diff --git a/mlir/lib/Dialect/SparseTensor/Transforms/Sparsification.cpp b/mlir/lib/Dialect/SparseTensor/Transforms/Sparsification.cpp --- a/mlir/lib/Dialect/SparseTensor/Transforms/Sparsification.cpp +++ b/mlir/lib/Dialect/SparseTensor/Transforms/Sparsification.cpp @@ -1287,6 +1287,7 @@ const BitVector &simple = env.lat(li).simple; unsigned numloopCond = 0; + bool hasNonUnique = false; // Converts bits to array + dim pair env.merger().foreachTidDimPairInBits( all, [&, idx](unsigned b, unsigned tid, std::optional dim, @@ -1305,6 +1306,7 @@ if (!dim) return; } + hasNonUnique = !isUniqueDLT(dlt) || hasNonUnique; tids.push_back(tid); dims.push_back(*dim); numloopCond++; @@ -1367,9 +1369,9 @@ } assert(numloopCond > 0); - // If we just need to one loop conditions, the loop can be generated by a for - // loop. - return numloopCond == 1; + // If we just need to one loop conditions and the conditions is not imposed on + // non-unique level, the loop can be generated by a for loop. + return numloopCond == 1 && !hasNonUnique; } /// Starts a single loop in current sequence. diff --git a/mlir/test/Dialect/SparseTensor/sorted_coo.mlir b/mlir/test/Dialect/SparseTensor/sorted_coo.mlir --- a/mlir/test/Dialect/SparseTensor/sorted_coo.mlir +++ b/mlir/test/Dialect/SparseTensor/sorted_coo.mlir @@ -1,4 +1,4 @@ -// RUN: mlir-opt %s -sparsification | FileCheck %s +// RUN: mlir-opt %s -sparsification --canonicalize | FileCheck %s #SortedCOO = #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] @@ -36,23 +36,48 @@ // Kernels that operate on SortedCOO format. // -// CHECK-LABEL: func.func @sparse_scale( -// CHECK-SAME: %[[VAL_0:.*]]: tensor>) -> tensor> { -// CHECK-DAG: %[[VAL_1:.*]] = arith.constant 0 : index -// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 1 : index -// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 2.000000e+00 : f32 -// CHECK-DAG: %[[VAL_4:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor> to memref -// CHECK-DAG: %[[VAL_5:.*]] = sparse_tensor.values %[[VAL_0]] : tensor> to memref -// CHECK: %[[VAL_6:.*]] = memref.load %[[VAL_4]]{{\[}}%[[VAL_1]]] : memref -// CHECK: %[[VAL_7:.*]] = memref.load %[[VAL_4]]{{\[}}%[[VAL_2]]] : memref -// CHECK: scf.for %[[VAL_8:.*]] = %[[VAL_6]] to %[[VAL_7]] step %[[VAL_2]] { -// CHECK: %[[VAL_9:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_8]]] : memref -// CHECK: %[[VAL_10:.*]] = arith.mulf %[[VAL_9]], %[[VAL_3]] : f32 -// CHECK: memref.store %[[VAL_10]], %[[VAL_5]]{{\[}}%[[VAL_8]]] : memref +// CHECK-LABEL: func.func @sparse_scale( +// CHECK-SAME: %[[VAL_0:.*]]: tensor>) -> tensor> { +// CHECK-DAG: %[[VAL_1:.*]] = arith.constant false +// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 0 : index +// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 1 : index +// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 2.000000e+00 : f32 +// CHECK-DAG: %[[VAL_5:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor> to memref +// CHECK-DAG: %[[VAL_6:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor> to memref> +// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.values %[[VAL_0]] : tensor> to memref +// CHECK-DAG: %[[VAL_8:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_2]]] : memref +// CHECK-DAG: %[[VAL_9:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_3]]] : memref +// CHECK: %[[VAL_10:.*]] = scf.while (%[[VAL_11:.*]] = %[[VAL_8]]) : (index) -> index { +// CHECK: %[[VAL_12:.