4006e49db2c6849890fd Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Thu, 10 Oct 2024 11:35:17 -0700
Subject: [PATCH 12/14] review comments:

    - change intrinsic names to `wave_readlane` to align with AMD
      implementations
    - update semantic check to ensure that only a scalar/vector is
      allowed
    - add test case to illustrate this
---
 .../clang/Basic/DiagnosticSemaKinds.td        |  2 ++
 clang/lib/CodeGen/CGBuiltin.cpp               |  2 +-
 clang/lib/CodeGen/CGHLSLRuntime.h             |  2 +-
 clang/lib/Sema/SemaHLSL.cpp                   | 21 +++++++++++++++-
 .../CodeGenHLSL/builtins/WaveReadLaneAt.hlsl  | 16 ++++++-------
 .../BuiltIns/WaveReadLaneAt-errors.hlsl       | 16 +++++++++++--
 llvm/include/llvm/IR/IntrinsicsDirectX.td     |  2 +-
 llvm/include/llvm/IR/IntrinsicsSPIRV.td       |  2 +-
 llvm/lib/Target/DirectX/DXIL.td               |  2 +-
 .../Target/SPIRV/SPIRVInstructionSelector.cpp |  2 +-
 llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll   | 24 +++++++++----------
 .../SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll   | 12 +++++-----
 12 files changed, 68 insertions(+), 35 deletions(-)

diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index f4a2d4a3f0656a..ac4481ff6a7b56 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -9232,6 +9232,8 @@ def err_typecheck_cond_incompatible_operands : Error<
 def err_typecheck_expect_scalar_or_vector : Error<
   "invalid operand of type %0 where %1 or "
   "a vector of such type is required">;
+def err_typecheck_expect_any_scalar_or_vector : Error<
+  "invalid operand of type %0 where a scalar or vector is required">;
 def err_typecheck_expect_flt_or_vector : Error<
   "invalid operand of type %0 where floating, complex or "
   "a vector of such types is required">;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 8d71f343e2bca2..3fcb82d477d51e 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18903,7 +18903,7 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
     return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
                                                      /*Local=*/false,
                                                      /*AssumeConvergent=*/true),
-                           ArrayRef{OpExpr, OpIndex}, "hlsl.wave.readlaneAt");
+                           ArrayRef{OpExpr, OpIndex}, "hlsl.wave.readlane");
   }
   case Builtin::BI__builtin_hlsl_elementwise_sign: {
     auto *Arg0 = E->getArg(0);
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index 12ea3d8c709c2d..f0f7cbc8192865 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -89,7 +89,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveIsFirstLane, wave_is_first_lane)
-  GENERATE_HLSL_INTRINSIC_FUNCTION(WaveReadLaneAt, wave_readlaneat)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(WaveReadLaneAt, wave_readlane)
 
   //===----------------------------------------------------------------------===//
   // End of reserved area for HLSL intrinsic getters.
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 52b2b908207aeb..3fa359556e0490 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1751,6 +1751,22 @@ static bool CheckScalarOrVector(Sema *S, CallExpr *TheCall, QualType Scalar,
   return false;
 }
 
+static bool CheckAnyScalarOrVector(Sema *S, CallExpr *TheCall,
+                                   unsigned ArgIndex) {
+  assert(TheCall->getNumArgs() >= ArgIndex);
+  QualType ArgType = TheCall->getArg(ArgIndex)->getType();
+  auto *VTy = ArgType->getAs<VectorType>();
+  // not the scalar or vector<scalar>
+  if (!(ArgType->isScalarType() ||
+        (VTy && VTy->getElementType()->isScalarType()))) {
+    S->Diag(TheCall->getArg(0)->getBeginLoc(),
+            diag::err_typecheck_expect_any_scalar_or_vector)
+        << ArgType;
+    return true;
+  }
+  return false;
+}
+
 static bool CheckBoolSelect(Sema *S, CallExpr *TheCall) {
   assert(TheCall->getNumArgs() == 3);
   Expr *Arg1 = TheCall->getArg(1);
@@ -2006,7 +2022,10 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
       return true;
     }
 
