Skip to content

Upgrading llvm and stablehlo hash #3053

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 11 commits into from
Feb 14, 2025
Merged

Upgrading llvm and stablehlo hash #3053

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
3784890
upgrading llvm and stablehlo hash. Fixing mlir tests
christopherlmunoz Jan 29, 2025
c6307f4
commenting out failng tests
christopherlmunoz Feb 6, 2025
6dcd231
Merge branch 'main' into llvm_update
christopherlmunoz Feb 6, 2025
649afcb
commenting out failng tests
christopherlmunoz Feb 6, 2025
79e36a0
dynamic accType sizing
christopherlmunoz Feb 7, 2025
a65d3a9
Merge branch 'main' into llvm_update
christopherlmunoz Feb 7, 2025
84bcdb5
Merge branch 'main' into llvm_update
christopherlmunoz Feb 10, 2025
4e19da5
fixing vector shapecast bug introduced by upgraded llvm
christopherlmunoz Feb 14, 2025
6dae8b4
Merge branch 'llvm_update' of https://github.com/christopherlmunoz/on…
christopherlmunoz Feb 14, 2025
62a16c4
Merge branch 'main' into llvm_update
christopherlmunoz Feb 14, 2025
6a5dd80
formating fixes
christopherlmunoz Feb 14, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion docs/BuildOnLinuxOSX.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ Firstly, install MLIR (as a part of LLVM-Project):
``` bash
git clone -n https://github.com/llvm/llvm-project.git
# Check out a specific branch that is known to work with ONNX-MLIR.
cd llvm-project && git checkout e86910337f98e57f5b9253f7d80d5b916eb1d97e && cd ..
cd llvm-project && git checkout b270525f730be6e7196667925f5a9bfa153262e9 && cd ..
```

[same-as-file]: <> (utils/build-mlir.sh)
Expand Down
2 changes: 1 addition & 1 deletion docs/BuildOnWindows.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,7 +52,7 @@ Install MLIR (as a part of LLVM-Project):
```shell
git clone -n https://github.com/llvm/llvm-project.git
# Check out a specific branch that is known to work with ONNX-MLIR.
cd llvm-project && git checkout e86910337f98e57f5b9253f7d80d5b916eb1d97e && cd ..
cd llvm-project && git checkout b270525f730be6e7196667925f5a9bfa153262e9 && cd ..
```

[same-as-file]: <> (utils/build-mlir.cmd)
Expand Down
10 changes: 6 additions & 4 deletions src/Conversion/ONNXToTOSA/Math/Conv2D.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,7 @@ Value createConvInGroups(PatternRewriter &rewriter, Operation *op,
const llvm::ArrayRef<int64_t> weightShape, Value &newInput,
Value &newWeight, Value &bias, const int64_t groups,
DenseI64ArrayAttr &pads, DenseI64ArrayAttr &strides,
DenseI64ArrayAttr &dilations) {
DenseI64ArrayAttr &dilations, TypeAttr &accType) {
// Set up constants outside of loop
const int64_t sizeOfSliceInput = weightShape[1];
const int64_t sizeOfSliceKernel = weightShape[0] / groups;
Expand Down Expand Up @@ -72,7 +72,7 @@ Value createConvInGroups(PatternRewriter &rewriter, Operation *op,
mlir::cast<ShapedType>(resultType).getElementType());
Value tempConv2D = tosa::CreateOpAndInfer<mlir::tosa::Conv2DOp>(rewriter,
op->getLoc(), newConvOutputType, newSliceInput, newSliceWeight,
newSliceBias, pads, strides, dilations);
newSliceBias, pads, strides, dilations, accType);
// Add value to vector
sliceValues.push_back(tempConv2D);
}
Expand Down Expand Up @@ -147,6 +147,8 @@ class ONNXConvOpLoweringToTOSA : public ConversionPattern {
DenseI64ArrayAttr newPads =
rewriter.getDenseI64ArrayAttr({pads[0], pads[2], pads[1], pads[3]});

TypeAttr accType = mlir::TypeAttr::get(rewriter.getF32Type());

