diff options
Diffstat (limited to 'mlir/test/Conversion')
9 files changed, 260 insertions, 404 deletions
diff --git a/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir b/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir index 1abe0fd..6e2352e 100644 --- a/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir +++ b/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir @@ -559,6 +559,23 @@ func.func @constant() { return } +// CHECK-LABEL: @constant_8bit_float +func.func @constant_8bit_float() { + // CHECK: spirv.Constant 56 : i8 + %cst = arith.constant 1.0 : f8E4M3 + // CHECK: spirv.Constant 56 : i8 + %cst_i8 = arith.bitcast %cst : f8E4M3 to i8 + // CHECK: spirv.Constant dense<56> : vector<4xi8> + %cst_vector = arith.constant dense<1.0> : vector<4xf8E4M3> + // CHECK: spirv.Constant dense<56> : vector<4xi8> + %cst_vector_i8 = arith.bitcast %cst_vector : vector<4xf8E4M3> to vector<4xi8> + // CHECK: spirv.Constant dense<60> : tensor<4xi8> : !spirv.array<4 x i8> + %cst_tensor = arith.constant dense<1.0> : tensor<4xf8E5M2> + // CHECK: spirv.Constant dense<60> : tensor<4xi8> : !spirv.array<4 x i8> + %cst_tensor_i8 = arith.bitcast %cst_tensor : tensor<4xf8E5M2> to tensor<4xi8> + return +} + // CHECK-LABEL: @constant_16bit func.func @constant_16bit() { // CHECK: spirv.Constant 4 : i16 diff --git a/mlir/test/Conversion/ComplexToROCDLLibraryCalls/complex-to-rocdl-library-calls.mlir b/mlir/test/Conversion/ComplexToROCDLLibraryCalls/complex-to-rocdl-library-calls.mlir index bae7c59..ae59f28 100644 --- a/mlir/test/Conversion/ComplexToROCDLLibraryCalls/complex-to-rocdl-library-calls.mlir +++ b/mlir/test/Conversion/ComplexToROCDLLibraryCalls/complex-to-rocdl-library-calls.mlir @@ -2,8 +2,26 @@ // CHECK-DAG: @__ocml_cabs_f32(complex<f32>) -> f32 // CHECK-DAG: @__ocml_cabs_f64(complex<f64>) -> f64 +// CHECK-DAG: @__ocml_carg_f32(complex<f32>) -> f32 +// CHECK-DAG: @__ocml_carg_f64(complex<f64>) -> f64 +// CHECK-DAG: @__ocml_ccos_f32(complex<f32>) -> complex<f32> +// CHECK-DAG: @__ocml_ccos_f64(complex<f64>) -> complex<f64> // CHECK-DAG: @__ocml_cexp_f32(complex<f32>) -> complex<f32> // CHECK-DAG: @__ocml_cexp_f64(complex<f64>) -> complex<f64> +// CHECK-DAG: @__ocml_clog_f32(complex<f32>) -> complex<f32> +// CHECK-DAG: @__ocml_clog_f64(complex<f64>) -> complex<f64> +// CHECK-DAG: @__ocml_conj_f32(complex<f32>) -> complex<f32> +// CHECK-DAG: @__ocml_conj_f64(complex<f64>) -> complex<f64> +// CHECK-DAG: @__ocml_cpow_f32(complex<f32>, complex<f32>) -> complex<f32> +// CHECK-DAG: @__ocml_cpow_f64(complex<f64>, complex<f64>) -> complex<f64> +// CHECK-DAG: @__ocml_csin_f32(complex<f32>) -> complex<f32> +// CHECK-DAG: @__ocml_csin_f64(complex<f64>) -> complex<f64> +// CHECK-DAG: @__ocml_csqrt_f32(complex<f32>) -> complex<f32> +// CHECK-DAG: @__ocml_csqrt_f64(complex<f64>) -> complex<f64> +// CHECK-DAG: @__ocml_ctan_f32(complex<f32>) -> complex<f32> +// CHECK-DAG: @__ocml_ctan_f64(complex<f64>) -> complex<f64> +// CHECK-DAG: @__ocml_ctanh_f32(complex<f32>) -> complex<f32> +// CHECK-DAG: @__ocml_ctanh_f64(complex<f64>) -> complex<f64> //CHECK-LABEL: @abs_caller func.func @abs_caller(%f: complex<f32>, %d: complex<f64>) -> (f32, f64) { @@ -15,6 +33,26 @@ func.func @abs_caller(%f: complex<f32>, %d: complex<f64>) -> (f32, f64) { return %rf, %rd : f32, f64 } +//CHECK-LABEL: @angle_caller +func.func @angle_caller(%f: complex<f32>, %d: complex<f64>) -> (f32, f64) { + // CHECK: %[[AF:.*]] = call @__ocml_carg_f32(%{{.*}}) + %af = complex.angle %f : complex<f32> + // CHECK: %[[AD:.*]] = call @__ocml_carg_f64(%{{.*}}) + %ad = complex.angle %d : complex<f64> + // CHECK: return %[[AF]], %[[AD]] + return %af, %ad : f32, f64 +} + +//CHECK-LABEL: @cos_caller +func.func @cos_caller(%f: complex<f32>, %d: complex<f64>) -> (complex<f32>, complex<f64>) { + // CHECK: %[[CF:.*]] = call @__ocml_ccos_f32(%{{.*}}) + %cf = complex.cos %f : complex<f32> + // CHECK: %[[CD:.*]] = call @__ocml_ccos_f64(%{{.*}}) + %cd = complex.cos %d : complex<f64> + // CHECK: return %[[CF]], %[[CD]] + return %cf, %cd : complex<f32>, complex<f64> +} + //CHECK-LABEL: @exp_caller func.func @exp_caller(%f: complex<f32>, %d: complex<f64>) -> (complex<f32>, complex<f64>) { // CHECK: %[[EF:.*]] = call @__ocml_cexp_f32(%{{.*}}) @@ -24,3 +62,73 @@ func.func @exp_caller(%f: complex<f32>, %d: complex<f64>) -> (complex<f32>, comp // CHECK: return %[[EF]], %[[ED]] return %ef, %ed : complex<f32>, complex<f64> } + +//CHECK-LABEL: @log_caller +func.func @log_caller(%f: complex<f32>, %d: complex<f64>) -> (complex<f32>, complex<f64>) { + // CHECK: %[[LF:.