-    // Ensure return type is the same as the input expr type
+    // Ensure input expr type is a scalar/vector and the same as the return type
+    if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0))
+      return true;
+
     ExprResult Expr = TheCall->getArg(0);
     QualType ArgTyExpr = Expr.get()->getType();
     TheCall->setType(ArgTyExpr);
diff --git a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
index b2d572672161a5..3ab54d63670c8a 100644
--- a/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
@@ -10,27 +10,27 @@
 // CHECK-LABEL: test_int
 int test_int(int expr, uint idx) {
   // CHECK-SPIRV: %[[#entry_tok:]] = call token @llvm.experimental.convergence.entry()
-  // CHECK-SPIRV:  %[[RET:.*]] = call [[TY:.*]] @llvm.spv.wave.readlaneat.i32([[TY]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok]]) ]
-  // CHECK-DXIL:  %[[RET:.*]] = call [[TY:.*]] @llvm.dx.wave.readlaneat.i32([[TY]] %[[#]], i32 %[[#]])
+  // CHECK-SPIRV:  %[[RET:.*]] = call [[TY:.*]] @llvm.spv.wave.readlane.i32([[TY]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok]]) ]
+  // CHECK-DXIL:  %[[RET:.*]] = call [[TY:.*]] @llvm.dx.wave.readlane.i32([[TY]] %[[#]], i32 %[[#]])
   // CHECK:  ret [[TY]] %[[RET]]
   return WaveReadLaneAt(expr, idx);
 }
 
-// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.readlaneat.i32([[TY]], i32) #[[#attr:]]
-// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.readlaneat.i32([[TY]], i32) #[[#attr:]]
+// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.readlane.i32([[TY]], i32) #[[#attr:]]
+// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.readlane.i32([[TY]], i32) #[[#attr:]]
 
 // Test basic lowering to runtime function call with array and float value.
 
 // CHECK-LABEL: test_floatv4
 float4 test_floatv4(float4 expr, uint idx) {
   // CHECK-SPIRV: %[[#entry_tok1:]] = call token @llvm.experimental.convergence.entry()
-  // CHECK-SPIRV:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.spv.wave.readlaneat.v4f32([[TY1]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok1]]) ]
-  // CHECK-DXIL:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.dx.wave.readlaneat.v4f32([[TY1]] %[[#]], i32 %[[#]])
+  // CHECK-SPIRV:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.spv.wave.readlane.v4f32([[TY1]] %[[#]], i32 %[[#]]) [ "convergencectrl"(token %[[#entry_tok1]]) ]
+  // CHECK-DXIL:  %[[RET1:.*]] = call [[TY1:.*]] @llvm.dx.wave.readlane.v4f32([[TY1]] %[[#]], i32 %[[#]])
   // CHECK:  ret [[TY1]] %[[RET1]]
   return WaveReadLaneAt(expr, idx);
 }
 
-// CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.readlaneat.v4f32([[TY1]], i32) #[[#attr]]
-// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.readlaneat.v4f32([[TY1]], i32) #[[#attr]]
+// CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.readlane.v4f32([[TY1]], i32) #[[#attr]]
+// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.readlane.v4f32([[TY1]], i32) #[[#attr]]
 
 // CHECK: attributes #[[#attr]] = {{{.*}} convergent {{.*}}}
diff --git a/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
index e2de250c1bfa11..ef8299b59ca73f 100644
--- a/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/WaveReadLaneAt-errors.hlsl
@@ -15,12 +15,24 @@ float2 test_too_many_arg(float2 p0) {
   // expected-error at -1 {{too many arguments to function call, expected 2, have 3}}
 }
 
-float3 test_index_type_check(float3 p0, double idx) {
+float3 test_index_double_type_check(float3 p0, double idx) {
   return __builtin_hlsl_wave_read_lane_at(p0, idx);
   // expected-error at -1 {{passing 'double' to parameter of incompatible type 'unsigned int'}}
 }
 
