Rename + refactor + lit test fix

nod-ai · Aug 29, 2024 · f797a36 · f797a36
1 parent f99dd46
commit f797a36
Show file tree

Hide file tree

Showing 8 changed files with 23 additions and 25 deletions.
diff --git a/...nsforms/AMDAIESplitLogicalObjectFifos.cpp → ...plitLogicalObjFifosForConnectionReuse.cpp b/...nsforms/AMDAIESplitLogicalObjectFifos.cpp → ...plitLogicalObjFifosForConnectionReuse.cpp
@@ -9,7 +9,7 @@
 #include "mlir/IR/Iterators.h"
 #include "mlir/Pass/Pass.h"
 
-#define DEBUG_TYPE "iree-amdaie-split-logical-objectfifos"
+#define DEBUG_TYPE "iree-amdaie-split-logical-objectfifos-for-connection-reuse"
 
 namespace mlir::iree_compiler::AMDAIE {
 
@@ -33,19 +33,20 @@ static SmallVector<AMDAIE::DmaCpyNdOp> fetchDmaCpyNdOpsToSplit(
   return l2ToL1DmaOps;
 }
 
-class AMDAIESplitLogicalObjectFifosPass
-    : public impl::AMDAIESplitLogicalObjectFifosBase<
-          AMDAIESplitLogicalObjectFifosPass> {
+class AMDAIESplitLogicalObjFifosForConnectionReusePass
+    : public impl::AMDAIESplitLogicalObjFifosForConnectionReuseBase<
+          AMDAIESplitLogicalObjFifosForConnectionReusePass> {
  public:
-  using AMDAIESplitLogicalObjectFifosBase::AMDAIESplitLogicalObjectFifosBase;
+  using AMDAIESplitLogicalObjFifosForConnectionReuseBase::
+      AMDAIESplitLogicalObjFifosForConnectionReuseBase;
 
   void getDependentDialects(DialectRegistry &registry) const override {
     registry.insert<AMDAIEDialect>();
   }
   void runOnOperation() override;
 };
 
-void AMDAIESplitLogicalObjectFifosPass::runOnOperation() {
+void AMDAIESplitLogicalObjFifosForConnectionReusePass::runOnOperation() {
   ModuleOp moduleOp = getOperation();
   MLIRContext *context = &getContext();
   IRRewriter rewriter(context);
@@ -60,8 +61,8 @@ void AMDAIESplitLogicalObjectFifosPass::runOnOperation() {
 
 }  // namespace
 
-std::unique_ptr<Pass> createAMDAIESplitLogicalObjectFifosPass() {
-  return std::make_unique<AMDAIESplitLogicalObjectFifosPass>();
+std::unique_ptr<Pass> createAMDAIESplitLogicalObjFifosForConnectionReusePass() {
+  return std::make_unique<AMDAIESplitLogicalObjFifosForConnectionReusePass>();
 }
 
 }  // namespace mlir::iree_compiler::AMDAIE
diff --git a/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/CMakeLists.txt b/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/CMakeLists.txt
@@ -86,7 +86,7 @@ iree_cc_library(
     "AMDAIEPad.cpp"
     "AMDAIEPeelForLoop.cpp"
     "AMDAIEPropagateDataLayout.cpp"
-    "AMDAIESplitLogicalObjectFifos.cpp"
+    "AMDAIESplitLogicalObjFifosForConnectionReuse.cpp"
     "AMDAIETile.cpp"
     "AMDAIETileAndFuse.cpp"
     "AMDAIEUtils.cpp"

diff --git a/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/PassDetail.h b/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/PassDetail.h
@@ -65,7 +65,7 @@ namespace mlir::iree_compiler::AMDAIE {
 #define GEN_PASS_DEF_AMDAIEPAD
 #define GEN_PASS_DEF_AMDAIEPEELFORLOOP
 #define GEN_PASS_DEF_AMDAIEPROPAGATEDATALAYOUT
-#define GEN_PASS_DEF_AMDAIESPLITLOGICALOBJECTFIFOS
+#define GEN_PASS_DEF_AMDAIESPLITLOGICALOBJFIFOSFORCONNECTIONREUSE
 #define GEN_PASS_DEF_AMDAIETILE
 #define GEN_PASS_DEF_AMDAIETILEANDFUSE
 #define GEN_PASS_DEF_AMDAIEVECTORIZATION