*]] = call @__ocml_clog_f32(%{{.*}}) + %lf = complex.log %f : complex<f32> + // CHECK: %[[LD:.*]] = call @__ocml_clog_f64(%{{.*}}) + %ld = complex.log %d : complex<f64> + // CHECK: return %[[LF]], %[[LD]] + return %lf, %ld : complex<f32>, complex<f64> +} + +//CHECK-LABEL: @conj_caller +func.func @conj_caller(%f: complex<f32>, %d: complex<f64>) -> (complex<f32>, complex<f64>) { + // CHECK: %[[CF:.*]] = call @__ocml_conj_f32(%{{.*}}) + %cf2 = complex.conj %f : complex<f32> + // CHECK: %[[CD:.*]] = call @__ocml_conj_f64(%{{.*}}) + %cd2 = complex.conj %d : complex<f64> + // CHECK: return %[[CF]], %[[CD]] + return %cf2, %cd2 : complex<f32>, complex<f64> +} + +//CHECK-LABEL: @pow_caller +func.func @pow_caller(%f: complex<f32>, %d: complex<f64>) -> (complex<f32>, complex<f64>) { + // CHECK: %[[PF:.*]] = call @__ocml_cpow_f32(%{{.*}}, %{{.*}}) + %pf = complex.pow %f, %f : complex<f32> + // CHECK: %[[PD:.*]] = call @__ocml_cpow_f64(%{{.*}}, %{{.*}}) + %pd = complex.pow %d, %d : complex<f64> + // CHECK: return %[[PF]], %[[PD]] + return %pf, %pd : complex<f32>, complex<f64> +} + +//CHECK-LABEL: @sin_caller +func.func @sin_caller(%f: complex<f32>, %d: complex<f64>) -> (complex<f32>, complex<f64>) { + // CHECK: %[[SF:.*]] = call @__ocml_csin_f32(%{{.*}}) + %sf2 = complex.sin %f : complex<f32> + // CHECK: %[[SD:.*]] = call @__ocml_csin_f64(%{{.*}}) + %sd2 = complex.sin %d : complex<f64> + // CHECK: return %[[SF]], %[[SD]] + return %sf2, %sd2 : complex<f32>, complex<f64> +} + +//CHECK-LABEL: @sqrt_caller +func.func @sqrt_caller(%f: complex<f32>, %d: complex<f64>) -> (complex<f32>, complex<f64>) { + // CHECK: %[[SF:.*]] = call @__ocml_csqrt_f32(%{{.*}}) + %sf = complex.sqrt %f : complex<f32> + // CHECK: %[[SD:.*]] = call @__ocml_csqrt_f64(%{{.*}}) + %sd = complex.sqrt %d : complex<f64> + // CHECK: return %[[SF]], %[[SD]] + return %sf, %sd : complex<f32>, complex<f64> +} + +//CHECK-LABEL: @tan_caller +func.func @tan_caller(%f: complex<f32>, %d: complex<f64>) -> (complex<f32>, complex<f64>) { + // CHECK: %[[TF:.*]] = call @__ocml_ctan_f32(%{{.*}}) + %tf2 = complex.tan %f : complex<f32> + // CHECK: %[[TD:.*]] = call @__ocml_ctan_f64(%{{.*}}) + %td2 = complex.tan %d : complex<f64> + // CHECK: return %[[TF]], %[[TD]] + return %tf2, %td2 : complex<f32>, complex<f64> +} + +//CHECK-LABEL: @tanh_caller +func.func @tanh_caller(%f: complex<f32>, %d: complex<f64>) -> (complex<f32>, complex<f64>) { + // CHECK: %[[TF:.*]] = call @__ocml_ctanh_f32(%{{.*}}) + %tf = complex.tanh %f : complex<f32> + // CHECK: %[[TD:.*]] = call @__ocml_ctanh_f64(%{{.*}}) + %td = complex.tanh %d : complex<f64> + // CHECK: return %[[TF]], %[[TD]] + return %tf, %td : complex<f32>, complex<f64> +} diff --git a/mlir/test/Conversion/ConvertToSPIRV/convert-gpu-modules.mlir b/mlir/test/Conversion/ConvertToSPIRV/convert-gpu-modules.mlir index 00bbd1c..96ad107 100644 --- a/mlir/test/Conversion/ConvertToSPIRV/convert-gpu-modules.mlir +++ b/mlir/test/Conversion/ConvertToSPIRV/convert-gpu-modules.mlir @@ -85,11 +85,10 @@ module attributes { // CHECK: spirv.Load "StorageBuffer" %val = memref.load %arg0[%idx0] : memref<2xi32> // CHECK: spirv.CompositeInsert - %vec = vector.insertelement %val, %vec0[%idx0 : index] : vector<2xi32> + %vec = vector.insert %val, %vec0[%idx0] : i32 into vector<2xi32> // CHECK: spirv.VectorShuffle %shuffle = vector.shuffle %vec, %vec[3, 2, 1, 0] : vector<2xi32>, vector<2xi32> - // CHECK: spirv.CompositeExtract - %res = vector.extractelement %shuffle[%idx0 : index] : vector<4xi32> + %res = vector.extract %shuffle[%idx0] : i32 from vector<4xi32> // CHECK: spirv.AccessChain // CHECK: spirv.Store "StorageBuffer" memref.store %res, %arg1[%idx0]: memref<4xi32> @@ -102,9 +101,9 @@ module attributes { // CHECK-SAME: %{{.*}}: memref<2xi32>, %{{.*}}: memref<4xi32> // CHECK: arith.constant // CHECK: memref.load - // CHECK: vector.insertelement + // CHECK: vector.insert // CHECK: vector.shuffle - // CHECK: vector.extractelement + // CHECK: vector.extract // CHECK: memref.store // CHECK: gpu.return } diff --git a/mlir/test/Conversion/ConvertToSPIRV/vector.mlir b/mlir/test/Conversion/ConvertToSPIRV/vector.mlir index fb14feb..eb9feaa 100644 --- a/mlir/test/Conversion/ConvertToSPIRV/vector.mlir +++ b/mlir/test/Conversion/ConvertToSPIRV/vector.mlir @@ -51,108 +51,6 @@ func.func @insert_size1_vector(%arg0 : vector<1xf32>, %arg1: f32) -> vector<1xf3 // ----- -// CHECK-LABEL: @extract_element -// CHECK-SAME: %[[V:.