-float3 test_index_type_check(float3 p0, int3 idxs) {
+float3 test_index_int3_type_check(float3 p0, int3 idxs) {
   return __builtin_hlsl_wave_read_lane_at(p0, idxs);
   // expected-error at -1 {{passing 'int3' (aka 'vector<int, 3>') to parameter of incompatible type 'unsigned int'}}
 }
+
+struct S { float f; };
+
+float3 test_index_S_type_check(float3 p0, S idx) {
+  return __builtin_hlsl_wave_read_lane_at(p0, idx);
+  // expected-error at -1 {{passing 'S' to parameter of incompatible type 'unsigned int'}}
+}
+
+S test_expr_struct_type_check(S p0, int idx) {
+  return __builtin_hlsl_wave_read_lane_at(p0, idx);
+  // expected-error at -1 {{invalid operand of type 'S' where a scalar or vector is required}}
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index 760359a664789f..b00922c084bc7a 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -85,7 +85,7 @@ def int_dx_normalize : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_
 def int_dx_rsqrt  : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
 def int_dx_wave_getlaneindex : DefaultAttrsIntrinsic<[llvm_i32_ty], [], [IntrConvergent, IntrNoMem]>;
 def int_dx_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
-def int_dx_wave_readlaneat : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
+def int_dx_wave_readlane : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
 def int_dx_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
 def int_dx_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
 def int_dx_radians : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 0b3c35574428f8..80366b8b9bb6e9 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -83,7 +83,7 @@ let TargetPrefix = "spv" in {
     [llvm_anyint_ty, LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
     [IntrNoMem, Commutative] >;
   def int_spv_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
-  def int_spv_wave_readlaneat : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
+  def int_spv_wave_readlane : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
   def int_spv_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
   def int_spv_radians : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty], [IntrNoMem]>;
 
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index c79e629a9fc1ac..7b59943c2cdc65 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -804,7 +804,7 @@ def WaveIsFirstLane :  DXILOp<110, waveIsFirstLane> {
 
 def WaveReadLaneAt:  DXILOp<117, waveReadLaneAt> {
   let Doc = "returns the value from the specified lane";
-  let LLVMIntrinsic = int_dx_wave_readlaneat;
+  let LLVMIntrinsic = int_dx_wave_readlane;
   let arguments = [OverloadTy, Int32Ty];
   let result = OverloadTy;
   let overloads = [Overloads<DXIL1_0, [HalfTy, FloatTy, DoubleTy, Int1Ty, Int16Ty, Int32Ty]>];
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 3905503d6d4c95..df251805d550f7 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -2565,7 +2565,7 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg,
         .addUse(GR.getSPIRVTypeID(ResType))
         .addUse(GR.getOrCreateConstInt(3, I, IntTy, TII));
   }
-  case Intrinsic::spv_wave_readlaneat:
+  case Intrinsic::spv_wave_readlane:
     return selectWaveReadLaneAt(ResVReg, ResType, I);
   case Intrinsic::spv_step:
     return selectExtInst(ResVReg, ResType, I, CL::step, GL::Step);
diff --git a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
index c390804ffdeae4..b05294678dad81 100644
--- a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
@@ -5,49 +5,49 @@
 define noundef half @wave_rla_half(half noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call half @dx.op.waveReadLaneAt.f16(i32 117, half %expr, i32 %idx)
-  %ret = call half @llvm.dx.wave.readlaneat.f16(half %expr, i32 %idx)
+  %ret = call half @llvm.dx.wave.readlane.f16(half %expr, i32 %idx)
   ret half %ret
 }
 
 define noundef float @wave_rla_float(float noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call float @dx.op.waveReadLaneAt.f32(i32 117, float %expr, i32 %idx)
-  %ret = call float @llvm.dx.wave.readlaneat(float %expr, i32 %idx)
+  %ret = call float @llvm.dx.wave.readlane(float %expr, i32 %idx)
   ret float %ret
 }
 
 define noundef double @wave_rla_double(double noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call double @dx.op.waveReadLaneAt.f64(i32 117, double %expr, i32 %idx)
-  %ret = call double @llvm.dx.wave.readlaneat(double %expr, i32 %idx)
+  %ret = call double @llvm.dx.wave.readlane(double %expr, i32 %idx)
   ret double %ret
 }
 
