llvm · aadeshps-mcw · Apr 1, 2025
diff --git a/llvm/lib/Target/SPIRV/SPIRVBuiltins.cpp b/llvm/lib/Target/SPIRV/SPIRVBuiltins.cpp
@@ -697,7 +697,8 @@ static bool buildAtomicStoreInst(const SPIRV::IncomingCall *Call,
                                  MachineIRBuilder &MIRBuilder,
                                  SPIRVGlobalRegistry *GR) {
   if (Call->isSpirvOp())
-    return buildOpFromWrapper(MIRBuilder, SPIRV::OpAtomicStore, Call, Register(0));
+    return buildOpFromWrapper(MIRBuilder, SPIRV::OpAtomicStore, Call,
+                              Register(0));
 
   Register ScopeRegister =
       buildConstantIntReg32(SPIRV::Scope::Device, MIRBuilder, GR);
@@ -2252,6 +2253,17 @@ static bool generateExtendedBitOpsInst(const SPIRV::IncomingCall *Call,
 
   return buildExtendedBitOpsInst(Call, Opcode, MIRBuilder, GR);
 }
+static bool generateSubgroup2DBlockInst(const SPIRV::IncomingCall *Call,
+                                        MachineIRBuilder &MIRBuilder,
+                                        SPIRVGlobalRegistry *GR) {
+  const SPIRV::DemangledBuiltin *Builtin = Call->Builtin;
+  unsigned Opcode =
+      SPIRV::lookupNativeBuiltin(Builtin->Name, Builtin->Set)->Opcode;
+  auto MIB = MIRBuilder.buildInstr(Opcode);
+  for (unsigned i = 0; i < Call->Arguments.size(); i++)
+    MIB.addUse(Call->Arguments[i]);
+  return true;
+}
 
 static bool generateBindlessImageINTELInst(const SPIRV::IncomingCall *Call,
                                            MachineIRBuilder &MIRBuilder,
@@ -2845,6 +2857,8 @@ std::optional<bool> lowerBuiltin(const StringRef DemangledCall,
     return generateExtendedBitOpsInst(Call.get(), MIRBuilder, GR);
   case SPIRV::BindlessINTEL:
     return generateBindlessImageINTELInst(Call.get(), MIRBuilder, GR);
+  case SPIRV::Subgroup2DBlock:
+    return generateSubgroup2DBlockInst(Call.get(), MIRBuilder, GR);
   }
   return false;
 }

diff --git a/llvm/lib/Target/SPIRV/SPIRVBuiltins.td b/llvm/lib/Target/SPIRV/SPIRVBuiltins.td
@@ -67,6 +67,7 @@ def CoopMatr : BuiltinGroup;
 def ICarryBorrow : BuiltinGroup;
 def ExtendedBitOps : BuiltinGroup;
 def BindlessINTEL : BuiltinGroup;
+def Subgroup2DBlock: BuiltinGroup;
 
 //===----------------------------------------------------------------------===//
 // Class defining a demangled builtin record. The information in the record
@@ -714,6 +715,13 @@ defm : DemangledNativeBuiltin<"__spirv_ConvertHandleToImageINTEL", OpenCL_std, B
 defm : DemangledNativeBuiltin<"__spirv_ConvertHandleToSamplerINTEL", OpenCL_std, BindlessINTEL, 1, 1, OpConvertHandleToSamplerINTEL>;
 defm : DemangledNativeBuiltin<"__spirv_ConvertHandleToSampledImageINTEL", OpenCL_std, BindlessINTEL, 1, 1, OpConvertHandleToSampledImageINTEL>;
 