*]]: vector<4xf32>, %[[ID:.*]]: i32 -// CHECK: spirv.VectorExtractDynamic %[[V]][%[[ID]]] : vector<4xf32>, i32 -func.func @extract_element(%arg0 : vector<4xf32>, %id : i32) -> f32 { - %0 = vector.extractelement %arg0[%id : i32] : vector<4xf32> - return %0: f32 -} - -// ----- - -// CHECK-LABEL: @extract_element_cst -// CHECK-SAME: %[[V:.*]]: vector<4xf32> -// CHECK: spirv.CompositeExtract %[[V]][1 : i32] : vector<4xf32> -func.func @extract_element_cst(%arg0 : vector<4xf32>) -> f32 { - %idx = arith.constant 1 : i32 - %0 = vector.extractelement %arg0[%idx : i32] : vector<4xf32> - return %0: f32 -} - -// ----- - -// CHECK-LABEL: @extract_element_index -func.func @extract_element_index(%arg0 : vector<4xf32>, %id : index) -> f32 { - // CHECK: spirv.VectorExtractDynamic - %0 = vector.extractelement %arg0[%id : index] : vector<4xf32> - return %0: f32 -} - -// ----- - -// CHECK-LABEL: @extract_element_size1_vector -// CHECK-SAME:(%[[S:.+]]: f32, -func.func @extract_element_size1_vector(%arg0 : f32, %i: index) -> f32 { - %bcast = vector.broadcast %arg0 : f32 to vector<1xf32> - %0 = vector.extractelement %bcast[%i : index] : vector<1xf32> - // CHECK: spirv.ReturnValue %[[S]] - return %0: f32 -} - -// ----- - -// CHECK-LABEL: @extract_element_0d_vector -// CHECK-SAME: (%[[S:.+]]: f32) -func.func @extract_element_0d_vector(%arg0 : f32) -> f32 { - %bcast = vector.broadcast %arg0 : f32 to vector<f32> - %0 = vector.extractelement %bcast[] : vector<f32> - // CHECK: spirv.ReturnValue %[[S]] - return %0: f32 -} - -// ----- - -// CHECK-LABEL: @insert_element -// CHECK-SAME: %[[VAL:.*]]: f32, %[[V:.*]]: vector<4xf32>, %[[ID:.*]]: i32 -// CHECK: spirv.VectorInsertDynamic %[[VAL]], %[[V]][%[[ID]]] : vector<4xf32>, i32 -func.func @insert_element(%val: f32, %arg0 : vector<4xf32>, %id : i32) -> vector<4xf32> { - %0 = vector.insertelement %val, %arg0[%id : i32] : vector<4xf32> - return %0: vector<4xf32> -} - -// ----- - -// CHECK-LABEL: @insert_element_cst -// CHECK-SAME: %[[VAL:.*]]: f32, %[[V:.*]]: vector<4xf32> -// CHECK: spirv.CompositeInsert %[[VAL]], %[[V]][2 : i32] : f32 into vector<4xf32> -func.func @insert_element_cst(%val: f32, %arg0 : vector<4xf32>) -> vector<4xf32> { - %idx = arith.constant 2 : i32 - %0 = vector.insertelement %val, %arg0[%idx : i32] : vector<4xf32> - return %0: vector<4xf32> -} - -// ----- - -// CHECK-LABEL: @insert_element_index -func.func @insert_element_index(%val: f32, %arg0 : vector<4xf32>, %id : index) -> vector<4xf32> { - // CHECK: spirv.VectorInsertDynamic - %0 = vector.insertelement %val, %arg0[%id : index] : vector<4xf32> - return %0: vector<4xf32> -} - -// ----- - -// CHECK-LABEL: @insert_element_size1_vector -// CHECK-SAME: (%[[S:[a-z0-9]+]]: f32 -func.func @insert_element_size1_vector(%scalar: f32, %vector : vector<1xf32>, %i: index) -> vector<1xf32> { - %0 = vector.insertelement %scalar, %vector[%i : index] : vector<1xf32> - // CHECK: spirv.ReturnValue %[[S]] - return %0: vector<1xf32> -} - -// ----- - -// CHECK-LABEL: @insert_element_0d_vector -// CHECK-SAME: (%[[S:[a-z0-9]+]]: f32 -func.func @insert_element_0d_vector(%scalar: f32, %vector : vector<f32>) -> vector<f32> { - %0 = vector.insertelement %scalar, %vector[] : vector<f32> - // CHECK: spirv.ReturnValue %[[S]] - return %0: vector<f32> -} - -// ----- - // CHECK-LABEL: @insert_size1_vector // CHECK-SAME: %[[SUB:.*]]: f32, %[[FULL:.*]]: vector<3xf32> // CHECK: %[[RET:.*]] = spirv.CompositeInsert %[[SUB]], %[[FULL]][2 : i32] : f32 into vector<3xf32> diff --git a/mlir/test/Conversion/FuncToSPIRV/types-to-spirv.mlir b/mlir/test/Conversion/FuncToSPIRV/types-to-spirv.mlir index 1737f4a..0c77c88 100644 --- a/mlir/test/Conversion/FuncToSPIRV/types-to-spirv.mlir +++ b/mlir/test/Conversion/FuncToSPIRV/types-to-spirv.mlir @@ -1,6 +1,8 @@ // RUN: mlir-opt -split-input-file -convert-func-to-spirv %s -o - | FileCheck %s // RUN: mlir-opt -split-input-file -convert-func-to-spirv="emulate-lt-32-bit-scalar-types=false" %s | \ // RUN: FileCheck %s --check-prefix=NOEMU +// RUN: mlir-opt -split-input-file -convert-func-to-spirv="emulate-unsupported-float-types=false" %s | \ +// RUN: FileCheck %s --check-prefix=UNSUPPORTED_FLOAT //===----------------------------------------------------------------------===// // Integer