// Handle group parameter by creating multiple convs
const int64_t group = adaptor.getGroup();
Value conv2D = NULL;
Expand All @@ -157,11 +159,11 @@ class ONNXConvOpLoweringToTOSA : public ConversionPattern {

conv2D = tosa::CreateOpAndInfer<mlir::tosa::Conv2DOp>(rewriter,
convOp->getLoc(), newConvOutputType, newInput, newWeight, bias,
newPads, strides, dilations);
newPads, strides, dilations, accType);
} else {
conv2D = createConvInGroups(rewriter, convOp, tosaBuilder, resultType,
weightShape, newInput, newWeight, bias, group, newPads, strides,
dilations);
dilations, accType);
}

// Convert output [N,OH,OW,OC] -> [N,OC,OH,OW]
Expand Down
8 changes: 3 additions & 5 deletions src/Conversion/ONNXToTOSA/ONNXToTOSALegalizeUtils.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,7 +48,7 @@ Value buildOnnxToTosaPaddingConstOp(mlir::PatternRewriter &rewriter,

// Create a new pad vec in the right format
// ONNX : [b1, b2, b3, b4, e1, e2, e3, e4]
// TOSA :[[b1, e1], [b2, e2], [b3, e3], [b4, e4]]
// TOSA :[b1, e1, b2, e2, b3, e3, b4, e4]

// Adds any initial or last vals, not included in onnxPads.
llvm::SmallVector<int64_t, 8> tosaPads{initialVals};
Expand All @@ -59,11 +59,9 @@ Value buildOnnxToTosaPaddingConstOp(mlir::PatternRewriter &rewriter,
tosaPads.push_back(onnxPads[i + dimSize]);
}
tosaPads.insert(tosaPads.end(), lastVals.begin(), lastVals.end());

// TOSA format groups dimensions by 2.
const unsigned int numberOfDims = tosaPads.size() / 2;
TosaBuilder tosaBuilder(rewriter, loc);
return tosaBuilder.getConst(tosaPads, {numberOfDims, 2});
return tosaBuilder.getConst(
tosaPads, {static_cast<int64_t>(tosaPads.size())});
}