+//SPV_INTEL_2d_block_io
+defm : DemangledNativeBuiltin<"__spirv_Subgroup2DBlockLoadINTEL", OpenCL_std, Subgroup2DBlock, 10, 10, OpSubgroup2DBlockLoadINTEL>;
+defm : DemangledNativeBuiltin<"__spirv_Subgroup2DBlockLoadTransposeINTEL", OpenCL_std, Subgroup2DBlock, 10, 10, OpSubgroup2DBlockLoadTransposeINTEL>;
+defm : DemangledNativeBuiltin<"__spirv_Subgroup2DBlockLoadTransformINTEL", OpenCL_std, Subgroup2DBlock, 10, 10, OpSubgroup2DBlockLoadTransformINTEL>;
+defm : DemangledNativeBuiltin<"__spirv_Subgroup2DBlockPrefetchINTEL", OpenCL_std, Subgroup2DBlock, 9, 9, OpSubgroup2DBlockPrefetchINTEL>;
+defm : DemangledNativeBuiltin<"__spirv_Subgroup2DBlockStoreINTEL", OpenCL_std, Subgroup2DBlock, 10, 10, OpSubgroup2DBlockStoreINTEL>;
+
 //===----------------------------------------------------------------------===//
 // Class defining a work/sub group builtin that should be translated into a
 // SPIR-V instruction using the defined properties.

diff --git a/llvm/lib/Target/SPIRV/SPIRVCommandLine.cpp b/llvm/lib/Target/SPIRV/SPIRVCommandLine.cpp
@@ -92,7 +92,9 @@ static const std::map<std::string, SPIRV::Extension::Extension, std::less<>>
         {"SPV_INTEL_long_composites",
          SPIRV::Extension::Extension::SPV_INTEL_long_composites},
         {"SPV_INTEL_fp_max_error",
-         SPIRV::Extension::Extension::SPV_INTEL_fp_max_error}};
+         SPIRV::Extension::Extension::SPV_INTEL_fp_max_error},
+        {"SPV_INTEL_2d_block_io",
+         SPIRV::Extension::Extension::SPV_INTEL_2d_block_io}};
 
 bool SPIRVExtensionsParser::parse(cl::Option &O, llvm::StringRef ArgName,
                                   llvm::StringRef ArgValue,

