[Backport to 16] Implement SPV_INTEL_ternary_bitwise_function (KhronosGroup#3104)

This commit implements support for bi-directional translation of the
`BitwiseFunctionINTEL` operation added in

https://github.khronos.org/SPIRV-Registry/extensions/INTEL/SPV_INTEL_ternary_bitwise_function.html
together with the corresponding capability.

Signed-off-by: Larsen, Steffen <[email protected]>

Author: steffenlarsen

include/LLVMSPIRVExtensions.inc

@@ -75,3 +75,4 @@ EXT(SPV_INTEL_bindless_images)
 EXT(SPV_INTEL_2d_block_io)
 EXT(SPV_INTEL_subgroup_matrix_multiply_accumulate)
 EXT(SPV_KHR_bfloat16)
+EXT(SPV_INTEL_ternary_bitwise_function)

lib/SPIRV/libSPIRV/SPIRVInstruction.h

@@ -4040,5 +4040,60 @@ class SPIRVSubgroupMatrixMultiplyAccumulateINTELInst
 _SPIRV_OP(SubgroupMatrixMultiplyAccumulate, true, 7, true, 4)
 #undef _SPIRV_OP
 
+class SPIRVTernaryBitwiseFunctionINTELInst : public SPIRVInstTemplateBase {
+public:
+  void validate() const override {
+    SPIRVInstruction::validate();
+    SPIRVErrorLog &SPVErrLog = this->getModule()->getErrorLog();
+    std::string InstName = "BitwiseFunctionINTEL";
+
+    const SPIRVType *ResTy = this->getType();
+    SPVErrLog.checkError(
+        ResTy->isTypeInt() || (ResTy->isTypeVector() &&
+                               ResTy->getVectorComponentType()->isTypeInt()),
+        SPIRVEC_InvalidInstruction,
+        InstName + "\nResult type must be an integer scalar or vector.\n");
+
+    auto CommonArgCheck = [this, ResTy, &InstName,
+                           &SPVErrLog](size_t ArgI, const char *ArgPlacement) {
+      SPIRVValue *Arg =
+          const_cast<SPIRVTernaryBitwiseFunctionINTELInst *>(this)->getOperand(
+              ArgI);
+      SPVErrLog.checkError(
+          Arg->getType() == ResTy, SPIRVEC_InvalidInstruction,
+          InstName + "\n" + ArgPlacement +
+              " argument must be the same as the result type.\n");
+    };
+
+    CommonArgCheck(0, "First");
+    CommonArgCheck(1, "Second");
+    CommonArgCheck(2, "Third");
+
+    SPIRVValue *LUTIndexArg =
+        const_cast<SPIRVTernaryBitwiseFunctionINTELInst *>(this)->getOperand(3);
+    const SPIRVType *LUTIndexArgTy = LUTIndexArg->getType();
+    SPVErrLog.checkError(
+        LUTIndexArgTy->isTypeInt(32), SPIRVEC_InvalidInstruction,
+        InstName + "\nFourth argument must be a 32-bit integer scalar.\n");
+    SPVErrLog.checkError(
+        isConstantOpCode(LUTIndexArg->getOpCode()), SPIRVEC_InvalidInstruction,
+        InstName + "\nFourth argument must be constant instruction.\n");
+  }
+
+  std::optional<ExtensionID> getRequiredExtension() const override {
+    return ExtensionID::SPV_INTEL_ternary_bitwise_function;
+  }
+  SPIRVCapVec getRequiredCapability() const override {
+    return getVec(CapabilityTernaryBitwiseFunctionINTEL);
+  }
+};
+
+#define _SPIRV_OP(x, ...)                                                      \
+  typedef SPIRVInstTemplate<SPIRVTernaryBitwiseFunctionINTELInst,              \
+                            Op##x##INTEL, __VA_ARGS__>                         \
+      SPIRV##x##INTEL;
+_SPIRV_OP(BitwiseFunction, true, 7)
+#undef _SPIRV_OP
+
 } // namespace SPIRV
 #endif // SPIRV_LIBSPIRV_SPIRVINSTRUCTION_H

lib/SPIRV/libSPIRV/SPIRVNameMapEnum.h

@@ -639,6 +639,7 @@ template <> inline void SPIRVMap<Capability, std::string>::init() {
   add(CapabilitySubgroup2DBlockTransposeINTEL, "Subgroup2DBlockTransposeINTEL");
   add(CapabilitySubgroupMatrixMultiplyAccumulateINTEL,
       "SubgroupMatrixMultiplyAccumulateINTEL");
+  add(CapabilityTernaryBitwiseFunctionINTEL, "TernaryBitwiseFunctionINTEL");
 