 define noundef i1 @wave_rla_i1(i1 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i1 @dx.op.waveReadLaneAt.i1(i32 117, i1 %expr, i32 %idx)
-  %ret = call i1 @llvm.dx.wave.readlaneat.i1(i1 %expr, i32 %idx)
+  %ret = call i1 @llvm.dx.wave.readlane.i1(i1 %expr, i32 %idx)
   ret i1 %ret
 }
 
 define noundef i16 @wave_rla_i16(i16 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i16 @dx.op.waveReadLaneAt.i16(i32 117, i16 %expr, i32 %idx)
-  %ret = call i16 @llvm.dx.wave.readlaneat.i16(i16 %expr, i32 %idx)
+  %ret = call i16 @llvm.dx.wave.readlane.i16(i16 %expr, i32 %idx)
   ret i16 %ret
 }
 
 define noundef i32 @wave_rla_i32(i32 noundef %expr, i32 noundef %idx) {
 entry:
 ; CHECK: call i32 @dx.op.waveReadLaneAt.i32(i32 117, i32 %expr, i32 %idx)
-  %ret = call i32 @llvm.dx.wave.readlaneat.i32(i32 %expr, i32 %idx)
+  %ret = call i32 @llvm.dx.wave.readlane.i32(i32 %expr, i32 %idx)
   ret i32 %ret
 }
 
-declare half @llvm.dx.wave.readlaneat.f16(half, i32)
-declare float @llvm.dx.wave.readlaneat.f32(float, i32)
-declare double @llvm.dx.wave.readlaneat.f64(double, i32)
+declare half @llvm.dx.wave.readlane.f16(half, i32)
+declare float @llvm.dx.wave.readlane.f32(float, i32)
+declare double @llvm.dx.wave.readlane.f64(double, i32)
 
-declare i1 @llvm.dx.wave.readlaneat.i1(i1, i32)
-declare i16 @llvm.dx.wave.readlaneat.i16(i16, i32)
-declare i32 @llvm.dx.wave.readlaneat.i32(i32, i32)
+declare i1 @llvm.dx.wave.readlane.i1(i1, i32)
+declare i16 @llvm.dx.wave.readlane.i16(i16, i32)
+declare i32 @llvm.dx.wave.readlane.i32(i32, i32)
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
index 2810eca8907c20..44d9024e0a44bd 100644
--- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
@@ -14,7 +14,7 @@
 define float @test_1(float %fexpr, i32 %idx) {
 entry:
 ; CHECK:   %[[#fret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#fexpr]] %[[#idx1]] %[[#scope]]
-  %0 = call float @llvm.spv.wave.readlaneat.f32(float %fexpr, i32 %idx)
+  %0 = call float @llvm.spv.wave.readlane.f32(float %fexpr, i32 %idx)
   ret float %0
 }
 
@@ -23,7 +23,7 @@ entry:
 define i32 @test_2(i32 %iexpr, i32 %idx) {
 entry:
 ; CHECK:   %[[#iret:]] = OpGroupNonUniformShuffle %[[#uint]] %[[#iexpr]] %[[#idx2]] %[[#scope]]
-  %0 = call i32 @llvm.spv.wave.readlaneat.i32(i32 %iexpr, i32 %idx)
+  %0 = call i32 @llvm.spv.wave.readlane.i32(i32 %iexpr, i32 %idx)
   ret i32 %0
 }
 
@@ -32,10 +32,10 @@ entry:
 define <4 x i1> @test_3(<4 x i1> %vbexpr, i32 %idx) {
 entry:
 ; CHECK:   %[[#vbret:]] = OpGroupNonUniformShuffle %[[#v4_bool]] %[[#vbexpr]] %[[#idx3]] %[[#scope]]
-  %0 = call <4 x i1> @llvm.spv.wave.readlaneat.v4i1(<4 x i1> %vbexpr, i32 %idx)
+  %0 = call <4 x i1> @llvm.spv.wave.readlane.v4i1(<4 x i1> %vbexpr, i32 %idx)
   ret <4 x i1> %0
 }
 