types @@ -944,3 +946,55 @@ func.func @unranked_tensor(%arg0: tensor<*xi32>) { return } func.func @dynamic_dim_tensor(%arg0: tensor<8x?xi32>) { return } } // end module + + +// ----- + +// Check that 8-bit float types are emulated as i8. +module attributes { + spirv.target_env = #spirv.target_env<#spirv.vce<v1.0, [Int8], []>, #spirv.resource_limits<>> +} { + + // CHECK: spirv.func @float8_to_integer8 + // CHECK-SAME: (%arg0: i8 + // CHECK-SAME: %arg1: i8 + // CHECK-SAME: %arg2: i8 + // CHECK-SAME: %arg3: i8 + // CHECK-SAME: %arg4: i8 + // CHECK-SAME: %arg5: i8 + // CHECK-SAME: %arg6: i8 + // CHECK-SAME: %arg7: i8 + // CHECK-SAME: %arg8: vector<4xi8> + // CHECK-SAME: %arg9: !spirv.ptr<!spirv.struct<(!spirv.array<8 x i8, stride=1> [0])>, StorageBuffer> + // CHECK-SAME: %arg10: !spirv.array<4 x i8> + // UNSUPPORTED_FLOAT-LABEL: func.func @float8_to_integer8 + // UNSUPPORTED_FLOAT-SAME: (%arg0: f8E5M2 + // UNSUPPORTED_FLOAT-SAME: %arg1: f8E4M3 + // UNSUPPORTED_FLOAT-SAME: %arg2: f8E4M3FN + // UNSUPPORTED_FLOAT-SAME: %arg3: f8E5M2FNUZ + // UNSUPPORTED_FLOAT-SAME: %arg4: f8E4M3FNUZ + // UNSUPPORTED_FLOAT-SAME: %arg5: f8E4M3B11FNUZ + // UNSUPPORTED_FLOAT-SAME: %arg6: f8E3M4 + // UNSUPPORTED_FLOAT-SAME: %arg7: f8E8M0FNU + // UNSUPPORTED_FLOAT-SAME: %arg8: vector<4xf8E4M3B11FNUZ> + // UNSUPPORTED_FLOAT-SAME: %arg9: memref<8xf8E4M3, #spirv.storage_class<StorageBuffer>> + // UNSUPPORTED_FLOAT-SAME: %arg10: tensor<4xf8E5M2> + // UNSUPPORTED_FLOAT-SAME: ) { + + func.func @float8_to_integer8( + %arg0: f8E5M2, // CHECK-NOT: f8E5M2 + %arg1: f8E4M3, // CHECK-NOT: f8E4M3 + %arg2: f8E4M3FN, // CHECK-NOT: f8E4M3FN + %arg3: f8E5M2FNUZ, // CHECK-NOT: f8E5M2FNUZ + %arg4: f8E4M3FNUZ, // CHECK-NOT: f8E4M3FNUZ + %arg5: f8E4M3B11FNUZ, // CHECK-NOT: f8E4M3B11FNUZ + %arg6: f8E3M4, // CHECK-NOT: f8E3M4 + %arg7: f8E8M0FNU, // CHECK-NOT: f8E8M0FNU + %arg8: vector<4xf8E4M3B11FNUZ>, // CHECK-NOT: vector<4xf8E4M3B11FNUZ> + %arg9: memref<8xf8E4M3, #spirv.storage_class<StorageBuffer>>, // CHECK-NOT: memref + %arg10: tensor<4xf8E5M2> // CHECK-NOT: tensor + ) { + // CHECK: spirv.Return + return + } +} diff --git a/mlir/test/Conversion/GPUToSPIRV/rotate.mlir b/mlir/test/Conversion/GPUToSPIRV/rotate.mlir index b96dd37..c71d220 100644 --- a/mlir/test/Conversion/GPUToSPIRV/rotate.mlir +++ b/mlir/test/Conversion/GPUToSPIRV/rotate.mlir @@ -10,16 +10,14 @@ gpu.module @kernels { // CHECK-LABEL: spirv.func @rotate() gpu.func @rotate() kernel attributes {spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [16, 1, 1]>} { - %offset = arith.constant 4 : i32 - %width = arith.constant 16 : i32 %val = arith.constant 42.0 : f32 + // CHECK: %[[VAL:.+]] = spirv.Constant 4.200000e+01 : f32 // CHECK: %[[OFFSET:.+]] = spirv.Constant 4 : i32 // CHECK: %[[WIDTH:.+]] = spirv.Constant 16 : i32 - // CHECK: %[[VAL:.+]] = spirv.Constant 4.200000e+01 : f32 // CHECK: %{{.+}} = spirv.GroupNonUniformRotateKHR <Subgroup> %[[VAL]], %[[OFFSET]], cluster_size(%[[WIDTH]]) : f32, i32, i32 -> f32 // CHECK: %{{.+}} = spirv.Constant true - %result, %valid = gpu.rotate %val, %offset, %width : f32 + %result, %valid = gpu.rotate %val, 4, 16 : f32 gpu.return } } @@ -38,18 +36,16 @@ gpu.module @kernels { // CHECK-LABEL: spirv.func @rotate_width_less_than_subgroup_size() gpu.func @rotate_width_less_than_subgroup_size() kernel attributes {spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [16, 1, 1]>} { - %offset = arith.constant 4 : i32 - %width = arith.constant 8 : i32 %val = arith.constant 42.0 : f32 + // CHECK: %[[VAL:.+]] = spirv.Constant 4.200000e+01 : f32 // CHECK: %[[OFFSET:.+]] = spirv.Constant 4 : i32 // CHECK: %[[WIDTH:.+]] = spirv.Constant 8 : i32 - // CHECK: %[[VAL:.+]] = spirv.Constant 4.200000e+01 : f32 // CHECK: %{{.+}} = spirv.GroupNonUniformRotateKHR <Subgroup> %[[VAL]], %[[OFFSET]], cluster_size(%[[WIDTH]]) : f32, i32, i32 -> f32 // CHECK: %[[INVOCATION_ID_ADDR:.+]] = spirv.mlir.addressof @__builtin__SubgroupLocalInvocationId__ // CHECK: %[[INVOCATION_ID:.+]] = spirv.Load "Input" %[[INVOCATION_ID_ADDR]] // CHECK: %{{.