[SPIRV] Enable OpenCL max_work_group_size translation via SPV_INTEL_kernel_attributes (#165891)

This adds BE support for the
[`SPV_INTEL_kernel_attributes`](https://github.khronos.org/SPIRV-Registry/extensions/INTEL/SPV_INTEL_kernel_attributes.html)
extension. The extension is necessary to encode the rather useful
`max_work_group_size` kernel attribute, via `OpExecutionMode
MaxWorkgroupSizeINTEL`, which is the only Execution Mode added by the
extension that this patch adds full processing for. Future patches will
add the other Execution Modes and Capabilities. The test is adapted from
the equivalent Translator test; it depends on #165815.

Author: AlexVlx

llvm/docs/SPIRVUsage.rst

@@ -187,6 +187,8 @@ Below is a list of supported SPIR-V extensions, sorted alphabetically by their e
      - Adds decorations that can be applied to global (module scope) variables.
    * - ``SPV_INTEL_global_variable_fpga_decorations``
      - Adds decorations that can be applied to global (module scope) variables to help code generation for FPGA devices.
+   * - ``SPV_INTEL_kernel_attributes``
+     - Adds execution modes that can be applied to entry points to inform scheduling.
    * - ``SPV_INTEL_media_block_io``
      - Adds additional subgroup block read and write functionality that allow applications to flexibly specify the width and height of the block to read from or write to a 2D image.
    * - ``SPV_INTEL_memory_access_aliasing``
@@ -226,9 +228,9 @@ Below is a list of supported SPIR-V extensions, sorted alphabetically by their e
    * - ``SPV_INTEL_fp_max_error``
      - Adds the ability to specify the maximum error for floating-point operations.
    * - ``SPV_INTEL_ternary_bitwise_function``
-     - Adds a bitwise instruction on three operands and a look-up table index for specifying the bitwise operation to perform. 
+     - Adds a bitwise instruction on three operands and a look-up table index for specifying the bitwise operation to perform.
    * - ``SPV_INTEL_subgroup_matrix_multiply_accumulate``
-     - Adds an instruction to compute the matrix product of an M x K matrix with a K x N matrix and then add an M x N matrix. 
+     - Adds an instruction to compute the matrix product of an M x K matrix with a K x N matrix and then add an M x N matrix.
    * - ``SPV_INTEL_int4``
      - Adds support for 4-bit integer type, and allow this type to be used in cooperative matrices.
    * - ``SPV_KHR_float_controls2``

llvm/lib/Target/SPIRV/SPIRVAsmPrinter.cpp

@@ -577,6 +577,11 @@ void SPIRVAsmPrinter::outputExecutionMode(const Module &M) {
     if (MDNode *Node = F.getMetadata("intel_reqd_sub_group_size"))
       outputExecutionModeFromMDNode(FReg, Node,
                                     SPIRV::ExecutionMode::SubgroupSize, 0, 0);
+    if (MDNode *Node = F.getMetadata("max_work_group_size")) {
+      if (ST->canUseExtension(SPIRV::Extension::SPV_INTEL_kernel_attributes))
+        outputExecutionModeFromMDNode(
+            FReg, Node, SPIRV::ExecutionMode::MaxWorkgroupSizeINTEL, 3, 1);
+    }
     if (MDNode *Node = F.getMetadata("vec_type_hint")) {
       MCInst Inst;
       Inst.setOpcode(SPIRV::OpExecutionMode);

llvm/lib/Target/SPIRV/SPIRVCommandLine.cpp

@@ -155,7 +155,9 @@ static const std::map<std::string, SPIRV::Extension::Extension, std::less<>>
         {"SPV_INTEL_predicated_io",
          SPIRV::Extension::Extension::SPV_INTEL_predicated_io},
         {"SPV_KHR_maximal_reconvergence",
-         SPIRV::Extension::Extension::SPV_KHR_maximal_reconvergence}};
+         SPIRV::Extension::Extension::SPV_KHR_maximal_reconvergence},
+        {"SPV_INTEL_kernel_attributes",
+         SPIRV::Extension::Extension::SPV_INTEL_kernel_attributes}};
 
 bool SPIRVExtensionsParser::parse(cl::Option &O, StringRef ArgName,
                                   StringRef ArgValue,

llvm/lib/Target/SPIRV/SPIRVModuleAnalysis.cpp

@@ -2180,6 +2180,10 @@ static void collectReqs(const Module &M, SPIRV::ModuleAnalysisInfo &MAI,
       MAI.Reqs.getAndAddRequirements(
           SPIRV::OperandCategory::ExecutionModeOperand,
           SPIRV::ExecutionMode::SubgroupSize, ST);
+    if (F.getMetadata("max_work_group_size"))
+      MAI.Reqs.getAndAddRequirements(
+          SPIRV::OperandCategory::ExecutionModeOperand,
+          SPIRV::ExecutionMode::MaxWorkgroupSizeINTEL, ST);
     if (F.getMetadata("vec_type_hint"))
       MAI.Reqs.getAndAddRequirements(
           SPIRV::OperandCategory::ExecutionModeOperand,