-declare float @llvm.spv.wave.readlaneat.f32(float, i32)
-declare i32 @llvm.spv.wave.readlaneat.i32(i32, i32)
-declare <4 x i1> @llvm.spv.wave.readlaneat.v4i1(<4 x i1>, i32)
+declare float @llvm.spv.wave.readlane.f32(float, i32)
+declare i32 @llvm.spv.wave.readlane.i32(i32, i32)
+declare <4 x i1> @llvm.spv.wave.readlane.v4i1(<4 x i1>, i32)

>From 8a755587ad9e60a03ddb3162d57a03968602db6f Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Thu, 10 Oct 2024 13:05:50 -0700
Subject: [PATCH 13/14] clang formatting

---
 llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index df251805d550f7..b90961ee7cb8fa 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -229,7 +229,7 @@ class SPIRVInstructionSelector : public InstructionSelector {
 
   bool selectSpvThreadId(Register ResVReg, const SPIRVType *ResType,
                          MachineInstr &I) const;
-  
+
   bool selectWaveReadLaneAt(Register ResVReg, const SPIRVType *ResType,
                             MachineInstr &I) const;
 

>From 78f7e5dfb7ee4e5650f1d1ecd0a6182a8f8dacb7 Mon Sep 17 00:00:00 2001
From: Finn Plummer <canadienfinn at gmail.com>
Date: Thu, 10 Oct 2024 16:33:38 -0700
Subject: [PATCH 14/14] self review:

    - the execution scope must be the second operand and not the final
      operand of the SPIRV instruction. this was missed as I did not
      realize that SPIRV-TOOLS was not enabled on my local machine but
      luckily got caught by the spir-v github tests

    - missing testcase for int64 in directx lowering and caught that
      Int64Ty was missing from the overload types of dxilop
---
 llvm/lib/Target/DirectX/DXIL.td                 |  4 ++--
 .../Target/SPIRV/SPIRVInstructionSelector.cpp   |  4 ++--
 llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll     |  8 ++++++++
 .../SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll     | 17 ++++++++++-------
 4 files changed, 22 insertions(+), 11 deletions(-)

diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index 7b59943c2cdc65..147b32b1ca9903 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -807,8 +807,8 @@ def WaveReadLaneAt:  DXILOp<117, waveReadLaneAt> {
   let LLVMIntrinsic = int_dx_wave_readlane;
   let arguments = [OverloadTy, Int32Ty];
   let result = OverloadTy;
-  let overloads = [Overloads<DXIL1_0, [HalfTy, FloatTy, DoubleTy, Int1Ty, Int16Ty, Int32Ty]>];
-    let stages = [Stages<DXIL1_0, [all_stages]>];
+  let overloads = [Overloads<DXIL1_0, [HalfTy, FloatTy, DoubleTy, Int1Ty, Int16Ty, Int32Ty, Int64Ty]>];
+  let stages = [Stages<DXIL1_0, [all_stages]>];
   let attributes = [Attributes<DXIL1_0, [ReadNone]>];
 }
 
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index b90961ee7cb8fa..8b040c4815b0d8 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -1777,9 +1777,9 @@ bool SPIRVInstructionSelector::selectWaveReadLaneAt(Register ResVReg,
                  TII.get(SPIRV::OpGroupNonUniformShuffle))
       .addDef(ResVReg)
       .addUse(GR.getSPIRVTypeID(ResType))
+      .addUse(GR.getOrCreateConstInt(3, I, IntTy, TII))
       .addUse(I.getOperand(2).getReg())
-      .addUse(I.getOperand(3).getReg())
-      .addUse(GR.getOrCreateConstInt(3, I, IntTy, TII));
+      .addUse(I.getOperand(3).getReg());
 }
 
 bool SPIRVInstructionSelector::selectBitreverse(Register ResVReg,
diff --git a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
index b05294678dad81..0024ba66c0cad8 100644
--- a/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/DirectX/WaveReadLaneAt.ll
@@ -44,6 +44,13 @@ entry:
   ret i32 %ret
 }
 