diff --git a/llvm/lib/Target/SPIRV/SPIRVInstrInfo.td b/llvm/lib/Target/SPIRV/SPIRVInstrInfo.td
@@ -928,3 +928,20 @@ def OpAliasScopeDeclINTEL: Op<5912, (outs ID:$res), (ins ID:$AliasDomain, variab
                   "$res = OpAliasScopeDeclINTEL $AliasDomain">;
 def OpAliasScopeListDeclINTEL: Op<5913, (outs ID:$res), (ins variable_ops),
                   "$res = OpAliasScopeListDeclINTEL">;
+
+//SPV_INTEL_2d_block_io
+def OpSubgroup2DBlockLoadINTEL: Op<6231, (outs),
+          (ins ID:$elementSize, ID:$blockWidth, ID:$blockHeight, ID:$blockCount, ID:$srcBasePointer, ID:$memoryWidth, ID:$memoryHeight, ID:$memoryPitch, ID:$coordinate, ID:$dstPointer),
+          "OpSubgroup2DBlockLoadINTEL $elementSize $blockWidth $blockHeight $blockCount $srcBasePointer $memoryWidth $memoryHeight $memoryPitch $coordinate $dstPointer">;
+def OpSubgroup2DBlockLoadTransposeINTEL: Op<6232, (outs),
+          (ins ID:$elementSize, ID:$blockWidth, ID:$blockHeight, ID:$blockCount, ID:$srcBasePointer, ID:$memoryWidth, ID:$memoryHeight, ID:$memoryPitch, ID:$coordinate, ID:$dstPointer),
+          "OpSubgroup2DBlockLoadTransposeINTEL $elementSize $blockWidth $blockHeight $blockCount $srcBasePointer $memoryWidth $memoryHeight $memoryPitch $coordinate $dstPointer">;
+def OpSubgroup2DBlockLoadTransformINTEL: Op<6233, (outs),
+          (ins ID:$elementSize, ID:$blockWidth, ID:$blockHeight, ID:$blockCount, ID:$srcBasePointer, ID:$memoryWidth, ID:$memoryHeight, ID:$memoryPitch, ID:$coordinate, ID:$dstPointer),
+          "OpSubgroup2DBlockLoadTransformINTEL $elementSize $blockWidth $blockHeight $blockCount $srcBasePointer $memoryWidth $memoryHeight $memoryPitch $coordinate $dstPointer">;
+def OpSubgroup2DBlockPrefetchINTEL: Op<6234, (outs),
+          (ins ID:$elementSize, ID:$blockWidth, ID:$blockHeight, ID:$blockCount, ID:$srcPointer, ID:$memoryWidth, ID:$memoryHeight, ID:$memoryPitch, ID:$coordinate),
+          "OpSubgroup2DBlockPrefetchINTEL $elementSize $blockWidth $blockHeight $blockCount $srcPointer $memoryWidth $memoryHeight $memoryPitch $coordinate">;
+def OpSubgroup2DBlockStoreINTEL: Op<6235, (outs),
+          (ins ID:$elementSize, ID:$blockWidth, ID:$blockHeight, ID:$blockCount, ID:$srcPointer, ID:$dstBasePointer, ID:$memoryWidth, ID:$memoryHeight, ID:$memoryPitch, ID:$coordinate),
+          "OpSubgroup2DBlockStoreINTEL $elementSize $blockWidth $blockHeight $blockCount $srcPointer $dstBasePointer $memoryWidth $memoryHeight $memoryPitch $coordinate">;
diff --git a/llvm/lib/Target/SPIRV/SPIRVModuleAnalysis.cpp b/llvm/lib/Target/SPIRV/SPIRVModuleAnalysis.cpp
@@ -1517,6 +1517,39 @@ void addInstrRequirements(const MachineInstr &MI,
       Reqs.addCapability(SPIRV::Capability::FunctionPointersINTEL);
     }
     break;
+
+  case SPIRV::OpSubgroup2DBlockLoadINTEL:
+  case SPIRV::OpSubgroup2DBlockPrefetchINTEL:
+  case SPIRV::OpSubgroup2DBlockStoreINTEL: {
+    if (!ST.canUseExtension(SPIRV::Extension::SPV_INTEL_2d_block_io))
+      report_fatal_error(
+          "OpSubgroup2DBlockLoadTransposeINTEL instruction requires the "
+          "following SPIR-V extension: SPV_INTEL_2d_block_io",
+          false);
+    Reqs.addExtension(SPIRV::Extension::SPV_INTEL_2d_block_io);
+    Reqs.addCapability(SPIRV::Capability::Subgroup2DBlockIOINTEL);
+    break;
+  }
+  case SPIRV::OpSubgroup2DBlockLoadTransformINTEL: {
+    if (!ST.canUseExtension(SPIRV::Extension::SPV_INTEL_2d_block_io))
+      report_fatal_error(
+          "OpSubgroup2DBlockLoadTransformINTEL instruction requires the "
+          "following SPIR-V extension: SPV_INTEL_2d_block_io",
+          false);
+    Reqs.addExtension(SPIRV::Extension::SPV_INTEL_2d_block_io);
+    Reqs.addCapability(SPIRV::Capability::Subgroup2DBlockTransformINTEL);
+    break;
+  }
+  case SPIRV::OpSubgroup2DBlockLoadTransposeINTEL: {
+    if (!ST.canUseExtension(SPIRV::Extension::SPV_INTEL_2d_block_io))
+      report_fatal_error(
+          "OpSubgroup2DBlockLoadTransposeINTEL instruction requires the "
+          "following SPIR-V extension: SPV_INTEL_2d_block_io",
+          false);
+    Reqs.addExtension(SPIRV::Extension::SPV_INTEL_2d_block_io);
+    Reqs.addCapability(SPIRV::Capability::Subgroup2DBlockTransposeINTEL);
+    break;
+  }
   case SPIRV::OpAtomicFAddEXT:
   case SPIRV::OpAtomicFMinEXT:
   case SPIRV::OpAtomicFMaxEXT:

diff --git a/llvm/lib/Target/SPIRV/SPIRVSymbolicOperands.td b/llvm/lib/Target/SPIRV/SPIRVSymbolicOperands.td
@@ -313,6 +313,7 @@ defm SPV_INTEL_bindless_images : ExtensionOperand<116>;
 defm SPV_INTEL_long_composites : ExtensionOperand<117>;
 defm SPV_INTEL_memory_access_aliasing : ExtensionOperand<118>;
 defm SPV_INTEL_fp_max_error : ExtensionOperand<119>;
+defm SPV_INTEL_2d_block_io : ExtensionOperand<120>;
 
 //===----------------------------------------------------------------------===//
 // Multiclass used to define Capabilities enum values and at the same time
@@ -513,6 +514,9 @@ defm LongCompositesINTEL : CapabilityOperand<6089, 0, 0, [SPV_INTEL_long_composi
 defm BindlessImagesINTEL : CapabilityOperand<6528, 0, 0, [SPV_INTEL_bindless_images], []>;
 defm MemoryAccessAliasingINTEL : CapabilityOperand<5910, 0, 0, [SPV_INTEL_memory_access_aliasing], []>;
 defm FPMaxErrorINTEL : CapabilityOperand<6169, 0, 0, [SPV_INTEL_fp_max_error], []>;
+defm Subgroup2DBlockIOINTEL : CapabilityOperand<6228, 0, 0, [SPV_INTEL_2d_block_io], []>;
+defm Subgroup2DBlockTransformINTEL : CapabilityOperand<6229, 0, 0, [SPV_INTEL_2d_block_io], []>;
+defm Subgroup2DBlockTransposeINTEL : CapabilityOperand<6230, 0, 0, [SPV_INTEL_2d_block_io], []>;
 
 //===----------------------------------------------------------------------===//
 // Multiclass used to define SourceLanguage enum values and at the same time