+}} = spirv.ULessThan %[[INVOCATION_ID]], %[[WIDTH]] - %result, %valid = gpu.rotate %val, %offset, %width : f32 + %result, %valid = gpu.rotate %val, 4, 8 : f32 gpu.return } } @@ -67,34 +63,10 @@ module attributes { gpu.module @kernels { gpu.func @rotate_with_bigger_than_subgroup_size() kernel attributes {spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [16, 1, 1]>} { - %offset = arith.constant 4 : i32 - %width = arith.constant 32 : i32 %val = arith.constant 42.0 : f32 // expected-error @+1 {{failed to legalize operation 'gpu.rotate'}} - %result, %valid = gpu.rotate %val, %offset, %width : f32 - gpu.return - } -} - -} - -// ----- - -module attributes { - gpu.container_module, - spirv.target_env = #spirv.target_env<#spirv.vce<v1.4, [Shader, GroupNonUniformRotateKHR], []>, - #spirv.resource_limits<subgroup_size = 16>> -} { - -gpu.module @kernels { - gpu.func @rotate_non_const_width(%width: i32) kernel - attributes {spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [16, 1, 1]>} { - %offset = arith.constant 4 : i32 - %val = arith.constant 42.0 : f32 - - // expected-error @+1 {{'gpu.rotate' op width is not a constant value}} - %result, %valid = gpu.rotate %val, %offset, %width : f32 + %result, %valid = gpu.rotate %val, 4, 32 : f32 gpu.return } } diff --git a/mlir/test/Conversion/MemRefToEmitC/memref-to-emitc-alloc.mlir b/mlir/test/Conversion/MemRefToEmitC/memref-to-emitc-alloc.mlir new file mode 100644 index 0000000..e391a89 --- /dev/null +++ b/mlir/test/Conversion/MemRefToEmitC/memref-to-emitc-alloc.mlir @@ -0,0 +1,72 @@ +// RUN: mlir-opt -convert-memref-to-emitc="lower-to-cpp=true" %s -split-input-file | FileCheck %s --check-prefix=CPP +// RUN: mlir-opt -convert-memref-to-emitc="lower-to-cpp=false" %s -split-input-file | FileCheck %s --check-prefix=NOCPP + +func.func @alloc() { + %alloc = memref.alloc() : memref<999xi32> + return +} + +// CPP: module { +// CPP-NEXT: emitc.include <"cstdlib"> +// CPP-LABEL: alloc() +// CPP-NEXT: %[[ALLOC:.*]] = emitc.call_opaque "sizeof"() {args = [i32]} : () -> !emitc.size_t +// CPP-NEXT: %[[ALLOC_SIZE:.*]] = "emitc.constant"() <{value = 999 : index}> : () -> index +// CPP-NEXT: %[[ALLOC_TOTAL_SIZE:.*]] = emitc.mul %[[ALLOC]], %[[ALLOC_SIZE]] : (!emitc.size_t, index) -> !emitc.size_t +// CPP-NEXT: %[[ALLOC_PTR:.*]] = emitc.call_opaque "malloc"(%[[ALLOC_TOTAL_SIZE]]) : (!emitc.size_t) -> !emitc.ptr<!emitc.opaque<"void">> +// CPP-NEXT: %[[ALLOC_CAST:.*]] = emitc.cast %[[ALLOC_PTR]] : !emitc.ptr<!emitc.opaque<"void">> to !emitc.ptr<i32> +// CPP-NEXT: return + +// NOCPP: module { +// NOCPP-NEXT: emitc.include <"stdlib.h"> +// NOCPP-LABEL: alloc() +// NOCPP-NEXT: %[[ALLOC:.*]] = emitc.call_opaque "sizeof"() {args = [i32]} : () -> !emitc.size_t +// NOCPP-NEXT: %[[ALLOC_SIZE:.*]] = "emitc.constant"() <{value = 999 : index}> : () -> index +// NOCPP-NEXT: %[[ALLOC_TOTAL_SIZE:.*]] = emitc.mul %[[ALLOC]], %[[ALLOC_SIZE]] : (!emitc.size_t, index) -> !emitc.size_t +// NOCPP-NEXT: %[[ALLOC_PTR:.*]] = emitc.call_opaque "malloc"(%[[ALLOC_TOTAL_SIZE]]) : (!emitc.size_t) -> !emitc.ptr<!emitc.opaque<"void">> +// NOCPP-NEXT: %[[ALLOC_CAST:.*]] = emitc.cast %[[ALLOC_PTR]] : !emitc.ptr<!emitc.opaque<"void">> to !emitc.ptr<i32> +// NOCPP-NEXT: return + +func.func @alloc_aligned() { + %alloc = memref.alloc() {alignment = 64 : i64} : memref<999xf32> + return +} + +// CPP-LABEL: alloc_aligned +// CPP-NEXT: %[[ALLOC:.*]] = emitc.call_opaque "sizeof"() {args = [f32]} : () -> !emitc.size_t +// CPP-NEXT: %[[ALLOC_SIZE:.*]] = "emitc.constant"() <{value = 999 : index}> : () -> index +// CPP-NEXT: %[[ALLOC_TOTAL_SIZE:.*]] = emitc.mul %[[ALLOC]], %[[ALLOC_SIZE]] : (!emitc.size_t, index) -> !emitc.size_t +// CPP-NEXT: %[[ALIGNMENT:.*]] = "emitc.constant"() <{value = 64 : index}> : () -> !emitc.size_t +// CPP-NEXT: %[[ALLOC_PTR:.*]] = emitc.call_opaque "aligned_alloc"(%[[ALIGNMENT]], %[[ALLOC_TOTAL_SIZE]]) : (!emitc.size_t, !emitc.size_t) -> !emitc.ptr<!emitc.opaque<"void">> +// CPP-NEXT: %[[ALLOC_CAST:.*]] = emitc.cast %[[ALLOC_PTR]] : !emitc.ptr<!emitc.opaque<"void">> to !emitc.ptr<f32> +// CPP-NEXT: return + +// NOCPP-LABEL: alloc_aligned +// NOCPP-NEXT: %[[ALLOC:.*]] = emitc.call_opaque "sizeof"() {args = [f32]} : () -> !emitc.size_t +// NOCPP-NEXT: %[[ALLOC_SIZE:.*]] = "emitc.constant"() <{value = 999 : index}> : () -> index +// NOCPP-NEXT: %[[ALLOC_TOTAL_SIZE:.*]] = emitc.mul %[[ALLOC]], %[[ALLOC_SIZE]] : (!emitc.size_t, index) -> !emitc.size_t +// NOCPP-NEXT: %[[ALIGNMENT:.*]] = "emitc.constant"() <{value = 64 : index}> : () -> !emitc.size_t +// NOCPP-NEXT: %[[ALLOC_PTR:.*]] = emitc.call_opaque "aligned_alloc"(%[[ALIGNMENT]], %[[ALLOC_TOTAL_SIZE]]) : (!emitc.size_t, !emitc.size_t) -> !emitc.ptr<!emitc.opaque<"void">> +// NOCPP-NEXT: %[[ALLOC_CAST:.