} // namespace tosa
Expand Down
96 changes: 56 additions & 40 deletions test/backend/inference_backend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -847,36 +847,48 @@ def get_test_models():
# "test_training_dropout_zero_ratio_mask_cpu": {STATIC_SHAPE:{}, DYNAMIC_SHAPE:{-1:{-1}}},
# ==OP== DynamicQuantizeLinear
# ==MIN== 11
"test_dynamicquantizelinear_cpu": {
STATIC_SHAPE: {},
DYNAMIC_SHAPE: {-1: {-1}},
CONSTANT_INPUT: {-1},
},
"test_dynamicquantizelinear_expanded_cpu": {
STATIC_SHAPE: {},
DYNAMIC_SHAPE: {-1: {-1}},
CONSTANT_INPUT: {-1},
},
"test_dynamicquantizelinear_max_adjusted_cpu": {
STATIC_SHAPE: {},
DYNAMIC_SHAPE: {-1: {-1}},
CONSTANT_INPUT: {-1},
},
"test_dynamicquantizelinear_max_adjusted_expanded_cpu": {
STATIC_SHAPE: {},
DYNAMIC_SHAPE: {-1: {-1}},
CONSTANT_INPUT: {-1},
},
"test_dynamicquantizelinear_min_adjusted_cpu": {
STATIC_SHAPE: {},
DYNAMIC_SHAPE: {-1: {-1}},
CONSTANT_INPUT: {-1},
},
"test_dynamicquantizelinear_min_adjusted_expanded_cpu": {
STATIC_SHAPE: {},
DYNAMIC_SHAPE: {-1: {-1}},
CONSTANT_INPUT: {-1},
},
# TODO uncomment when issue #3068 is resolved.
# "vector.shape_cast operation fails on s390x"
# "test_dynamicquantizelinear_cpu": {
# STATIC_SHAPE: {},
# DYNAMIC_SHAPE: {-1: {-1}},
# CONSTANT_INPUT: {-1},
# },
# TODO uncomment when issue #3068 is resolved.
# "vector.shape_cast operation fails on s390x"
# "test_dynamicquantizelinear_expanded_cpu": {
# STATIC_SHAPE: {},
# DYNAMIC_SHAPE: {-1: {-1}},
# CONSTANT_INPUT: {-1},
# },
# TODO uncomment when issue #3068 is resolved.
# "vector.shape_cast operation fails on s390x"
# "test_dynamicquantizelinear_max_adjusted_cpu": {
# STATIC_SHAPE: {},
# DYNAMIC_SHAPE: {-1: {-1}},
# CONSTANT_INPUT: {-1},
# },
# TODO uncomment when issue #3068 is resolved.
# "vector.shape_cast operation fails on s390x"
# "test_dynamicquantizelinear_max_adjusted_expanded_cpu": {
# STATIC_SHAPE: {},
# DYNAMIC_SHAPE: {-1: {-1}},
# CONSTANT_INPUT: {-1},
# },
# TODO uncomment when issue #3068 is resolved.
# "vector.shape_cast operation fails on s390x"
# "test_dynamicquantizelinear_min_adjusted_cpu": {
# STATIC_SHAPE: {},
# DYNAMIC_SHAPE: {-1: {-1}},
# CONSTANT_INPUT: {-1},
# },
# TODO uncomment when issue #3068 is resolved.
# "vector.shape_cast operation fails on s390x"
# "test_dynamicquantizelinear_min_adjusted_expanded_cpu": {
# STATIC_SHAPE: {},
# DYNAMIC_SHAPE: {-1: {-1}},
# CONSTANT_INPUT: {-1},
# },
# ==OP== Einsum
# ==MIN== 12
# ==LIM== Limited to the types supported by ReduceSum and MatMul (which we decompose to in most cases) which exclude integers with width < 32. `inputs` must have static dimensions.
Expand Down Expand Up @@ -2323,11 +2335,13 @@ def get_test_models():
# ==MIN== 10
# ==LIM== Does not support per-axis and i8 quantization. Does not support int4 and uint4.
# "test_quantizelinear_axis_cpu": {STATIC_SHAPE:{}, DYNAMIC_SHAPE:{-1:{-1}}, CONSTANT_INPUT:{-1}},
"test_quantizelinear_cpu": {
STATIC_SHAPE: {},
DYNAMIC_SHAPE: {-1: {-1}},
CONSTANT_INPUT: {-1},
},
# TODO uncomment when issue #3068 is resolved.
# "vector.shape_cast operation fails on s390x"
# "test_quantizelinear_cpu": {
# STATIC_SHAPE: {},
# DYNAMIC_SHAPE: {-1: {-1}},
# CONSTANT_INPUT: {-1},
# },
# ==OP== Range
# ==MIN== 11
"test_range_float_type_positive_delta_cpu": {
Expand Down Expand Up @@ -2815,11 +2829,13 @@ def get_test_models():
},
# ==OP== Round
# ==MIN== 11
"test_round_cpu": {
STATIC_SHAPE: {},
DYNAMIC_SHAPE: {-1: {-1}},
CONSTANT_INPUT: {-1},
},
# TODO uncomment when issue #3068 is resolved.
# "vector.shape_cast operation fails on s390x"
# "test_round_cpu": {
# STATIC_SHAPE: {},
# DYNAMIC_SHAPE: {-1: {-1}},
# CONSTANT_INPUT: {-1},
# },
# ==OP== Scan
# ==MIN== 8
# ==LIM== Does not support dynamic shapes. Does not support int4 and uint4.
Expand Down
33 changes: 8 additions & 25 deletions test/mlir/conversion/krnl_to_llvm/krnl_math_function_lowering.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,9 +16,7 @@ func.func @test_krnl_erf_lowering(%arg0: memref<10x10xf32>) -> memref<10x10xf32>

// CHECK-LABEL: test_krnl_erf_lowering
// CHECK: [[MEMREF_IN:%.+]] = llvm.insertvalue %arg6, {{.*}}[4, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[BUILTIN_CAST_0:%.+]] = builtin.unrealized_conversion_cast [[MEMREF_IN]] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> to memref<10x10xf32>
// CHECK: [[BUILTIN_CAST_1:%.+]] = builtin.unrealized_conversion_cast [[BUILTIN_CAST_0]] : memref<10x10xf32> to !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[BUILTIN_CAST_1]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[MEMREF_IN]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA_IN:%.+]] = llvm.getelementptr [[DATA]]{{.*}} : (!llvm.ptr, i64) -> !llvm.ptr, f32
// CHECK: [[SCALAR_IN:%.+]] = llvm.load [[DATA_IN]] : !llvm.ptr
// CHECK: [[ERF_RES:%.+]] = llvm.call @erff([[SCALAR_IN]]) : (f32) -> f32
Expand All @@ -41,9 +39,7 @@ func.func @test_krnl_acos_lowering(%arg0: memref<10x10xf32>) -> memref<10x10xf32