diff --git a/llvm/test/CodeGen/SPIRV/extensions/SPV_INTEL_2d_block_io/2d_block_io_generic.ll b/llvm/test/CodeGen/SPIRV/extensions/SPV_INTEL_2d_block_io/2d_block_io_generic.ll
@@ -0,0 +1,74 @@
+; Generated with:
+; source.cl:
+; void __spirv_Subgroup2DBlockLoadINTEL(         int element_size, int block_width, int block_height, int block_count, const __global void* src_base_pointer, int memory_width,                int memory_height, int memory_pitch,  int2 coordinate,  private void* dst_pointer);
+; void __spirv_Subgroup2DBlockLoadTransposeINTEL(int element_size, int block_width, int block_height, int block_count, const __global void* src_base_pointer, int memory_width,                int memory_height, int memory_pitch,  int2 coordinate,  private void* dst_pointer);
+; void __spirv_Subgroup2DBlockLoadTransformINTEL(int element_size, int block_width, int block_height, int block_count, const __global void* src_base_pointer, int memory_width,                int memory_height, int memory_pitch,  int2 coordinate,  private void* dst_pointer);
+; void __spirv_Subgroup2DBlockPrefetchINTEL(     int element_size, int block_width, int block_height, int block_count, const __global void* src_base_pointer, int memory_width,                int memory_height, int memory_pitch,  int2 coordinate                            );
+; void __spirv_Subgroup2DBlockStoreINTEL(        int element_size, int block_width, int block_height, int block_count, const  private void* src_pointer,      __global void* dst_base_pointer, int memory_width,  int memory_height, int memory_pitch, int2 coordinate          );
+;
+; void foo(const __global void* base_address, __global void* dst_base_pointer, int width, int height, int pitch, int2 coord, private void* dst_pointer, const private void* src_pointer) {
+;     const int i = 42;
+;     __spirv_Subgroup2DBlockLoadINTEL(i, i, i, i, base_address, width, height, pitch, coord, dst_pointer);
+;     __spirv_Subgroup2DBlockLoadTransformINTEL(i, i, i, i, base_address, width, height, pitch, coord, dst_pointer);
+;     __spirv_Subgroup2DBlockLoadTransposeINTEL(i, i, i, i, base_address, width, height, pitch, coord, dst_pointer);
+;     __spirv_Subgroup2DBlockPrefetchINTEL(i, i, i, i, base_address, width, height, pitch, coord);
+;     __spirv_Subgroup2DBlockStoreINTEL(i, i, i, i, src_pointer, dst_base_pointer, width, height, pitch, coord);
+;   }
+; clang -cc1 -cl-std=clc++2021 -triple spir64-unknown-unknown -emit-llvm -finclude-default-header source.cl -o tmp.ll
+
+
+
+; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv64-unknown-unknown --spirv-ext=+SPV_INTEL_2d_block_io %s -o %t.spt
+; RUN: FileCheck %s --input-file=%t.spt
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv64-unknown-unknown --spirv-ext=+SPV_INTEL_2d_block_io %s -o - -filetype=obj | spirv-val %}
+
+; CHECK: OpCapability Subgroup2DBlockIOINTEL
+; CHECK: OpCapability Subgroup2DBlockTransformINTEL
+; CHECK: OpCapability Subgroup2DBlockTransposeINTEL
+; CHECK: OpExtension "SPV_INTEL_2d_block_io"
+; CHECK: %[[#Int8Ty:]] = OpTypeInt 8 0
+; CHECK: %[[#GlbPtrTy:]] = OpTypePointer CrossWorkgroup %[[#Int8Ty]]
+; CHECK: %[[#Int32Ty:]] = OpTypeInt 32 0
+; CHECK: %[[#VectorTy:]] = OpTypeVector %[[#Int32Ty]] 2
+; CHECK: %[[#PrvPtrTy:]] = OpTypePointer Function %[[#Int8Ty]]
+; CHECK: %[[#VoidTy:]] = OpTypeVoid
+; CHECK: %[[#Const42:]] = OpConstant %[[#Int32Ty]] 42
+; CHECK: %[[#BaseSrc:]] = OpFunctionParameter %[[#GlbPtrTy]]
+; CHECK: %[[#BaseDst:]] = OpFunctionParameter %[[#GlbPtrTy]]