+define noundef i64 @wave_rla_i64(i64 noundef %expr, i32 noundef %idx) {
+entry:
+; CHECK: call i64 @dx.op.waveReadLaneAt.i64(i32 117, i64 %expr, i32 %idx)
+  %ret = call i64 @llvm.dx.wave.readlane.i64(i64 %expr, i32 %idx)
+  ret i64 %ret
+}
+
 declare half @llvm.dx.wave.readlane.f16(half, i32)
 declare float @llvm.dx.wave.readlane.f32(float, i32)
 declare double @llvm.dx.wave.readlane.f64(double, i32)
@@ -51,3 +58,4 @@ declare double @llvm.dx.wave.readlane.f64(double, i32)
 declare i1 @llvm.dx.wave.readlane.i1(i1, i32)
 declare i16 @llvm.dx.wave.readlane.i16(i16, i32)
 declare i32 @llvm.dx.wave.readlane.i32(i32, i32)
+declare i64 @llvm.dx.wave.readlane.i64(i64, i32)
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
index 44d9024e0a44bd..ea0616eaf39dcc 100644
--- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WaveReadLaneAt.ll
@@ -1,5 +1,5 @@
 ; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv32v1.3-vulkan-unknown %s -o - | FileCheck %s
-; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-vulkan-unknown %s -o - -filetype=obj | spirv-val %}
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv32v1.3-vulkan-unknown %s -o - -filetype=obj | spirv-val %}
 
 ; Test lowering to spir-v backend for various types and scalar/vector
 
@@ -9,29 +9,32 @@
 ; CHECK-DAG:   %[[#v4_bool:]] = OpTypeVector %[[#bool]] 4
 ; CHECK-DAG:   %[[#scope:]] = OpConstant %[[#uint]] 3
 
+; CHECK-LABEL: Begin function test_float
 ; CHECK:   %[[#fexpr:]] = OpFunctionParameter %[[#f32]]
 ; CHECK:   %[[#idx1:]] = OpFunctionParameter %[[#uint]]
-define float @test_1(float %fexpr, i32 %idx) {
+define float @test_float(float %fexpr, i32 %idx) {
 entry:
-; CHECK:   %[[#fret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#fexpr]] %[[#idx1]] %[[#scope]]
+; CHECK:   %[[#fret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#scope]] %[[#fexpr]] %[[#idx1]]
   %0 = call float @llvm.spv.wave.readlane.f32(float %fexpr, i32 %idx)
   ret float %0
 }
 
+; CHECK-LABEL: Begin function test_int
 ; CHECK:   %[[#iexpr:]] = OpFunctionParameter %[[#uint]]
 ; CHECK:   %[[#idx2:]] = OpFunctionParameter %[[#uint]]
-define i32 @test_2(i32 %iexpr, i32 %idx) {
+define i32 @test_int(i32 %iexpr, i32 %idx) {
 entry:
-; CHECK:   %[[#iret:]] = OpGroupNonUniformShuffle %[[#uint]] %[[#iexpr]] %[[#idx2]] %[[#scope]]
+; CHECK:   %[[#iret:]] = OpGroupNonUniformShuffle %[[#uint]] %[[#scope]] %[[#iexpr]] %[[#idx2]]
   %0 = call i32 @llvm.spv.wave.readlane.i32(i32 %iexpr, i32 %idx)
   ret i32 %0
 }
 
+; CHECK-LABEL: Begin function test_vbool
 ; CHECK:   %[[#vbexpr:]] = OpFunctionParameter %[[#v4_bool]]
 ; CHECK:   %[[#idx3:]] = OpFunctionParameter %[[#uint]]
-define <4 x i1> @test_3(<4 x i1> %vbexpr, i32 %idx) {
+define <4 x i1> @test_vbool(<4 x i1> %vbexpr, i32 %idx) {
 entry:
-; CHECK:   %[[#vbret:]] = OpGroupNonUniformShuffle %[[#v4_bool]] %[[#vbexpr]] %[[#idx3]] %[[#scope]]
+; CHECK:   %[[#vbret:]] = OpGroupNonUniformShuffle %[[#v4_bool]] %[[#scope]] %[[#vbexpr]] %[[#idx3]]
   %0 = call <4 x i1> @llvm.spv.wave.readlane.v4i1(<4 x i1> %vbexpr, i32 %idx)
   ret <4 x i1> %0
 }



More information about the cfe-commits mailing list