*]] = arith.cmpi ult, %[[VAL_11]], %[[VAL_9]] : index +// CHECK: scf.condition(%[[VAL_12]]) %[[VAL_11]] : index +// CHECK: } do { +// CHECK: ^bb0(%[[VAL_13:.*]]: index): +// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_13]]] : memref> +// CHECK: %[[VAL_15:.*]] = scf.while (%[[VAL_16:.*]] = %[[VAL_13]]) : (index) -> index { +// CHECK: %[[VAL_17:.*]] = arith.cmpi ult, %[[VAL_16]], %[[VAL_9]] : index +// CHECK: %[[VAL_18:.*]] = scf.if %[[VAL_17]] -> (i1) { +// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_16]]] : memref> +// CHECK: %[[VAL_20:.*]] = arith.cmpi eq, %[[VAL_19]], %[[VAL_14]] : index +// CHECK: scf.yield %[[VAL_20]] : i1 +// CHECK: } else { +// CHECK: scf.yield %[[VAL_1]] : i1 +// CHECK: } +// CHECK: scf.condition(%[[VAL_21:.*]]) %[[VAL_16]] : index +// CHECK: } do { +// CHECK: ^bb0(%[[VAL_22:.*]]: index): +// CHECK: %[[VAL_23:.*]] = arith.addi %[[VAL_22]], %[[VAL_3]] : index +// CHECK: scf.yield %[[VAL_23]] : index +// CHECK: } +// CHECK: scf.for %[[VAL_24:.*]] = %[[VAL_13]] to %[[VAL_25:.*]] step %[[VAL_3]] { +// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_24]]] : memref +// CHECK: %[[VAL_27:.*]] = arith.mulf %[[VAL_26]], %[[VAL_4]] : f32 +// CHECK: memref.store %[[VAL_27]], %[[VAL_7]]{{\[}}%[[VAL_24]]] : memref +// CHECK: } {"Emitted from" = "linalg.generic"} +// CHECK: scf.yield %[[VAL_28:.*]] : index +// CHECK: } attributes {"Emitted from" = "linalg.generic"} +// CHECK: %[[VAL_29:.*]] = sparse_tensor.load %[[VAL_0]] : tensor> +// CHECK: return %[[VAL_29]] : tensor> // CHECK: } -// CHECK: %[[VAL_11:.*]] = sparse_tensor.load %[[VAL_0]] : tensor> -// CHECK: return %[[VAL_11]] : tensor> -// CHECK: } func.func @sparse_scale(%argx: tensor) -> tensor { %c = arith.constant 2.0 : f32 %0 = linalg.generic #trait_scale @@ -64,33 +89,57 @@ return %0 : tensor } -// CHECK-LABEL: func.func @matvec( -// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x64xf64, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] }>>, -// CHECK-SAME: %[[VAL_1:.*]]: tensor<64xf64>, -// CHECK-SAME: %[[VAL_2:.*]]: tensor<32xf64>) -> tensor<32xf64> { -// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 0 : index -// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 1 : index -// CHECK-DAG: %[[VAL_5:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<32x64xf64, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] }>> to memref -// CHECK-DAG: %[[VAL_6:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<32x64xf64, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] }>> to memref> -// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 1 : index} : tensor<32x64xf64, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] }>> to memref> -// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x64xf64, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] }>> to memref -// CHECK-DAG: %[[VAL_9:.*]] = bufferization.to_memref %[[VAL_1]] : memref<64xf64> -// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32xf64> -// CHECK-DAG: %[[VAL_11:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_3]]] : memref -// CHECK-DAG: %[[VAL_12:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_4]]] : memref -// CHECK: scf.for %[[VAL_13:.