diff --git a/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/Passes.cpp b/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/Passes.cpp
@@ -582,7 +582,7 @@ void addAMDAIEObjectFifoLoweringPasses(OpPassManager &passManager) {
   passManager.addPass(createAMDAIEDistributeCoresAndObjectFifosPass());
   passManager.addPass(createCSEPass());
   passManager.addPass(createCanonicalizerPass());
-  passManager.addPass(createAMDAIESplitLogicalObjectFifosPass());
+  passManager.addPass(createAMDAIESplitLogicalObjFifosForConnectionReusePass());
 
   passManager.addPass(createAMDAIEDmaToCircularDmaPass());
   passManager.addNestedPass<func::FuncOp>(createAMDAIECreateAIEWorkgroupPass());

diff --git a/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/Passes.h b/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/Passes.h
@@ -212,7 +212,7 @@ std::unique_ptr<Pass> createAMDAIEPeelForLoopPass(
     AMDAIEPeelForLoopOptions options = {});
 
 /// Create a pass to split buffers.
-std::unique_ptr<Pass> createAMDAIESplitLogicalObjectFifosPass();
+std::unique_ptr<Pass> createAMDAIESplitLogicalObjFifosForConnectionReusePass();
 
 /// Create pass to tile TilingInterface operations.
 std::unique_ptr<Pass> createAMDAIETilePass(AMDAIETileOptions options = {});

diff --git a/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/Passes.td b/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/Passes.td
@@ -430,10 +430,10 @@ def AMDAIEPropagateDataLayout :
   let constructor = "mlir::iree_compiler::AMDAIE::createAMDAIEPropagateDataLayoutPass()";
 }
 
-def AMDAIESplitLogicalObjectFifos :
-  Pass<"iree-amdaie-split-logical-objectfifos", "ModuleOp"> {
+def AMDAIESplitLogicalObjFifosForConnectionReuse :
+  Pass<"iree-amdaie-split-logical-objectfifos-for-connection-reuse", "ModuleOp"> {
   let summary = "Pass to split L2 buffers to share inputs of Matmul and Elementwise operations.";
-  let constructor = "mlir::iree_compiler::AMDAIE::createAMDAIESplitLogicalObjectFifosPass()";
+  let constructor = "mlir::iree_compiler::AMDAIE::createAMDAIESplitLogicalObjFifosForConnectionReusePass()";
 }
 
 def AMDAIETile :

diff --git a/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/test/CMakeLists.txt b/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/test/CMakeLists.txt
@@ -57,7 +57,7 @@ iree_lit_test_suite(
     "pad.mlir"
     "peel_for_loop.mlir"
     "propagate_data_layout.mlir"
-    "split_logical_objectfifos.mlir"
+    "split_logicalobjfifos_for_connection_reuse.mlir"
     "tile_and_fuse_using_scf_for.mlir"
     "tile_and_fuse_using_scf_forall.mlir"
     "tile_copy_using_scf_for.mlir"

diff --git a/...forms/test/split_logical_objectfifos.mlir → ...logicalobjfifos_for_connection_reuse.mlir b/...forms/test/split_logical_objectfifos.mlir → ...logicalobjfifos_for_connection_reuse.mlir
@@ -1,4 +1,4 @@
-// RUN: iree-opt --pass-pipeline="builtin.module(iree-amdaie-split-logical-objectfifos,cse)" --split-input-file --verify-diagnostics %s | FileCheck %s
+// RUN: iree-opt --pass-pipeline="builtin.module(iree-amdaie-split-logical-objectfifos-for-connection-reuse,cse)" --split-input-file --verify-diagnostics %s | FileCheck %s
 