*]] = emitc.cast %[[ALLOC_PTR]] : !emitc.ptr<!emitc.opaque<"void">> to !emitc.ptr<f32> +// NOCPP-NEXT: return + +func.func @allocating_multi() { + %alloc_5 = memref.alloc() : memref<7x999xi32> + return +} + +// CPP-LABEL: allocating_multi +// CPP-NEXT: %[[ALLOC:.*]] = emitc.call_opaque "sizeof"() {args = [i32]} : () -> !emitc.size_t +// CPP-NEXT: %[[ALLOC_SIZE:.*]] = "emitc.constant"() <{value = 6993 : index}> : () -> index +// CPP-NEXT: %[[ALLOC_TOTAL_SIZE:.*]] = emitc.mul %[[ALLOC]], %[[ALLOC_SIZE]] : (!emitc.size_t, index) -> !emitc.size_t +// CPP-NEXT: %[[ALLOC_PTR:.*]] = emitc.call_opaque "malloc"(%[[ALLOC_TOTAL_SIZE]]) : (!emitc.size_t) -> !emitc.ptr<!emitc.opaque<"void"> +// CPP-NEXT: %[[ALLOC_CAST:.*]] = emitc.cast %[[ALLOC_PTR]] : !emitc.ptr<!emitc.opaque<"void">> to !emitc.ptr<i32> +// CPP-NEXT: return + +// NOCPP-LABEL: allocating_multi +// NOCPP-NEXT: %[[ALLOC:.*]] = emitc.call_opaque "sizeof"() {args = [i32]} : () -> !emitc.size_t +// NOCPP-NEXT: %[[ALLOC_SIZE:.*]] = "emitc.constant"() <{value = 6993 : index}> : () -> index +// NOCPP-NEXT: %[[ALLOC_TOTAL_SIZE:.*]] = emitc.mul %[[ALLOC]], %[[ALLOC_SIZE]] : (!emitc.size_t, index) -> !emitc.size_t +// NOCPP-NEXT: %[[ALLOC_PTR:.*]] = emitc.call_opaque "malloc"(%[[ALLOC_TOTAL_SIZE]]) : (!emitc.size_t) -> !emitc.ptr<!emitc.opaque<"void">> +// NOCPP-NEXT: %[[ALLOC_CAST:.*]] = emitc.cast %[[ALLOC_PTR]] : !emitc.ptr<!emitc.opaque<"void">> to !emitc.ptr<i32> +// NOCPP-NEXT: return + diff --git a/mlir/test/Conversion/VectorToLLVM/vector-to-llvm-interface.mlir b/mlir/test/Conversion/VectorToLLVM/vector-to-llvm-interface.mlir index 8c135d5..31e17fb 100644 --- a/mlir/test/Conversion/VectorToLLVM/vector-to-llvm-interface.mlir +++ b/mlir/test/Conversion/VectorToLLVM/vector-to-llvm-interface.mlir @@ -274,73 +274,6 @@ func.func @shuffle_2D(%a: vector<1x4xf32>, %b: vector<2x4xf32>) -> vector<3x4xf3 // ----- //===----------------------------------------------------------------------===// -// vector.extractelement -//===----------------------------------------------------------------------===// - -func.func @extractelement_from_vec_0d_f32(%arg0: vector<f32>) -> f32 { - %1 = vector.extractelement %arg0[] : vector<f32> - return %1 : f32 -} -// CHECK-LABEL: @extractelement_from_vec_0d_f32 -// CHECK: %[[C0:.*]] = llvm.mlir.constant(0 : index) : i64 -// CHECK: llvm.extractelement %{{.*}}[%[[C0]] : {{.*}}] : vector<1xf32> - -// ----- - -func.func @extractelement_from_vec_1d_f32_idx_as_i32(%arg0: vector<16xf32>) -> f32 { - %0 = arith.constant 15 : i32 - %1 = vector.extractelement %arg0[%0 : i32]: vector<16xf32> - return %1 : f32 -} -// CHECK-LABEL: @extractelement_from_vec_1d_f32_idx_as_i32( -// CHECK-SAME: %[[A:.*]]: vector<16xf32>) -// CHECK: %[[C:.*]] = arith.constant 15 : i32 -// CHECK: %[[X:.*]] = llvm.extractelement %[[A]][%[[C]] : i32] : vector<16xf32> -// CHECK: return %[[X]] : f32 - -// ----- - -func.func @extractelement_from_vec_1d_f32_idx_as_i32_scalable(%arg0: vector<[16]xf32>) -> f32 { - %0 = arith.constant 15 : i32 - %1 = vector.extractelement %arg0[%0 : i32]: vector<[16]xf32> - return %1 : f32 -} -// CHECK-LABEL: @extractelement_from_vec_1d_f32_idx_as_i32_scalable( -// CHECK-SAME: %[[A:.*]]: vector<[16]xf32>) -// CHECK: %[[C:.*]] = arith.constant 15 : i32 -// CHECK: %[[X:.*]] = llvm.extractelement %[[A]][%[[C]] : i32] : vector<[16]xf32> -// CHECK: return %[[X]] : f32 - -// ----- -func.func @extractelement_from_vec_1d_f32_idx_as_index(%arg0: vector<16xf32>) -> f32 { - %0 = arith.constant 15 : index - %1 = vector.extractelement %arg0[%0 : index]: vector<16xf32> - return %1 : f32 -} -// CHECK-LABEL: @extractelement_from_vec_1d_f32_idx_as_index( -// CHECK-SAME: %[[A:.*]]: vector<16xf32>) -// CHECK: %[[C:.*]] = arith.constant 15 : index -// CHECK: %[[I:.*]] = builtin.unrealized_conversion_cast %[[C]] : index to i64 -// CHECK: %[[X:.*]] = llvm.extractelement %[[A]][%[[I]] : i64] : vector<16xf32> -// CHECK: return %[[X]] : f32 - -// ----- - -func.func @extractelement_from_vec_1d_f32_idx_as_index_scalable(%arg0: vector<[16]xf32>) -> f32 { - %0 = arith.constant 15 : index - %1 = vector.extractelement %arg0[%0 : index]: vector<[16]xf32> - return %1 : f32 -} -// CHECK-LABEL: @extractelement_from_vec_1d_f32_idx_as_index_scalable( -// CHECK-SAME: %[[A:.*]]: vector<[16]xf32>) -// CHECK: %[[C:.*]] = arith.constant 15 : index -// CHECK: %[[I:.*]] = builtin.unrealized_conversion_cast %[[C]] : index to i64 -// CHECK: %[[X:.