[midend][tests] Add batch level parallelism outer product BatchMatMul…

… optimization pass and tests.

examples/MLIRLinalg/linalg-batch-matmul.mlir

@@ -0,0 +1,29 @@
+// RUN: buddy-opt -batchmatmul-optimize -verify-diagnostics -expand-strided-metadata -lower-affine -convert-vector-to-llvm -finalize-memref-to-llvm -convert-scf-to-cf -convert-linalg-to-llvm -llvm-request-c-wrappers -convert-func-to-llvm -reconcile-unrealized-casts %s \
+// RUN: | mlir-cpu-runner -O0 -e buddy_batchmatmul_f32 \
+// RUN: -shared-libs=%mlir_runner_utils_dir/libmlir_runner_utils%shlibext,%mlir_runner_utils_dir/libmlir_c_runner_utils%shlibext \
+// RUN: | FileCheck %s
+
+memref.global "private" @A : memref<2x2x3xf32> = dense<[[[9., 4., 6.],[2., 4., 0.]],[[6., 3., 3.],[0., 4., 7.]]]>
+memref.global "private" @B : memref<2x3x4xf32> = dense<[[[1., 3., 8., 0.],[1., 8., 8., 7.], [6., 9., 7., 9.]],[[3., 8., 6., 8.],[2., 7., 0., 6.],[0., 4., 0., 4.]]]>
+memref.global "private" @C : memref<2x2x4xf32> = dense<[[[ 49., 113., 146.,  82.],[  6.,  38.,  48.,  28.]],[[ 24.,  81.,  36.,  78.],[  8.,  56.,   0.,  52.]]]>
+
+func.func private @printMemrefF32(memref<*xf32>) attributes { llvm.emit_c_interface }
+
+func.func @buddy_batchmatmul_f32() -> f32{
+  %a = memref.get_global @A : memref<2x2x3xf32>
+  %b = memref.get_global @B : memref<2x3x4xf32>
+  %c = memref.get_global @C : memref<2x2x4xf32>
+
+  linalg.batch_matmul 
+      ins(%a, %b: memref<2x2x3xf32>, memref<2x3x4xf32>)
+      outs(%c: memref<2x2x4xf32>)
+  %printed_c = memref.cast %c : memref<2x2x4xf32> to memref<*xf32>
+  call @printMemrefF32(%printed_c) : (memref<*xf32>) -> ()
+  // CHECK: {{Unranked Memref base@ = 0x[0-9A-Fa-f]{1,} rank = 3 offset = 0 sizes = \[2, 2, 4\] strides = \[8, 4, 1\] data =}}
+  // CHECK{LITERAL}: [[[98,    226,    292,    164], 
+  // CHECK{LITERAL}:   [12,    76,    96,    56]], 
+  // CHECK{LITERAL}:  [[48,    162,    72,    156], 
+  // CHECK{LITERAL}:   [16,    112,    0,    104]]]
+  %zero = arith.constant 0.0 :f32
+  return %zero :f32
+}

examples/MLIRLinalg/makefile

@@ -136,6 +136,61 @@ linalg-matmul-optimize-run:
 	${MLIR_CPU_RUNNER} ${OPT_FLAG} -e main -entry-point-result=void \
 		-shared-libs=${MLIR_RUNNER_UTILS} -shared-libs=${MLIR_C_RUNNER_UTILS}
 
+linalg-batch-matmul-optimize-run:
+	@${BUDDY_OPT} linalg-matmul.mlir ${MLIR_OPT_OPTIONS} \
+		-batchmatmul-optimize="step-placeholder=64" \
+		-convert-linalg-to-loops \
+		-expand-strided-metadata \
+		-lower-affine \
+		-convert-scf-to-cf \
+		-convert-vector-to-llvm \
+		-finalize-memref-to-llvm \
+		-convert-arith-to-llvm \
+		-convert-func-to-llvm \
+		-reconcile-unrealized-casts | \
+	${MLIR_CPU_RUNNER} ${OPT_FLAG} -e main -entry-point-result=void \
+		-shared-libs=${MLIR_RUNNER_UTILS} -shared-libs=${MLIR_C_RUNNER_UTILS}
+
+linalg-batch-matmul-lower:
+	@${MLIR_OPT} linalg-batch-matmul.mlir ${MLIR_OPT_OPTIONS} \
+		-convert-linalg-to-loops -lower-affine -convert-scf-to-cf \
+		-convert-vector-to-llvm -finalize-memref-to-llvm -convert-arith-to-llvm \
+		-convert-func-to-llvm -reconcile-unrealized-casts \
+		-o ./log.mlir
+
+linalg-batch-matmul-translate:
+	@${MLIR_OPT} linalg-batch-matmul.mlir ${MLIR_OPT_OPTIONS} \
+		-convert-linalg-to-loops -lower-affine -convert-scf-to-cf \
+		-convert-vector-to-llvm -finalize-memref-to-llvm -convert-arith-to-llvm \
+		-convert-func-to-llvm -reconcile-unrealized-casts | \
+	${MLIR_TRANSLATE} --mlir-to-llvmir -o log.ll
+
+linalg-batch-matmul-run:
+	@${MLIR_OPT} linalg-batch-matmul.mlir ${MLIR_OPT_OPTIONS} \
+		-convert-linalg-to-loops -lower-affine -convert-scf-to-cf \
+		-convert-vector-to-llvm -finalize-memref-to-llvm -convert-arith-to-llvm \
+		-convert-func-to-llvm -reconcile-unrealized-casts | \
+	${MLIR_CPU_RUNNER} ${OPT_FLAG} -e main -entry-point-result=void -shared-libs=${MLIR_RUNNER_UTILS} -shared-libs=${MLIR_C_RUNNER_UTILS}
+
+linalg-batch-matmul-optimize-lower:
+	@${BUDDY_OPT} linalg-batch-matmul.mlir ${MLIR_OPT_OPTIONS} \
+		-batchmatmul-optimize="step-placeholder=64" \
+		-o ./log.mlir
+
+linalg-batch-matmul-optimize-translate:
+	@${BUDDY_OPT} linalg-batch-matmul.mlir ${MLIR_OPT_OPTIONS} \
+		-batchmatmul-optimize="step-placeholder=64" \
+		-convert-linalg-to-loops \
+		-expand-strided-metadata \
+		-lower-affine \
+		-convert-scf-to-cf \
+		-convert-vector-to-llvm \
+		-finalize-memref-to-llvm \
+		-convert-arith-to-llvm \
+		-convert-func-to-llvm \
+		-reconcile-unrealized-casts | \
+	${MLIR_TRANSLATE} --mlir-to-llvmir -o log.ll
+
 linalg-conv2d_nchw_fchw-lower:
 	@${MLIR_OPT} ./linalg-conv2d_nchw_fchw.mlir \
 		-convert-linalg-to-loops -o ./log.mlir