*]] = %[[VAL_11]] to %[[VAL_12]] step %[[VAL_4]] { -// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_13]]] : memref> -// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_14]]] : memref<32xf64> -// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_13]]] : memref> -// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_13]]] : memref -// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_16]]] : memref<64xf64> -// CHECK: %[[VAL_19:.*]] = arith.mulf %[[VAL_17]], %[[VAL_18]] : f64 -// CHECK: %[[VAL_20:.*]] = arith.addf %[[VAL_15]], %[[VAL_19]] : f64 -// CHECK: memref.store %[[VAL_20]], %[[VAL_10]]{{\[}}%[[VAL_14]]] : memref<32xf64> +// CHECK-LABEL: func.func @matvec( +// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x64xf64, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] }>>, +// CHECK-SAME: %[[VAL_1:.*]]: tensor<64xf64>, +// CHECK-SAME: %[[VAL_2:.*]]: tensor<32xf64>) -> tensor<32xf64> { +// CHECK-DAG: %[[VAL_3:.*]] = arith.constant false +// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 0 : index +// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 1 : index +// CHECK-DAG: %[[VAL_6:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<32x64xf64, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] }>> to memref +// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<32x64xf64, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] }>> to memref> +// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 1 : index} : tensor<32x64xf64, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] }>> to memref> +// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x64xf64, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ] }>> to memref +// CHECK: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32xf64> +// CHECK: %[[VAL_11:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_4]]] : memref +// CHECK: %[[VAL_12:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_5]]] : memref +// CHECK: %[[VAL_13:.*]] = scf.while (%[[VAL_14:.*]] = %[[VAL_11]]) : (index) -> index { +// CHECK: %[[VAL_15:.*]] = arith.cmpi ult, %[[VAL_14]], %[[VAL_12]] : index +// CHECK: scf.condition(%[[VAL_15]]) %[[VAL_14]] : index +// CHECK: } do { +// CHECK: ^bb0(%[[VAL_16:.*]]: index): +// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_16]]] : memref> +// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_16]]] : memref> +// CHECK: %[[VAL_19:.*]] = scf.while (%[[VAL_20:.*]] = %[[VAL_16]]) : (index) -> index { +// CHECK: %[[VAL_21:.*]] = arith.cmpi ult, %[[VAL_20]], %[[VAL_12]] : index +// CHECK: %[[VAL_22:.*]] = scf.if %[[VAL_21]] -> (i1) { +// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_20]]] : memref> +// CHECK: %[[VAL_24:.*]] = arith.cmpi eq, %[[VAL_23]], %[[VAL_18]] : index +// CHECK: scf.yield %[[VAL_24]] : i1 +// CHECK: } else { +// CHECK: scf.yield %[[VAL_3]] : i1 +// CHECK: } +// CHECK: scf.condition(%[[VAL_25:.*]]) %[[VAL_20]] : index +// CHECK: } do { +// CHECK: ^bb0(%[[VAL_26:.*]]: index): +// CHECK: %[[VAL_27:.*]] = arith.addi %[[VAL_26]], %[[VAL_5]] : index +// CHECK: scf.yield %[[VAL_27]] : index +// CHECK: } +// CHECK: %[[VAL_28:.*]] = tensor.extract %[[VAL_2]]{{\[}}%[[VAL_17]]] : tensor<32xf64> +// CHECK: %[[VAL_29:.*]] = scf.for %[[VAL_30:.*]] = %[[VAL_16]] to %[[VAL_31:.*]] step %[[VAL_5]] iter_args(%[[VAL_32:.*]] = %[[VAL_28]]) -> (f64) { +// CHECK: %[[VAL_33:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_30]]] : memref> +// CHECK: %[[VAL_34:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_30]]] : memref +// CHECK: %[[VAL_35:.*]] = tensor.extract %[[VAL_1]]{{\[}}%[[VAL_33]]] : tensor<64xf64> +// CHECK: %[[VAL_36:.