*]] = llvm.extractelement %[[A]][%[[I]] : i64] : vector<[16]xf32> -// CHECK: return %[[X]] : f32 - -// ----- - -//===----------------------------------------------------------------------===// // vector.extract //===----------------------------------------------------------------------===// @@ -592,81 +525,6 @@ func.func @extract_scalar_from_vec_2d_f32_dynamic_idxs_compile_time_const(%arg : // ----- //===----------------------------------------------------------------------===// -// vector.insertelement -//===----------------------------------------------------------------------===// - -func.func @insertelement_into_vec_0d_f32(%arg0: f32, %arg1: vector<f32>) -> vector<f32> { - %1 = vector.insertelement %arg0, %arg1[] : vector<f32> - return %1 : vector<f32> -} -// CHECK-LABEL: @insertelement_into_vec_0d_f32 -// CHECK-SAME: %[[A:.*]]: f32, -// CHECK: %[[B:.*]] = builtin.unrealized_conversion_cast %{{.*}} : -// CHECK: vector<f32> to vector<1xf32> -// CHECK: %[[C0:.*]] = llvm.mlir.constant(0 : index) : i64 -// CHECK: %[[X:.*]] = llvm.insertelement %[[A]], %[[B]][%[[C0]] : {{.*}}] : vector<1xf32> - -// ----- - -func.func @insertelement_into_vec_1d_f32_idx_as_i32(%arg0: f32, %arg1: vector<4xf32>) -> vector<4xf32> { - %0 = arith.constant 3 : i32 - %1 = vector.insertelement %arg0, %arg1[%0 : i32] : vector<4xf32> - return %1 : vector<4xf32> -} -// CHECK-LABEL: @insertelement_into_vec_1d_f32_idx_as_i32( -// CHECK-SAME: %[[A:.*]]: f32, -// CHECK-SAME: %[[B:.*]]: vector<4xf32>) -// CHECK: %[[C:.*]] = arith.constant 3 : i32 -// CHECK: %[[X:.*]] = llvm.insertelement %[[A]], %[[B]][%[[C]] : i32] : vector<4xf32> -// CHECK: return %[[X]] : vector<4xf32> - -// ----- - -func.func @insertelement_into_vec_1d_f32_idx_as_i32_scalable(%arg0: f32, %arg1: vector<[4]xf32>) -> vector<[4]xf32> { - %0 = arith.constant 3 : i32 - %1 = vector.insertelement %arg0, %arg1[%0 : i32] : vector<[4]xf32> - return %1 : vector<[4]xf32> -} -// CHECK-LABEL: @insertelement_into_vec_1d_f32_idx_as_i32_scalable( -// CHECK-SAME: %[[A:.*]]: f32, -// CHECK-SAME: %[[B:.*]]: vector<[4]xf32>) -// CHECK: %[[C:.*]] = arith.constant 3 : i32 -// CHECK: %[[X:.*]] = llvm.insertelement %[[A]], %[[B]][%[[C]] : i32] : vector<[4]xf32> -// CHECK: return %[[X]] : vector<[4]xf32> - -// ----- - -func.func @insertelement_into_vec_1d_f32_scalable_idx_as_index(%arg0: f32, %arg1: vector<4xf32>) -> vector<4xf32> { - %0 = arith.constant 3 : index - %1 = vector.insertelement %arg0, %arg1[%0 : index] : vector<4xf32> - return %1 : vector<4xf32> -} -// CHECK-LABEL: @insertelement_into_vec_1d_f32_scalable_idx_as_index( -// CHECK-SAME: %[[A:.*]]: f32, -// CHECK-SAME: %[[B:.*]]: vector<4xf32>) -// CHECK: %[[C:.*]] = arith.constant 3 : index -// CHECK: %[[I:.*]] = builtin.unrealized_conversion_cast %[[C]] : index to i64 -// CHECK: %[[X:.*]] = llvm.insertelement %[[A]], %[[B]][%[[I]] : i64] : vector<4xf32> -// CHECK: return %[[X]] : vector<4xf32> - -// ----- - -func.func @insertelement_into_vec_1d_f32_scalable_idx_as_index_scalable(%arg0: f32, %arg1: vector<[4]xf32>) -> vector<[4]xf32> { - %0 = arith.constant 3 : index - %1 = vector.insertelement %arg0, %arg1[%0 : index] : vector<[4]xf32> - return %1 : vector<[4]xf32> -} -// CHECK-LABEL: @insertelement_into_vec_1d_f32_scalable_idx_as_index_scalable( -// CHECK-SAME: %[[A:.*]]: f32, -// CHECK-SAME: %[[B:.*]]: vector<[4]xf32>) -// CHECK: %[[C:.*]] = arith.constant 3 : index -// CHECK: %[[I:.*]] = builtin.unrealized_conversion_cast %[[C]] : index to i64 -// CHECK: %[[X:.*]] = llvm.insertelement %[[A]], %[[B]][%[[I]] : i64] : vector<[4]xf32> -// CHECK: return %[[X]] : vector<[4]xf32> - -// ----- - -//===----------------------------------------------------------------------===// // vector.insert //===----------------------------------------------------------------------===// diff --git a/mlir/test/Conversion/VectorToSPIRV/vector-to-spirv.mlir b/mlir/test/Conversion/VectorToSPIRV/vector-to-spirv.mlir index f43a41a..8918f91 100644 --- a/mlir/test/Conversion/VectorToSPIRV/vector-to-spirv.mlir +++ b/mlir/test/Conversion/VectorToSPIRV/vector-to-spirv.mlir @@ -400,67 +400,6 @@ func.func @insert_dynamic_cst(%val: f32, %arg0 : vector<4xf32>) -> vector<4xf32> // ----- -// CHECK-LABEL: @extract_element -// CHECK-SAME: %[[V:.*]]: vector<4xf32>, %[[ID:.