[mlir][spirv] Convert gpu.barrier to spv.ControlBarrier

[antiagainst]

Reorder CHECK

mlir/lib/Conversion/GPUToSPIRV/GPUToSPIRV.cpp

 //===- GPUToSPIRV.cpp - GPU to SPIR-V Patterns ----------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file implements patterns to convert GPU dialect to SPIR-V dialect.
 //
 //===----------------------------------------------------------------------===//
 
 #include "mlir/Conversion/GPUToSPIRV/GPUToSPIRV.h"
 #include "mlir/Dialect/GPU/GPUDialect.h"
 #include "mlir/Dialect/SPIRV/IR/SPIRVDialect.h"
+#include "mlir/Dialect/SPIRV/IR/SPIRVEnums.h"
 #include "mlir/Dialect/SPIRV/IR/SPIRVOps.h"
 #include "mlir/Dialect/SPIRV/IR/TargetAndABI.h"
 #include "mlir/Dialect/SPIRV/Transforms/SPIRVConversion.h"
 #include "mlir/IR/BuiltinOps.h"
 #include "mlir/Transforms/DialectConversion.h"
-#include "llvm/ADT/StringSwitch.h"
 
 using namespace mlir;
 
 static constexpr const char kSPIRVModule[] = "__spv__";
 
 namespace {
 /// Pattern lowering GPU block/thread size/id to loading SPIR-V invocation
 /// builtin variables.

 public:
   using OpConversionPattern<gpu::ReturnOp>::OpConversionPattern;
 
   LogicalResult
   matchAndRewrite(gpu::ReturnOp returnOp, OpAdaptor adaptor,
                   ConversionPatternRewriter &rewriter) const override;
 };
 
+/// Pattern to convert a gpu.barrier op into a spv.ControlBarrier op.
+class GPUBarrierConversion final : public OpConversionPattern<gpu::BarrierOp> {
+public:
+  using OpConversionPattern::OpConversionPattern;
+
+  LogicalResult
+  matchAndRewrite(gpu::BarrierOp barrierOp, OpAdaptor adaptor,
+                  ConversionPatternRewriter &rewriter) const override;
+};
+
 } // namespace
 
 //===----------------------------------------------------------------------===//
 // Builtins.
 //===----------------------------------------------------------------------===//
 
 template <typename SourceOp, spirv::BuiltIn builtin>
 LogicalResult LaunchConfigConversion<SourceOp, builtin>::matchAndRewrite(
LogicalResult GPUReturnOpConversion::matchAndRewrite(
   if (!adaptor.getOperands().empty())
     return failure();
 
   rewriter.replaceOpWithNewOp<spirv::ReturnOp>(returnOp);
   return success();
 }
 
 //===----------------------------------------------------------------------===//
+// Barrier.
+//===----------------------------------------------------------------------===//
+
+LogicalResult GPUBarrierConversion::matchAndRewrite(
+    gpu::BarrierOp barrierOp, OpAdaptor adaptor,
+    ConversionPatternRewriter &rewriter) const {
+  MLIRContext *context = getContext();
+  // Both execution and memory scope should be workgroup.
+  auto scope = spirv::ScopeAttr::get(context, spirv::Scope::Workgroup);
+  // Require acquire and release memory semantics for workgroup memory.
+  auto memorySemantics = spirv::MemorySemanticsAttr::get(
+      context, spirv::MemorySemantics::WorkgroupMemory |
+                   spirv::MemorySemantics::AcquireRelease);
+  rewriter.replaceOpWithNewOp<spirv::ControlBarrierOp>(barrierOp, scope, scope,
+                                                       memorySemantics);
+  return success();
+}
+
+//===----------------------------------------------------------------------===//
 // GPU To SPIRV Patterns.
 //===----------------------------------------------------------------------===//
 
 void mlir::populateGPUToSPIRVPatterns(SPIRVTypeConverter &typeConverter,
                                       RewritePatternSet &patterns) {
   patterns.add<
-      GPUFuncOpConversion, GPUModuleConversion, GPUModuleEndConversion,
-      GPUReturnOpConversion,
+      GPUBarrierConversion, GPUFuncOpConversion, GPUModuleConversion,
+      GPUModuleEndConversion, GPUReturnOpConversion,
       LaunchConfigConversion<gpu::BlockIdOp, spirv::BuiltIn::WorkgroupId>,
       LaunchConfigConversion<gpu::GridDimOp, spirv::BuiltIn::NumWorkgroups>,
       LaunchConfigConversion<gpu::ThreadIdOp,
                              spirv::BuiltIn::LocalInvocationId>,
       SingleDimLaunchConfigConversion<gpu::SubgroupIdOp,
                                       spirv::BuiltIn::SubgroupId>,
       SingleDimLaunchConfigConversion<gpu::NumSubgroupsOp,
                                       spirv::BuiltIn::NumSubgroups>,
       SingleDimLaunchConfigConversion<gpu::SubgroupSizeOp,
                                       spirv::BuiltIn::SubgroupSize>,
       WorkGroupSizeConversion>(typeConverter, patterns.getContext());
 }

mlir/test/Conversion/GPUToSPIRV/simple.mlir

gpu.func @missing_entry_point_abi(
       %arg1 : memref<12xf32>
         {spv.interface_var_abi = #spv.interface_var_abi<(3, 0)>}) kernel
     attributes
       {spv.entry_point_abi = {local_size = dense<[32, 4, 1]>: vector<3xi32>}} {
       gpu.return
     }
   }
 }
+
+// -----
+
+module attributes {gpu.container_module} {
+  gpu.module @kernels {
+    // CHECK-LABEL: spv.func @barrier
+    gpu.func @barrier(%arg0 : f32, %arg1 : memref<12xf32>) kernel
+      attributes {spv.entry_point_abi = {local_size = dense<[32, 4, 1]>: vector<3xi32>}} {
+      // CHECK: spv.ControlBarrier Workgroup, Workgroup, "AcquireRelease|WorkgroupMemory"
+      gpu.barrier
+      gpu.return
+    }
+  }
+
+  func @main() {
+    %0 = "op"() : () -> (f32)
+    %1 = "op"() : () -> (memref<12xf32>)
+    %cst = arith.constant 1 : index
+    gpu.launch_func @kernels::@barrier
+        blocks in (%cst, %cst, %cst) threads in (%cst, %cst, %cst)
+        args(%0 : f32, %1 : memref<12xf32>)
+    return
+  }
+}