 // Glossary:
 // candidate core op : they are those amdaie.core ops which have at least three input dma ops.
@@ -429,8 +429,6 @@ module {
 //   CHECK-DAG:   %[[C1:.*]] = arith.constant 1 : index
 //   CHECK-DAG:   %[[C2:.*]] = arith.constant 2 : index
 //   CHECK-DAG:   %[[C3:.*]] = arith.constant 3 : index
-//   CHECK-DAG:   %[[C34:.*]] = arith.constant 34 : index
-//   CHECK-DAG:   %[[C35:.*]] = arith.constant 35 : index
 //   CHECK-DAG:   %[[L3_ALLOC:.*]] = memref.alloc() : memref<128x128xi32>
 //   CHECK-DAG:   %[[L2_ALLOC_0:.*]] = memref.alloc() : memref<1x1x32x32xi32, 1 : i32>
 //   CHECK-DAG:   %[[L2_ALLOC_1:.*]] = memref.alloc() : memref<1x1x32x32xi32, 1 : i32>
@@ -456,16 +454,16 @@ module {
 //       CHECK:   scf.forall (%[[IV0:.*]], %[[IV1:.*]]) in (2, 2)
 //       CHECK:       %[[DMA_CPY_ND_L3_TO_L2_0:.*]] = amdaie.dma_cpy_nd(
 //  CHECK-SAME:                                         %[[L2_OBJECTFIFO_0]][0, 0, 0, 0] [1, 1, 32, 32] [2048, 1024, 32, 1]
-//  CHECK-SAME:                                         %[[L3_OBJECTFIFO]][0, 0, %[[C3]], %[[C2]]] [1, 1, 32, 32] [4096, 32, 128, 1]
+//  CHECK-SAME:                                         %[[L3_OBJECTFIFO]][0, 0, 3, 2] [1, 1, 32, 32] [4096, 32, 128, 1]
 //       CHECK:       %[[DMA_CPY_ND_L3_TO_L2_1:.*]] = amdaie.dma_cpy_nd(
 //  CHECK-SAME:                                         %[[L2_OBJECTFIFO_1]][0, 0, 0, 0] [1, 1, 32, 32] [2048, 1024, 32, 1]
-//  CHECK-SAME:                                         %[[L3_OBJECTFIFO]][0, 0, %[[C3]], %[[C34]]] [1, 1, 32, 32] [4096, 32, 128, 1]
+//  CHECK-SAME:                                         %[[L3_OBJECTFIFO]][0, 0, 3, 34] [1, 1, 32, 32] [4096, 32, 128, 1]
 //       CHECK:       %[[DMA_CPY_ND_L3_TO_L2_2:.*]] = amdaie.dma_cpy_nd(
 //  CHECK-SAME:                                         %[[L2_OBJECTFIFO_2]][0, 0, 0, 0] [1, 1, 32, 32] [2048, 1024, 32, 1]
-//  CHECK-SAME:                                         %[[L3_OBJECTFIFO]][0, 0, %[[C35]], %[[C2]]] [1, 1, 32, 32] [4096, 32, 128, 1]
+//  CHECK-SAME:                                         %[[L3_OBJECTFIFO]][0, 0, 35, 2] [1, 1, 32, 32] [4096, 32, 128, 1]
 //       CHECK:       %[[DMA_CPY_ND_L3_TO_L2_3:.*]] = amdaie.dma_cpy_nd(
 //  CHECK-SAME:                                         %[[L2_OBJECTFIFO_3]][0, 0, 0, 0] [1, 1, 32, 32] [2048, 1024, 32, 1]
-//  CHECK-SAME:                                         %[[L3_OBJECTFIFO]][0, 0, %[[C35]], %[[C34]]] [1, 1, 32, 32] [4096, 32, 128, 1]
+//  CHECK-SAME:                                         %[[L3_OBJECTFIFO]][0, 0, 35, 34] [1, 1, 32, 32] [4096, 32, 128, 1]
 //       CHECK:       amdaie.logicalobjectfifo.from_memref
 //       CHECK:       amdaie.logicalobjectfifo.from_memref
 //       CHECK:       amdaie.dma_cpy_nd
@@ -785,8 +783,7 @@ module {
 //   CHECK-DAG: #map = affine_map<(d0) -> (d0 + 32)>
 //   CHECK-DAG: #map1 = affine_map<(d0) -> (d0)>
 //       CHECK: @block_argument_of_funcOp_offset
-//  CHECK-SAME:   %[[ARG0:.*]]: index,
-//  CHECK-SAME:   %[[ARG1:.*]]: index,
+//  CHECK-SAME:   %[[ARG0:.*]]: index, %[[ARG1:.*]]: index,
 //   CHECK-DAG:   %[[IV1_32:.*]] = affine.apply #map(%[[ARG1]])
 //   CHECK-DAG:   %[[IV0_32:.*]] = affine.apply #map(%[[ARG0]])
 //   CHECK-DAG:   %[[IV1_0:.*]] = affine.apply #map1(%[[ARG1]])