*]] = arith.mulf %[[VAL_34]], %[[VAL_35]] : f64 +// CHECK: %[[VAL_37:.*]] = arith.addf %[[VAL_32]], %[[VAL_36]] : f64 +// CHECK: scf.yield %[[VAL_37]] : f64 +// CHECK: } {"Emitted from" = "linalg.generic"} +// CHECK: memref.store %[[VAL_38:.*]], %[[VAL_10]]{{\[}}%[[VAL_17]]] : memref<32xf64> +// CHECK: scf.yield %[[VAL_39:.*]] : index +// CHECK: } attributes {"Emitted from" = "linalg.generic"} +// CHECK: %[[VAL_40:.*]] = bufferization.to_tensor %[[VAL_10]] : memref<32xf64> +// CHECK: return %[[VAL_40]] : tensor<32xf64> // CHECK: } -// CHECK: %[[VAL_21:.*]] = bufferization.to_tensor %[[VAL_10]] : memref<32xf64> -// CHECK: return %[[VAL_21]] : tensor<32xf64> -// CHECK: } func.func @matvec(%arga: tensor<32x64xf64, #SortedCOO>, %argb: tensor<64xf64>, %argx: tensor<32xf64>) -> tensor<32xf64> { @@ -193,7 +242,6 @@ // CHECK: %[[VAL_73:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_64]]] : memref // CHECK: %[[VAL_74:.*]] = arith.mulf %[[VAL_72]], %[[VAL_73]] : f64 // CHECK: memref.store %[[VAL_74]], %[[VAL_15]]{{\[}}%[[VAL_31]], %[[VAL_68]]] : memref<32x64xf64> -// CHECK: } else { // CHECK: } // CHECK: %[[VAL_75:.*]] = arith.cmpi eq, %[[VAL_65]], %[[VAL_68]] : index // CHECK: %[[VAL_76:.*]] = arith.addi %[[VAL_63]], %[[VAL_6]] : index @@ -203,7 +251,6 @@ // CHECK: %[[VAL_80:.*]] = arith.select %[[VAL_78]], %[[VAL_79]], %[[VAL_64]] : index // CHECK: scf.yield %[[VAL_77]], %[[VAL_80]] : index, index // CHECK: } attributes {"Emitted from" = "linalg.generic"} -// CHECK: } else { // CHECK: } // CHECK: %[[VAL_81:.*]] = arith.cmpi eq, %[[VAL_28]], %[[VAL_31]] : index // CHECK: %[[VAL_82:.*]] = arith.select %[[VAL_81]], %[[VAL_83:.*]], %[[VAL_26]] : index diff --git a/mlir/test/Dialect/SparseTensor/sparse_reshape_dot.mlir b/mlir/test/Dialect/SparseTensor/sparse_reshape_dot.mlir --- a/mlir/test/Dialect/SparseTensor/sparse_reshape_dot.mlir +++ b/mlir/test/Dialect/SparseTensor/sparse_reshape_dot.mlir @@ -3,51 +3,77 @@ #COO_2D = #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton" ], posWidth = 32, crdWidth = 32 }> #COO_3D = #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton-nu", "singleton" ], posWidth = 32, crdWidth = 32 }> + // CHECK-LABEL: func.func @sparse_reshape_fused( // CHECK-SAME: %[[VAL_0:.*]]: tensor<5x6xf32>, -// CHECK-SAME: %[[VAL_1:.*]]: tensor<6x2x3xf32, -// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 5 : index -// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 3 : index -// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 0 : index -// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 1 : index -// CHECK-DAG: %[[VAL_6:.*]] = tensor.empty() : tensor<5x6xf32> -// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.positions %[[VAL_1]] {level = 0 : index} -// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.coordinates %[[VAL_1]] {level = 0 : index} -// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.coordinates %[[VAL_1]] {level = 1 : index} -// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.coordinates %[[VAL_1]] {level = 2 : index} -// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_1]] -// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_6]] : memref<5x6xf32> -// CHECK: scf.for %[[VAL_13:.*]] = %[[VAL_4]] to %[[VAL_2]] step %[[VAL_5]] { -// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_4]]] : memref -// CHECK: %[[VAL_15:.*]] = arith.extui %[[VAL_14]] : i32 to i64 -// CHECK: %[[VAL_16:.