84 lines · plain
1// RUN: mlir-opt %s --linalg-generalize-named-ops \2// RUN: --sparsification-and-bufferization | FileCheck %s3 4#CSR_ones_complex = #sparse_tensor.encoding<{5 map = (d0, d1) -> (d0 : dense, d1 : compressed),6 explicitVal = #complex.number<:f32 1.0, 0.0>,7 implicitVal = #complex.number<:f32 0.0, 0.0>8}>9 10#CSR_ones_fp = #sparse_tensor.encoding<{11 map = (d0, d1) -> (d0 : dense, d1 : compressed),12 explicitVal = 1.0 : f32,13 implicitVal = 0.0 : f3214}>15 16#CSR_ones_int = #sparse_tensor.encoding<{17 map = (d0, d1) -> (d0 : dense, d1 : compressed),18 explicitVal = 1 : i32,19 implicitVal = 0 : i3220}>21 22// CHECK-LABEL: func.func @matmul_complex23// CHECK: scf.for24// CHECK: scf.for25// CHECK: %[[X:.*]] = memref.load26// CHECK: scf.for27// CHECK: %[[I:.*]] = memref.load28// CHECK: %[[Y:.*]] = memref.load29// CHECK: %[[M:.*]] = complex.add %[[Y]], %[[X]] : complex<f32>30// CHECK: memref.store %[[M]]31// CHECK: }32// CHECK: }33// CHECK: }34func.func @matmul_complex(%a: tensor<10x20xcomplex<f32>>,35 %b: tensor<20x30xcomplex<f32>, #CSR_ones_complex>,36 %c: tensor<10x30xcomplex<f32>>) -> tensor<10x30xcomplex<f32>> {37 %0 = linalg.matmul38 ins(%a, %b: tensor<10x20xcomplex<f32>>, tensor<20x30xcomplex<f32>,#CSR_ones_complex>)39 outs(%c: tensor<10x30xcomplex<f32>>) -> tensor<10x30xcomplex<f32>>40 return %0 : tensor<10x30xcomplex<f32>>41}42 43// CHECK-LABEL: func.func @matmul_fp44// CHECK: scf.for45// CHECK: scf.for46// CHECK: %[[X:.*]] = memref.load47// CHECK: scf.for48// CHECK: %[[I:.*]] = memref.load49// CHECK: %[[Y:.*]] = memref.load50// CHECK: %[[M:.*]] = arith.addf %[[Y]], %[[X]] : f3251// CHECK: memref.store %[[M]]52// CHECK: }53// CHECK: }54// CHECK: }55func.func @matmul_fp(%a: tensor<10x20xf32>,56 %b: tensor<20x30xf32, #CSR_ones_fp>,57 %c: tensor<10x30xf32>) -> tensor<10x30xf32> {58 %0 = linalg.matmul59 ins(%a, %b: tensor<10x20xf32>, tensor<20x30xf32,#CSR_ones_fp>)60 outs(%c: tensor<10x30xf32>) -> tensor<10x30xf32>61 return %0 : tensor<10x30xf32>62}63 64// CHECK-LABEL: func.func @matmul_int65// CHECK: scf.for66// CHECK: scf.for67// CHECK: %[[X:.*]] = memref.load68// CHECK: scf.for69// CHECK: %[[I:.*]] = memref.load70// CHECK: %[[Y:.*]] = memref.load71// CHECK: %[[M:.*]] = arith.addi %[[Y]], %[[X]] : i3272// CHECK: memref.store %[[M]]73// CHECK: }74// CHECK: }75// CHECK: }76func.func @matmul_int(%a: tensor<10x20xi32>,77 %b: tensor<20x30xi32, #CSR_ones_int>,78 %c: tensor<10x30xi32>) -> tensor<10x30xi32> {79 %0 = linalg.matmul80 ins(%a, %b: tensor<10x20xi32>, tensor<20x30xi32,#CSR_ones_int>)81 outs(%c: tensor<10x30xi32>) -> tensor<10x30xi32>82 return %0 : tensor<10x30xi32>83}84