fixes for full column, review comment addressing

nod-ai · Oct 8, 2024 · f51056d · f51056d
1 parent 3784918
commit f51056d
Show file tree

Hide file tree

Showing 4 changed files with 4 additions and 10 deletions.
diff --git a/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/AMDAIEPackAndTranspose.cpp b/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/AMDAIEPackAndTranspose.cpp
@@ -7,11 +7,8 @@
 #include "iree-amd-aie/IR/AMDAIEAttrs.h"
 #include "iree-amd-aie/Transforms/Passes.h"
 #include "iree/compiler/Codegen/Dialect/Codegen/IR/IREECodegenAttrs.h"
-#include "mlir/Dialect/Linalg/IR/LinalgInterfaces.h"
 #include "mlir/Dialect/Linalg/Transforms/Transforms.h"
-#include "mlir/Dialect/Tensor/IR/Tensor.h"
 #include "mlir/Dialect/Utils/StaticValueUtils.h"
-#include "mlir/IR/Diagnostics.h"
 #include "mlir/Pass/Pass.h"
 
 #define DEBUG_TYPE "iree-amdaie-pack-and-transpose"

diff --git a/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/KernelDispatch.cpp b/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/KernelDispatch.cpp
@@ -574,7 +574,7 @@ static LogicalResult setRootConfigForConvDecomposePipeline(
     // Target one column of 4 cores, each core processing a different
     // output image row. TODO(newling) use 4x4 array.
     // https://github.com/nod-ai/iree-amd-aie/issues/821
-    tileSizeLevel0 = {1, 4 * OH_1, OW_0, OC_0, 0, 0};
+    tileSizeLevel0 = {1, 4 * OH_1, OW_0, OC_1, 0, 0};
     tileSizeLevel1 = {1, OH_1, OW_0, OC_1, 0, 0};
     tileSizeLevel2 = {0, 0, 0, 0, 1, 1, 0};
   }

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
@@ -6,8 +6,6 @@
 
 #include "iree-amd-aie/Transforms/Passes.h"
 
-#include <memory>
-
 #include "aie/Passes.h"
 #include "aievec/Passes.h"
 #include "air/Conversion/AIRLoweringPass.h"
@@ -36,7 +34,6 @@
 #include "mlir/Dialect/Affine/Passes.h"
 #include "mlir/Dialect/Func/IR/FuncOps.h"
 #include "mlir/Dialect/Linalg/Passes.h"
-#include "mlir/Dialect/MemRef/IR/MemRef.h"
 #include "mlir/Dialect/MemRef/Transforms/Passes.h"
 #include "mlir/Pass/PassManager.h"
 #include "mlir/Transforms/Passes.h"

diff --git a/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/test/lowering_strategy_conv.mlir b/compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/test/lowering_strategy_conv.mlir
@@ -2,7 +2,7 @@
 
 
 
-// CHECK{LITERAL}: #config = #iree_codegen.lowering_config<tile_sizes = [[1, 1, 4, 4, 0, 0, 0], [1, 1, 4, 4, 0, 0, 0], [0, 0, 0, 0, 1, 1, 1, 0, 0]]>
+// CHECK{LITERAL}: #config = #iree_codegen.lowering_config<tile_sizes = [[1, 4, 4, 4, 0, 0, 0], [1, 1, 4, 4, 0, 0, 0], [0, 0, 0, 0, 1, 1, 1, 0, 0]]>
 // CHECK{LITERAL}: #packingConfig = #amdaie.packing_config<packing_config = [{packedSizes = [0, 0, 0, 4, 0, 0, 8], transposePackIndices = [0, 1, 2], unpackEmpty = [false, false, true], innerPerm = [[], [1, 0], []], outerPerm = [[0, 1, 3, 2], [], [0, 1, 2, 3]]}]>
 #pipeline_layout = #hal.pipeline.layout<bindings = [
   <storage_buffer>,
@@ -29,7 +29,7 @@ func.func @conv_static_dispatch_0_conv_2d_nhwc_hwcf_2x12x12x64x3x3x32_bf16xbf16x
 
 // -----
 
-// CHECK{LITERAL}: #config = #iree_codegen.lowering_config<tile_sizes = [[1, 1, 4, 4, 0, 0], [1, 1, 4, 4, 0, 0], [0, 0, 0, 0, 1, 1, 0]]>
+// CHECK{LITERAL}: #config = #iree_codegen.lowering_config<tile_sizes = [[1, 4, 4, 4, 0, 0], [1, 1, 4, 4, 0, 0], [0, 0, 0, 0, 1, 1, 0]]>
 // CHECK{LITERAL}: #packingConfig = #amdaie.packing_config<packing_config = [{packedSizes = [0, 0, 0, 4, 0, 0], transposePackIndices = [0, 1, 2], unpackEmpty = [false, false, true], innerPerm = [[], [], []], outerPerm = [[0, 1, 2, 3], [0, 1, 2], [0, 1, 2, 3]]}]>
 #pipeline_layout = #hal.pipeline.layout<bindings = [
   <storage_buffer>,
@@ -57,7 +57,7 @@ func.func @conv_depthwise_channel_last_bf16(){
 // -----
 // Same test as above, but where the operand type is i8. In this case we expect OC tile size 8  (not 4) at level 1. This is because of the instruction size of AIE.
 
-// CHECK{LITERAL}: #config = #iree_codegen.lowering_config<tile_sizes = [[1, 1, 4, 8, 0, 0], [1, 1, 4, 8, 0, 0], [0, 0, 0, 0, 1, 1, 0]]>
+// CHECK{LITERAL}: #config = #iree_codegen.lowering_config<tile_sizes = [[1, 4, 4, 8, 0, 0], [1, 1, 4, 8, 0, 0], [0, 0, 0, 0, 1, 1, 0]]>
 // CHECK{LITERAL}: #packingConfig = #amdaie.packing_config<packing_config = [{packedSizes = [0, 0, 0, 8, 0, 0], transposePackIndices = [0, 1, 2], unpackEmpty = [false, false, true], innerPerm = [[], [], []], outerPerm = [[0, 1, 2, 3], [0, 1, 2], [0, 1, 2, 3]]}]>
 #pipeline_layout = #hal.pipeline.layout<bindings = [
   <storage_buffer>,