*]] = arith.index_cast %[[VAL_15]] : i64 to index -// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_5]]] : memref -// CHECK: %[[VAL_18:.*]] = arith.extui %[[VAL_17]] : i32 to i64 -// CHECK: %[[VAL_19:.*]] = arith.index_cast %[[VAL_18]] : i64 to index -// CHECK: scf.for %[[VAL_20:.*]] = %[[VAL_16]] to %[[VAL_19]] step %[[VAL_5]] { -// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_20]]] : memref> -// CHECK: %[[VAL_22:.*]] = arith.extui %[[VAL_21]] : i32 to i64 -// CHECK: %[[VAL_23:.*]] = arith.index_cast %[[VAL_22]] : i64 to index -// CHECK: %[[VAL_24:.*]] = tensor.extract %[[VAL_0]]{{\[}}%[[VAL_13]], %[[VAL_23]]] : tensor<5x6xf32> -// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_20]]] : memref> +// CHECK-SAME: %[[VAL_1:.*]]: tensor<6x2x3xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed-nu", "singleton-nu", "singleton" ], posWidth = 32, crdWidth = 32 }>>) -> tensor { +// CHECK-DAG: %[[VAL_2:.*]] = arith.constant false +// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 5 : index +// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 3 : index +// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index +// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index +// CHECK-DAG: %[[VAL_7:.*]] = tensor.empty() : tensor<5x6xf32> +// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.positions %[[VAL_1]] {level = 0 : index} +// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.coordinates %[[VAL_1]] {level = 0 : index} +// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.coordinates %[[VAL_1]] {level = 1 : index} +// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.coordinates %[[VAL_1]] {level = 2 : index} +// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.values %[[VAL_1]] +// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_7]] : memref<5x6xf32> +// CHECK: scf.for %[[VAL_14:.*]] = %[[VAL_5]] to %[[VAL_3]] step %[[VAL_6]] { +// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_5]]] : memref +// CHECK: %[[VAL_16:.*]] = arith.extui %[[VAL_15]] : i32 to i64 +// CHECK: %[[VAL_17:.*]] = arith.index_cast %[[VAL_16]] : i64 to index +// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_6]]] : memref +// CHECK: %[[VAL_19:.*]] = arith.extui %[[VAL_18]] : i32 to i64 +// CHECK: %[[VAL_20:.*]] = arith.index_cast %[[VAL_19]] : i64 to index +// CHECK: %[[VAL_21:.*]] = scf.while (%[[VAL_22:.*]] = %[[VAL_17]]) : (index) -> index { +// CHECK: %[[VAL_23:.*]] = arith.cmpi ult, %[[VAL_22]], %[[VAL_20]] : index +// CHECK: scf.condition(%[[VAL_23]]) %[[VAL_22]] : index +// CHECK: } do { +// CHECK: ^bb0(%[[VAL_24:.*]]: index): +// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_24]]] : memref> // CHECK: %[[VAL_26:.*]] = arith.extui %[[VAL_25]] : i32 to i64 // CHECK: %[[VAL_27:.*]] = arith.index_cast %[[VAL_26]] : i64 to index -// CHECK: %[[VAL_28:.*]] = arith.muli %[[VAL_27]], %[[VAL_3]] : index -// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_20]]] : memref> -// CHECK: %[[VAL_30:.*]] = arith.extui %[[VAL_29]] : i32 to i64 -// CHECK: %[[VAL_31:.*]] = arith.index_cast %[[VAL_30]] : i64 to index -// CHECK: %[[VAL_32:.*]] = arith.addi %[[VAL_28]], %[[VAL_31]] : index -// CHECK: %[[VAL_33:.*]] = tensor.extract %[[VAL_6]]{{\[}}%[[VAL_13]], %[[VAL_32]]] : tensor<5x6xf32> -// CHECK: %[[VAL_34:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_20]]] : memref -// CHECK: %[[VAL_35:.