*]]: i32 -// CHECK: spirv.VectorExtractDynamic %[[V]][%[[ID]]] : vector<4xf32>, i32 -func.func @extract_element(%arg0 : vector<4xf32>, %id : i32) -> f32 { - %0 = vector.extractelement %arg0[%id : i32] : vector<4xf32> - return %0: f32 -} - -// ----- - -// CHECK-LABEL: @extract_element_cst -// CHECK-SAME: %[[V:.*]]: vector<4xf32> -// CHECK: spirv.CompositeExtract %[[V]][1 : i32] : vector<4xf32> -func.func @extract_element_cst(%arg0 : vector<4xf32>) -> f32 { - %idx = arith.constant 1 : i32 - %0 = vector.extractelement %arg0[%idx : i32] : vector<4xf32> - return %0: f32 -} - -// ----- - -// CHECK-LABEL: @extract_element_index -func.func @extract_element_index(%arg0 : vector<4xf32>, %id : index) -> f32 { - // CHECK: spirv.VectorExtractDynamic - %0 = vector.extractelement %arg0[%id : index] : vector<4xf32> - return %0: f32 -} - -// ----- - -// CHECK-LABEL: @extract_element_size5_vector -func.func @extract_element_size5_vector(%arg0 : vector<5xf32>, %id : i32) -> f32 { - // CHECK: vector.extractelement - %0 = vector.extractelement %arg0[%id : i32] : vector<5xf32> - return %0: f32 -} - -// ----- - -// CHECK-LABEL: @extract_element_size1_vector -// CHECK-SAME: (%[[S:.+]]: f32 -func.func @extract_element_size1_vector(%arg0 : f32, %i: index) -> f32 { - %bcast = vector.broadcast %arg0 : f32 to vector<1xf32> - %0 = vector.extractelement %bcast[%i : index] : vector<1xf32> - // CHECK: return %[[S]] - return %0: f32 -} - -// ----- - -// CHECK-LABEL: @extract_element_0d_vector -// CHECK-SAME: (%[[S:.+]]: f32) -func.func @extract_element_0d_vector(%arg0 : f32) -> f32 { - %bcast = vector.broadcast %arg0 : f32 to vector<f32> - %0 = vector.extractelement %bcast[] : vector<f32> - // CHECK: return %[[S]] - return %0: f32 -} - -// ----- - // CHECK-LABEL: @extract_strided_slice // CHECK-SAME: %[[ARG:.+]]: vector<4xf32> // CHECK: spirv.VectorShuffle [1 : i32, 2 : i32] %[[ARG]], %[[ARG]] : vector<4xf32>, vector<4xf32> -> vector<2xf32> @@ -473,67 +412,6 @@ func.func @extract_strided_slice(%arg0: vector<4xf32>) -> (vector<2xf32>, vector // ----- -// CHECK-LABEL: @insert_element -// CHECK-SAME: %[[VAL:.*]]: f32, %[[V:.*]]: vector<4xf32>, %[[ID:.*]]: i32 -// CHECK: spirv.VectorInsertDynamic %[[VAL]], %[[V]][%[[ID]]] : vector<4xf32>, i32 -func.func @insert_element(%val: f32, %arg0 : vector<4xf32>, %id : i32) -> vector<4xf32> { - %0 = vector.insertelement %val, %arg0[%id : i32] : vector<4xf32> - return %0: vector<4xf32> -} - -// ----- - -// CHECK-LABEL: @insert_element_cst -// CHECK-SAME: %[[VAL:.*]]: f32, %[[V:.*]]: vector<4xf32> -// CHECK: spirv.CompositeInsert %[[VAL]], %[[V]][2 : i32] : f32 into vector<4xf32> -func.func @insert_element_cst(%val: f32, %arg0 : vector<4xf32>) -> vector<4xf32> { - %idx = arith.constant 2 : i32 - %0 = vector.insertelement %val, %arg0[%idx : i32] : vector<4xf32> - return %0: vector<4xf32> -} - -// ----- - -// CHECK-LABEL: @insert_element_index -func.func @insert_element_index(%val: f32, %arg0 : vector<4xf32>, %id : index) -> vector<4xf32> { - // CHECK: spirv.VectorInsertDynamic - %0 = vector.insertelement %val, %arg0[%id : index] : vector<4xf32> - return %0: vector<4xf32> -} - -// ----- - -// CHECK-LABEL: @insert_element_size5_vector -func.func @insert_element_size5_vector(%val: f32, %arg0 : vector<5xf32>, %id : i32) -> vector<5xf32> { - // CHECK: vector.insertelement - %0 = vector.insertelement %val, %arg0[%id : i32] : vector<5xf32> - return %0 : vector<5xf32> -} - -// ----- - -// CHECK-LABEL: @insert_element_size1_vector -// CHECK-SAME: (%[[S:[a-z0-9]+]]: f32 -func.func @insert_element_size1_vector(%scalar: f32, %vector : vector<1xf32>, %i: index) -> vector<1xf32> { - %0 = vector.insertelement %scalar, %vector[%i : index] : vector<1xf32> - // CHECK: %[[V:.+]] = builtin.unrealized_conversion_cast %arg0 : f32 to vector<1xf32> - // CHECK: return %[[V]] - return %0: vector<1xf32> -} - -// ----- - -// CHECK-LABEL: @insert_element_0d_vector -// CHECK-SAME: (%[[S:[a-z0-9]+]]: f32 -func.func @insert_element_0d_vector(%scalar: f32, %vector : vector<f32>) -> vector<f32> { - %0 = vector.insertelement %scalar, %vector[] : vector<f32> - // CHECK: %[[V:.+]] = builtin.unrealized_conversion_cast %arg0 : f32 to vector<f32> - // CHECK: return %[[V]] - return %0: vector<f32> -} - -// ----- - // CHECK-LABEL: @insert_strided_slice // CHECK-SAME: %[[PART:.+]]: vector<2xf32>, %[[ALL:.+]]: vector<4xf32> // CHECK: spirv.VectorShuffle [0 : i32, 4 : i32, 5 : i32, 3 : i32] %[[ALL]], %[[PART]] : vector<4xf32>, vector<2xf32> -> vector<4xf32> |