llvm/lib/Target/SPIRV/SPIRVSymbolicOperands.td

@@ -587,6 +587,11 @@ defm CooperativeMatrixBFloat16ComponentTypeINTEL : CapabilityOperand<6437, 0, 0,
 defm RoundToInfinityINTEL : CapabilityOperand<5582, 0, 0, [SPV_INTEL_float_controls2], []>;
 defm FloatingPointModeINTEL : CapabilityOperand<5583, 0, 0, [SPV_INTEL_float_controls2], []>;
 defm FunctionFloatControlINTEL : CapabilityOperand<5821, 0, 0, [SPV_INTEL_float_controls2], []>;
+defm KernelAttributesINTEL : CapabilityOperand<5892, 0, 0, [SPV_INTEL_kernel_attributes], [Kernel]>;
+// TODO-SPIRV: add these once they are used / tested.
+// defm FPGAKernelAttributesINTEL : CapabilityOperand<5897, 0, 0, [SPV_INTEL_kernel_attributes], [Kernel]>;
+// defm FPGAKernelAttributesv2INTEL : CapabilityOperand<6161, 0, 0, [SPV_INTEL_kernel_attributes], [Kernel]>;
+// END TODO-SPIRV
 defm LongCompositesINTEL : CapabilityOperand<6089, 0, 0, [SPV_INTEL_long_composites], []>;
 defm BindlessImagesINTEL : CapabilityOperand<6528, 0, 0, [SPV_INTEL_bindless_images], []>;
 defm MemoryAccessAliasingINTEL : CapabilityOperand<5910, 0, 0, [SPV_INTEL_memory_access_aliasing], []>;
@@ -805,6 +810,15 @@ defm RoundingModeRTPINTEL : ExecutionModeOperand<5620, [RoundToInfinityINTEL]>;
 defm RoundingModeRTNINTEL : ExecutionModeOperand<5621, [RoundToInfinityINTEL]>;
 defm FloatingPointModeALTINTEL : ExecutionModeOperand<5622, [FloatingPointModeINTEL]>;
 defm FloatingPointModeIEEEINTEL : ExecutionModeOperand<5623, [FloatingPointModeINTEL]>;
+defm MaxWorkgroupSizeINTEL : ExecutionModeOperand<5893, [KernelAttributesINTEL]>;
+// TODO-SPIRV: Add the following once they are used / tested.
+// defm MaxWorkDimINTEL : ExecutionModeOperand<5894, [KernelAttributesINTEL]>;
+// defm NoGlobalOffsetINTEL : ExecutionModeOperand<5895, [KernelAttributesINTEL]>;
+// defm NumSIMDWorkitemsINTEL : ExecutionModeOperand<5896, [FPGAKernelAttributesINTEL]>;
+// defm SchedulerTargetFmaxMhzINTEL : ExecutionModeOperand<5903, [FPGAKernelAttributesINTEL]>;
+// defm StreamingInterfaceINTEL : ExecutionModeOperand<6154, [FPGAKernelAttributesv2INTEL]>;
+// defm RegisterMapInterfaceINTEL : ExecutionModeOperand<6160, [FPGAKernelAttributesv2INTEL]>;
+// END TODO-SPIRV
 defm FPFastMathDefault : ExecutionModeOperand<6028, [FloatControls2]>;
 defm MaximallyReconvergesKHR : ExecutionModeOperand<6023, [Shader]>;
 
@@ -1919,7 +1933,7 @@ defm GenericCastToPtr :  SpecConstantOpOperandsOperand<122, [], [Kernel]>;
 defm PtrCastToGeneric :  SpecConstantOpOperandsOperand<121, [], [Kernel]>;
 defm Bitcast :  SpecConstantOpOperandsOperand<124, [], []>;
 defm QuantizeToF16 :  SpecConstantOpOperandsOperand<116, [], [Shader]>;
-// Arithmetic 
+// Arithmetic
 defm SNegate :  SpecConstantOpOperandsOperand<126, [], []>;
 defm Not :  SpecConstantOpOperandsOperand<200, [], []>;
 defm IAdd :  SpecConstantOpOperandsOperand<128, [], []>;

llvm/test/CodeGen/SPIRV/extensions/SPV_INTEL_kernel_attributes/max_work_group_size.ll

@@ -0,0 +1,32 @@
+; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv64-unknown-unknown --spirv-ext=+SPV_INTEL_kernel_attributes %s -o - | FileCheck %s
+; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv64-amd-amdhsa %s -o - | FileCheck %s
+; %if spirv-tools %{ llc -O0 -mtriple=spirv64-unknown-unknown --spirv-ext=+SPV_INTEL_kernel_attributes %s -o - -filetype=obj | spirv-val %}
+; %if spirv-tools %{ llc -O0 -mtriple=spirv64-amd-amdhsa %s -o - -filetype=obj | spirv-val %}
+
+; CHECK: OpCapability KernelAttributesINTEL
+; CHECK: OpExtension "SPV_INTEL_kernel_attributes"
+; CHECK: OpEntryPoint {{.*}} %[[DIM1:[0-9]+]] "Dim1"
+; CHECK: OpEntryPoint {{.*}} %[[DIM2:[0-9]+]] "Dim2"
+; CHECK: OpEntryPoint {{.*}} %[[DIM3:[0-9]+]] "Dim3"
+; CHECK: OpExecutionMode %[[DIM1]] MaxWorkgroupSizeINTEL 4 1 1
+; CHECK: OpExecutionMode %[[DIM2]] MaxWorkgroupSizeINTEL 8 4 1
+; CHECK: OpExecutionMode %[[DIM3]] MaxWorkgroupSizeINTEL 16 8 4
+; CHECK: %[[DIM1]] = OpFunction
+; CHECK: %[[DIM2]] = OpFunction
+; CHECK: %[[DIM3]] = OpFunction
+
+define spir_kernel void @Dim1() !max_work_group_size !0 {
+  ret void
+}
+
+define spir_kernel void @Dim2() !max_work_group_size !1 {
+  ret void
+}
+
+define spir_kernel void @Dim3() !max_work_group_size !2 {
+  ret void
+}
+
+!0 = !{i32 4}
+!1 = !{i32 8, i32 4}
+!2 = !{i32 16, i32 8, i32 4}