// CHECK-LABEL: test_krnl_acos_lowering
// CHECK: [[MEMREF_IN:%.+]] = llvm.insertvalue %arg6, {{.*}}[4, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[BUILTIN_CAST_0:%.+]] = builtin.unrealized_conversion_cast [[MEMREF_IN]] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> to memref<10x10xf32>
// CHECK: [[BUILTIN_CAST_1:%.+]] = builtin.unrealized_conversion_cast [[BUILTIN_CAST_0]] : memref<10x10xf32> to !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[BUILTIN_CAST_1]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[MEMREF_IN]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA_IN:%.+]] = llvm.getelementptr [[DATA]]{{.*}} : (!llvm.ptr, i64) -> !llvm.ptr, f32
// CHECK: [[SCALAR_IN:%.+]] = llvm.load [[DATA_IN]] : !llvm.ptr
// CHECK: [[ACOS_RES:%.+]] = llvm.call @acosf([[SCALAR_IN]]) : (f32) -> f32
Expand All @@ -66,9 +62,7 @@ func.func @test_krnl_acosh_lowering(%arg0: memref<10x10xf32>) -> memref<10x10xf3

// CHECK-LABEL: test_krnl_acosh_lowering
// CHECK: [[MEMREF_IN:%.+]] = llvm.insertvalue %arg6, {{.*}}[4, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[BUILTIN_CAST_0:%.+]] = builtin.unrealized_conversion_cast [[MEMREF_IN]] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> to memref<10x10xf32>
// CHECK: [[BUILTIN_CAST_1:%.+]] = builtin.unrealized_conversion_cast [[BUILTIN_CAST_0]] : memref<10x10xf32> to !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[BUILTIN_CAST_1]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[MEMREF_IN]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA_IN:%.+]] = llvm.getelementptr [[DATA]]{{.*}} : (!llvm.ptr, i64) -> !llvm.ptr, f32
// CHECK: [[SCALAR_IN:%.+]] = llvm.load [[DATA_IN]] : !llvm.ptr
// CHECK: [[ACOS_RES:%.+]] = llvm.call @acoshf([[SCALAR_IN]]) : (f32) -> f32
Expand All @@ -91,9 +85,7 @@ func.func @test_krnl_asin_lowering(%arg0: memref<10x10xf32>) -> memref<10x10xf32

// CHECK-LABEL: test_krnl_asin_lowering
// CHECK: [[MEMREF_IN:%.+]] = llvm.insertvalue %arg6, {{.*}}[4, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[BUILTIN_CAST_0:%.+]] = builtin.unrealized_conversion_cast [[MEMREF_IN]] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> to memref<10x10xf32>
// CHECK: [[BUILTIN_CAST_1:%.+]] = builtin.unrealized_conversion_cast [[BUILTIN_CAST_0]] : memref<10x10xf32> to !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[BUILTIN_CAST_1]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[MEMREF_IN]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA_IN:%.+]] = llvm.getelementptr [[DATA]]{{.*}} : (!llvm.ptr, i64) -> !llvm.ptr, f32
// CHECK: [[SCALAR_IN:%.+]] = llvm.load [[DATA_IN]] : !llvm.ptr
// CHECK: [[ACOS_RES:%.+]] = llvm.call @asinf([[SCALAR_IN]]) : (f32) -> f32
Expand All @@ -116,9 +108,7 @@ func.func @test_krnl_asinh_lowering(%arg0: memref<10x10xf32>) -> memref<10x10xf3