+; CHECK: %[[#Width:]] = OpFunctionParameter %[[#Int32Ty]]
+; CHECK: %[[#Height:]] = OpFunctionParameter %[[#Int32Ty]]
+; CHECK: %[[#Pitch:]] = OpFunctionParameter %[[#Int32Ty]]
+; CHECK: %[[#Coord:]] = OpFunctionParameter %[[#VectorTy]]
+; CHECK: %[[#Dst:]] = OpFunctionParameter %[[#PrvPtrTy]]
+; CHECK: %[[#Src:]] = OpFunctionParameter %[[#PrvPtrTy]]
+; CHECK: OpSubgroup2DBlockLoadINTEL %[[#Const42]] %[[#Const42]] %[[#Const42]] %[[#Const42]] %[[#BaseSrc]] %[[#Width]] %[[#Height]] %[[#Pitch]] %[[#Coord]] %[[#Dst]]
+; CHECK: OpSubgroup2DBlockLoadTransformINTEL %[[#Const42]] %[[#Const42]] %[[#Const42]] %[[#Const42]] %[[#BaseSrc]] %[[#Width]] %[[#Height]] %[[#Pitch]] %[[#Coord]] %[[#Dst]]
+; CHECK: OpSubgroup2DBlockLoadTransposeINTEL %[[#Const42]] %[[#Const42]] %[[#Const42]] %[[#Const42]] %[[#BaseSrc]] %[[#Width]] %[[#Height]] %[[#Pitch]] %[[#Coord]] %[[#Dst]]
+; CHECK: OpSubgroup2DBlockPrefetchINTEL %[[#Const42]] %[[#Const42]] %[[#Const42]] %[[#Const42]] %[[#BaseSrc]] %[[#Width]] %[[#Height]] %[[#Pitch]] %[[#Coord]]
+; CHECK: OpSubgroup2DBlockStoreINTEL %[[#Const42]] %[[#Const42]] %[[#Const42]] %[[#Const42]] %[[#Src]] %[[#BaseDst]] %[[#Width]] %[[#Height]] %[[#Pitch]] %[[#Coord]]
+
+
+
+define spir_func void @foo(ptr addrspace(1) %base_address, ptr addrspace(1) %dst_base_pointer, i32 %width, i32 %height, i32 %pitch, <2 x i32> %coord, ptr %dst_pointer, ptr %src_pointer) {
+entry:
+  call spir_func void @_Z32__spirv_Subgroup2DBlockLoadINTELiiiiPU3AS1KviiiDv2_iPv(i32 42, i32 42, i32 42, i32 42, ptr addrspace(1) %base_address, i32 %width, i32 %height, i32 %pitch, <2 x i32> %coord, ptr %dst_pointer)
+  call spir_func void @_Z41__spirv_Subgroup2DBlockLoadTransformINTELiiiiPU3AS1KviiiDv2_iPv(i32 42, i32 42, i32 42, i32 42, ptr addrspace(1) %base_address, i32 %width, i32 %height, i32 %pitch, <2 x i32> %coord, ptr %dst_pointer)
+  call spir_func void @_Z41__spirv_Subgroup2DBlockLoadTransposeINTELiiiiPU3AS1KviiiDv2_iPv(i32 42, i32 42, i32 42, i32 42, ptr addrspace(1) %base_address, i32 %width, i32 %height, i32 %pitch, <2 x i32> %coord, ptr %dst_pointer)
+  call spir_func void @_Z36__spirv_Subgroup2DBlockPrefetchINTELiiiiPU3AS1KviiiDv2_i(i32 42, i32 42, i32 42, i32 42, ptr addrspace(1) %base_address, i32 %width, i32 %height, i32 %pitch, <2 x i32> %coord)
+  call spir_func void @_Z33__spirv_Subgroup2DBlockStoreINTELiiiiPKvPU3AS1viiiDv2_i(i32 42, i32 42, i32 42, i32 42, ptr %src_pointer, ptr addrspace(1) %dst_base_pointer, i32 %width, i32 %height, i32 %pitch, <2 x i32> %coord)
+  ret void
+}
+
+declare spir_func void @_Z32__spirv_Subgroup2DBlockLoadINTELiiiiPU3AS1KviiiDv2_iPv(i32, i32, i32, i32, ptr addrspace(1), i32, i32, i32, <2 x i32>, ptr)
+declare spir_func void @_Z41__spirv_Subgroup2DBlockLoadTransformINTELiiiiPU3AS1KviiiDv2_iPv(i32, i32, i32, i32, ptr addrspace(1), i32, i32, i32, <2 x i32>, ptr)
+declare spir_func void @_Z41__spirv_Subgroup2DBlockLoadTransposeINTELiiiiPU3AS1KviiiDv2_iPv(i32, i32, i32, i32, ptr addrspace(1), i32, i32, i32, <2 x i32>, ptr)
+declare spir_func void @_Z36__spirv_Subgroup2DBlockPrefetchINTELiiiiPU3AS1KviiiDv2_i(i32, i32, i32, i32, ptr addrspace(1), i32, i32, i32, <2 x i32>)
+declare spir_func void @_Z33__spirv_Subgroup2DBlockStoreINTELiiiiPKvPU3AS1viiiDv2_i(i32, i32, i32, i32, ptr, ptr addrspace(1), i32, i32, i32, <2 x i32>)
+
+!opencl.spir.version = !{!0}
+!spirv.Source = !{!1}
+!llvm.ident = !{!2}
+
+!0 = !{i32 1, i32 0}
+!1 = !{i32 4, i32 100000}
+!2 = !{!"clang version 17.0.0"}