   // From spirv_internal.hpp
   add(internal::CapabilityFastCompositeINTEL, "FastCompositeINTEL");

lib/SPIRV/libSPIRV/SPIRVOpCodeEnum.h

@@ -570,6 +570,7 @@ _SPIRV_OP(Subgroup2DBlockLoadTransposeINTEL, 6233)
 _SPIRV_OP(Subgroup2DBlockPrefetchINTEL, 6234)
 _SPIRV_OP(Subgroup2DBlockStoreINTEL, 6235)
 _SPIRV_OP(SubgroupMatrixMultiplyAccumulateINTEL, 6237)
+_SPIRV_OP(BitwiseFunctionINTEL, 6242)
 _SPIRV_OP(GroupIMulKHR, 6401)
 _SPIRV_OP(GroupFMulKHR, 6402)
 _SPIRV_OP(GroupBitwiseAndKHR, 6403)

test/extensions/INTEL/SPV_INTEL_ternary_bitwise_function/bitwise_function.ll

@@ -0,0 +1,78 @@
+; RUN: llvm-as %s -o %t.bc
+; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_INTEL_ternary_bitwise_function -o %t.spv
+; RUN: llvm-spirv %t.spv --to-text -o %t.spt
+; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
+
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
+
+; RUN: not llvm-spirv %t.bc 2>&1 | FileCheck %s --check-prefix=CHECK-ERROR
+; CHECK-ERROR: RequiresExtension: Feature requires the following SPIR-V extension:
+; CHECK-ERROR-NEXT: SPV_INTEL_ternary_bitwise_function
+
+; CHECK-SPIRV-NOT: Name [[#]] "_Z28__spirv_BitwiseFunctionINTELiiij"
+; CHECK-SPIRV-NOT: Name [[#]] "_Z28__spirv_BitwiseFunctionINTELDv4_iS_S_j"
+
+; CHECK-SPIRV-DAG: Capability TernaryBitwiseFunctionINTEL
+; CHECK-SPIRV-DAG: Extension "SPV_INTEL_ternary_bitwise_function"
+
+; CHECK-SPIRV-DAG: TypeInt [[#TYPEINT:]] 32 0
+; CHECK-SPIRV-DAG: TypeVector [[#TYPEINTVEC4:]] [[#TYPEINT]] 4
+; CHECK-SPIRV-DAG: Constant [[#TYPEINT]] [[#ScalarLUT:]] 24
+; CHECK-SPIRV-DAG: Constant [[#TYPEINT]] [[#VecLUT:]] 42
+
+; CHECK-SPIRV: Load [[#TYPEINT]] [[#ScalarA:]]
+; CHECK-SPIRV: Load [[#TYPEINT]] [[#ScalarB:]]
+; CHECK-SPIRV: Load [[#TYPEINT]] [[#ScalarC:]]
+; CHECK-SPIRV: BitwiseFunctionINTEL [[#TYPEINT]] {{.*}} [[#ScalarA]] [[#ScalarB]] [[#ScalarC]] [[#ScalarLUT]]
+; CHECK-SPIRV: Load [[#TYPEINTVEC4]] [[#VecA:]]
+; CHECK-SPIRV: Load [[#TYPEINTVEC4]] [[#VecB:]]
+; CHECK-SPIRV: Load [[#TYPEINTVEC4]] [[#VecC:]]
+; CHECK-SPIRV: BitwiseFunctionINTEL [[#TYPEINTVEC4]] {{.*}} [[#VecA]] [[#VecB]] [[#VecC]] [[#VecLUT]]
+
+; CHECK-LLVM: %[[ScalarA:.*]] = load i32, ptr
+; CHECK-LLVM: %[[ScalarB:.*]] = load i32, ptr
+; CHECK-LLVM: %[[ScalarC:.*]] = load i32, ptr
+; CHECK-LLVM: call spir_func i32 @_Z28__spirv_BitwiseFunctionINTELiiii(i32 %[[ScalarA]], i32 %[[ScalarB]], i32 %[[ScalarC]], i32 24)
+; CHECK-LLVM: %[[VecA:.*]] = load <4 x i32>, ptr
+; CHECK-LLVM: %[[VecB:.*]] = load <4 x i32>, ptr
+; CHECK-LLVM: %[[VecC:.*]] = load <4 x i32>, ptr
+; CHECK-LLVM: call spir_func <4 x i32> @_Z28__spirv_BitwiseFunctionINTELDv4_iS_S_i(<4 x i32> %[[VecA]], <4 x i32> %[[VecB]], <4 x i32> %[[VecC]], i32 42)
+
+target datalayout = "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir"
+
+; Function Attrs: nounwind readnone
+define spir_kernel void @fooScalar() {
+entry:
+  %argA = alloca i32
+  %argB = alloca i32
+  %argC = alloca i32
+  %A = load i32, ptr %argA
+  %B = load i32, ptr %argB
+  %C = load i32, ptr %argC
+  %res = call spir_func i32 @_Z28__spirv_BitwiseFunctionINTELiiii(i32 %A, i32 %B, i32 %C, i32 24)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+define spir_kernel void @fooVec() {
+entry:
+  %argA = alloca <4 x i32>
+  %argB = alloca <4 x i32>
+  %argC = alloca <4 x i32>
+  %A = load <4 x i32>, ptr %argA
+  %B = load <4 x i32>, ptr %argB
+  %C = load <4 x i32>, ptr %argC
+  %res = call spir_func <4 x i32> @_Z28__spirv_BitwiseFunctionINTELDv4_iS_S_i(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, i32 42)
+  ret void
+}
+
+declare dso_local spir_func i32 @_Z28__spirv_BitwiseFunctionINTELiiii(i32, i32, i32, i32)
+declare dso_local spir_func <4 x i32> @_Z28__spirv_BitwiseFunctionINTELDv4_iS_S_i(<4 x i32>, <4 x i32>, <4 x i32>, i32)
+
+!llvm.module.flags = !{!0}
+!opencl.spir.version = !{!1}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 1, i32 2}