// CHECK-LABEL: test_krnl_asinh_lowering
// CHECK: [[MEMREF_IN:%.+]] = llvm.insertvalue %arg6, {{.*}}[4, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[BUILTIN_CAST_0:%.+]] = builtin.unrealized_conversion_cast [[MEMREF_IN]] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> to memref<10x10xf32>
// CHECK: [[BUILTIN_CAST_1:%.+]] = builtin.unrealized_conversion_cast [[BUILTIN_CAST_0]] : memref<10x10xf32> to !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[BUILTIN_CAST_1]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[MEMREF_IN]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA_IN:%.+]] = llvm.getelementptr [[DATA]]{{.*}} : (!llvm.ptr, i64) -> !llvm.ptr, f32
// CHECK: [[SCALAR_IN:%.+]] = llvm.load [[DATA_IN]] : !llvm.ptr
// CHECK: [[ACOS_RES:%.+]] = llvm.call @asinhf([[SCALAR_IN]]) : (f32) -> f32
Expand All @@ -141,9 +131,7 @@ func.func @test_krnl_atan_lowering(%arg0: memref<10x10xf32>) -> memref<10x10xf32

// CHECK-LABEL: test_krnl_atan_lowering
// CHECK: [[MEMREF_IN:%.+]] = llvm.insertvalue %arg6, {{.*}}[4, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[BUILTIN_CAST_0:%.+]] = builtin.unrealized_conversion_cast [[MEMREF_IN]] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> to memref<10x10xf32>
// CHECK: [[BUILTIN_CAST_1:%.+]] = builtin.unrealized_conversion_cast [[BUILTIN_CAST_0]] : memref<10x10xf32> to !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[BUILTIN_CAST_1]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[MEMREF_IN]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA_IN:%.+]] = llvm.getelementptr [[DATA]]{{.*}} : (!llvm.ptr, i64) -> !llvm.ptr, f32
// CHECK: [[SCALAR_IN:%.+]] = llvm.load [[DATA_IN]] : !llvm.ptr
// CHECK: [[ACOS_RES:%.+]] = llvm.call @atanf([[SCALAR_IN]]) : (f32) -> f32
Expand All @@ -165,9 +153,7 @@ func.func @test_krnl_atanh_lowering(%arg0: memref<10x10xf32>) -> memref<10x10xf3

// CHECK-LABEL: test_krnl_atanh_lowering
// CHECK: [[MEMREF_IN:%.+]] = llvm.insertvalue %arg6, {{.*}}[4, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[BUILTIN_CAST_0:%.+]] = builtin.unrealized_conversion_cast [[MEMREF_IN]] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> to memref<10x10xf32>
// CHECK: [[BUILTIN_CAST_1:%.+]] = builtin.unrealized_conversion_cast [[BUILTIN_CAST_0]] : memref<10x10xf32> to !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[BUILTIN_CAST_1]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[MEMREF_IN]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA_IN:%.+]] = llvm.getelementptr [[DATA]]{{.*}} : (!llvm.ptr, i64) -> !llvm.ptr, f32
// CHECK: [[SCALAR_IN:%.+]] = llvm.load [[DATA_IN]] : !llvm.ptr
// CHECK: [[ACOS_RES:%.+]] = llvm.call @atanhf([[SCALAR_IN]]) : (f32) -> f32
Expand All @@ -189,12 +175,9 @@ func.func @test_krnl_tan_lowering(%arg0: memref<10x10xf32>) -> memref<10x10xf32>

// CHECK-LABEL: test_krnl_tan_lowering
// CHECK: [[MEMREF_IN:%.+]] = llvm.insertvalue %arg6, {{.*}}[4, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[BUILTIN_CAST_0:%.+]] = builtin.unrealized_conversion_cast [[MEMREF_IN]] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> to memref<10x10xf32>
// CHECK: [[BUILTIN_CAST_1:%.+]] = builtin.unrealized_conversion_cast [[BUILTIN_CAST_0]] : memref<10x10xf32> to !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[BUILTIN_CAST_1]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA:%.+]] = llvm.extractvalue [[MEMREF_IN]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: [[DATA_IN:%.+]] = llvm.getelementptr [[DATA]]{{.*}} : (!llvm.ptr, i64) -> !llvm.ptr, f32
// CHECK: [[SCALAR_IN:%.+]] = llvm.load [[DATA_IN]] : !llvm.ptr
// CHECK: [[ACOS_RES:%.+]] = llvm.call @tanf([[SCALAR_IN]]) : (f32) -> f32
// CHECK: [[DATA_OUT:%.+]] = llvm.getelementptr {{.*}} : (!llvm.ptr, i64) -> !llvm.ptr, f32
// CHECK: llvm.store [[ACOS_RES]], [[DATA_OUT]] : f32, !llvm.ptr

Loading