*]] = arith.mulf %[[VAL_24]], %[[VAL_34]] : f32 -// CHECK: %[[VAL_36:.*]] = arith.addf %[[VAL_33]], %[[VAL_35]] : f32 -// CHECK: memref.store %[[VAL_36]], %[[VAL_12]]{{\[}}%[[VAL_13]], %[[VAL_32]]] : memref<5x6xf32> -// CHECK: } {"Emitted from" = "linalg.generic"} +// CHECK: %[[VAL_28:.*]] = scf.while (%[[VAL_29:.*]] = %[[VAL_24]]) : (index) -> index { +// CHECK: %[[VAL_30:.*]] = arith.cmpi ult, %[[VAL_29]], %[[VAL_20]] : index +// CHECK: %[[VAL_31:.*]] = scf.if %[[VAL_30]] -> (i1) { +// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_29]]] : memref> +// CHECK: %[[VAL_33:.*]] = arith.extui %[[VAL_32]] : i32 to i64 +// CHECK: %[[VAL_34:.*]] = arith.index_cast %[[VAL_33]] : i64 to index +// CHECK: %[[VAL_35:.*]] = arith.cmpi eq, %[[VAL_34]], %[[VAL_27]] : index +// CHECK: scf.yield %[[VAL_35]] : i1 +// CHECK: } else { +// CHECK: scf.yield %[[VAL_2]] : i1 +// CHECK: } +// CHECK: scf.condition(%[[VAL_36:.*]]) %[[VAL_29]] : index +// CHECK: } do { +// CHECK: ^bb0(%[[VAL_37:.*]]: index): +// CHECK: %[[VAL_38:.*]] = arith.addi %[[VAL_37]], %[[VAL_6]] : index +// CHECK: scf.yield %[[VAL_38]] : index +// CHECK: } +// CHECK: %[[VAL_39:.*]] = tensor.extract %[[VAL_0]]{{\[}}%[[VAL_14]], %[[VAL_27]]] : tensor<5x6xf32> +// CHECK: scf.for %[[VAL_40:.*]] = %[[VAL_24]] to %[[VAL_41:.*]] step %[[VAL_6]] { +// CHECK: %[[VAL_42:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_40]]] : memref> +// CHECK: %[[VAL_43:.*]] = arith.extui %[[VAL_42]] : i32 to i64 +// CHECK: %[[VAL_44:.*]] = arith.index_cast %[[VAL_43]] : i64 to index +// CHECK: %[[VAL_45:.*]] = arith.muli %[[VAL_44]], %[[VAL_4]] : index +// CHECK: %[[VAL_46:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_40]]] : memref> +// CHECK: %[[VAL_47:.*]] = arith.extui %[[VAL_46]] : i32 to i64 +// CHECK: %[[VAL_48:.*]] = arith.index_cast %[[VAL_47]] : i64 to index +// CHECK: %[[VAL_49:.*]] = arith.addi %[[VAL_45]], %[[VAL_48]] : index +// CHECK: %[[VAL_50:.*]] = tensor.extract %[[VAL_7]]{{\[}}%[[VAL_14]], %[[VAL_49]]] : tensor<5x6xf32> +// CHECK: %[[VAL_51:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_40]]] : memref +// CHECK: %[[VAL_52:.*]] = arith.mulf %[[VAL_39]], %[[VAL_51]] : f32 +// CHECK: %[[VAL_53:.*]] = arith.addf %[[VAL_50]], %[[VAL_52]] : f32 +// CHECK: memref.store %[[VAL_53]], %[[VAL_13]]{{\[}}%[[VAL_14]], %[[VAL_49]]] : memref<5x6xf32> +// CHECK: } {"Emitted from" = "linalg.generic"} +// CHECK: scf.yield %[[VAL_54:.*]] : index +// CHECK: } attributes {"Emitted from" = "linalg.generic"} // CHECK: } {"Emitted from" = "linalg.generic"} -// CHECK: %[[VAL_37:.*]] = bufferization.to_tensor %[[VAL_12]] : memref<5x6xf32> -// CHECK: %[[VAL_38:.*]] = tensor.expand_shape %[[VAL_37]] {{\[\[}}0], [1, 2]] : tensor<5x6xf32> into tensor<5x2x3xf32> -// CHECK: %[[VAL_39:.*]] = tensor.cast %[[VAL_38]] : tensor<5x2x3xf32> to tensor -// CHECK: return %[[VAL_39]] : tensor +// CHECK: %[[VAL_55:.*]] = bufferization.to_tensor %[[VAL_13]] : memref<5x6xf32> +// CHECK: %[[VAL_56:.*]] = tensor.expand_shape %[[VAL_55]] {{\[\[}}0], [1, 2]] : tensor<5x6xf32> into tensor<5x2x3xf32> +// CHECK: %[[VAL_57:.*]] = tensor.cast %[[VAL_56]] : tensor<5x2x3xf32> to tensor +// CHECK: return %[[VAL_57]] : tensor // CHECK: } func.func @sparse_reshape_fused(%arg0: tensor<5x6xf32>, %arg1: tensor<6x2x3xf32, #COO_3D>) -> tensor { %collapsed = tensor.collapse_shape %arg1 [[0], [1, 2]] : tensor<6x2x3xf32, #COO_3D> into tensor<6x6xf32, #COO_2D>