Skip to content

[CPU] Populate to_elements unrolling patterns in LLVM conversion. #22010

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
hanhanW merged 4 commits into from
Sep 17, 2025
Merged

[CPU] Populate to_elements unrolling patterns in LLVM conversion. #22010

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
2f6d32a
[CPU] Populate to_elements unrolling patterns in LLVM conversion.
hanhanW Sep 16, 2025
361ae89
Add test to AMDGPU
hanhanW Sep 16, 2025
fbd367f
disable the test on riscv
hanhanW Sep 17, 2025
cac09f2
exclude the test properly
hanhanW Sep 17, 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: 2 additions & 0 deletions compiler/src/iree/compiler/Codegen/LLVMCPU/ConvertToLLVM.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1015,6 +1015,7 @@ void ConvertToLLVMPass::runOnOperation() {
vector::populateVectorMaskOpLoweringPatterns(patterns);
vector::populateVectorShapeCastLoweringPatterns(patterns);
vector::populateVectorFromElementsLoweringPatterns(patterns);
vector::populateVectorToElementsLoweringPatterns(patterns);
// TODO: doubtful that the "default" does what one want here, it is likely
// better to use shuffle.
vector::populateVectorTransposeLoweringPatterns(
Expand Down Expand Up @@ -1091,6 +1092,7 @@ void ConvertToLLVMPass::runOnOperation() {
populateVectorToLLVMConversionPatterns(typeConverter, patterns,
reassociateFpReductions);
vector::populateVectorFromElementsLoweringPatterns(patterns);
vector::populateVectorToElementsLoweringPatterns(patterns);
ub::populateUBToLLVMConversionPatterns(typeConverter, patterns);
vector::populateVectorTransferLoweringPatterns(patterns,
/*maxTransferRank=*/1);
Expand Down
15 changes: 15 additions & 0 deletions tests/e2e/regression/BUILD.bazel
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,6 +86,20 @@ iree_check_single_backend_test_suite(
target_backend = "llvm-cpu",
)

# TODO(22013): Merge the test suite into the main one, after the issue is fixed.
iree_check_single_backend_test_suite(
name = "check_regression_dynamic_gather_attention_llvm-cpu",
srcs = [
"dynamic_gather_attention.mlir",
],
compiler_flags = ["--iree-llvmcpu-target-cpu=generic"],
driver = "local-task",
tags = [
"noriscv",
],
target_backend = "llvm-cpu",
)

iree_check_single_backend_test_suite(
name = "check_regression_tosa_llvm-cpu",
srcs = [
Expand Down Expand Up @@ -141,6 +155,7 @@ iree_check_single_backend_test_suite(
iree_check_single_backend_test_suite(
name = "check_regression_hip",
srcs = [
"dynamic_gather_attention.mlir",
"linalg_ops_dynamic.mlir",
"split_reduction_using_tiling.mlir",
],
Expand Down
16 changes: 16 additions & 0 deletions tests/e2e/regression/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -80,6 +80,21 @@ iree_check_single_backend_test_suite(
"stablehlo"
)

iree_check_single_backend_test_suite(
NAME
check_regression_dynamic_gather_attention_llvm-cpu
SRCS
"dynamic_gather_attention.mlir"
TARGET_BACKEND
"llvm-cpu"
DRIVER
"local-task"
COMPILER_FLAGS
"--iree-llvmcpu-target-cpu=generic"
LABELS
"noriscv"
)

iree_check_single_backend_test_suite(
NAME
check_regression_tosa_llvm-cpu
Expand Down Expand Up @@ -189,6 +204,7 @@ iree_check_single_backend_test_suite(
NAME
check_regression_hip
SRCS
"dynamic_gather_attention.mlir"
"linalg_ops_dynamic.mlir"
"split_reduction_using_tiling.mlir"
TARGET_BACKEND
Expand Down
26 changes: 26 additions & 0 deletions tests/e2e/regression/dynamic_gather_attention.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,26 @@
#map = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d4)>
#map1 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d5, d1, d6, d4)>
#map2 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d5, d1, d6, d3)>
#map3 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> ()>
#map4 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d5, d6)>
#map5 = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3)>
func.func @gather_attention() {
%0 = util.unfoldable_constant dense<1.000000e+00> : tensor<32x4x2x32xf16>
%1 = flow.tensor.dynamic_constant dense<5.000000e-01> : tensor<2x4x16x32xf16> -> tensor<?x4x16x32xf16>
%2 = flow.tensor.dynamic_constant dense<1.500000e+00> : tensor<2x4x16x32xf16> -> tensor<?x4x16x32xf16>
%3 = flow.tensor.dynamic_constant dense<1> : tensor<32x2xi64> -> tensor<32x?xi64>
%4 = flow.tensor.dynamic_constant dense<1.500000e+00> : tensor<32x4x2x2x16xf16> -> tensor<32x4x2x?x16xf16>
%c1 = arith.constant 1 : index
%dim = tensor.dim %3, %c1 : tensor<32x?xi64>
%5 = tensor.empty(%dim) : tensor<32x?x4x16x32xf16>
%6 = iree_linalg_ext.gather dimension_map = [0] ins(%1, %3 : tensor<?x4x16x32xf16>, tensor<32x?xi64>) outs(%5 : tensor<32x?x4x16x32xf16>) -> tensor<32x?x4x16x32xf16>
%7 = iree_linalg_ext.gather dimension_map = [0] ins(%2, %3 : tensor<?x4x16x32xf16>, tensor<32x?xi64>) outs(%5 : tensor<32x?x4x16x32xf16>) -> tensor<32x?x4x16x32xf16>
%cst = arith.constant 1.767580e-01 : f16
%8 = tensor.empty() : tensor<32x4x2x32xf16>
%9 = iree_linalg_ext.attention {indexing_maps = [#map, #map1, #map2, #map3, #map4, #map5]} ins(%0, %6, %7, %cst, %4 : tensor<32x4x2x32xf16>, tensor<32x?x4x16x32xf16>, tensor<32x?x4x16x32xf16>, f16, tensor<32x4x2x?x16xf16>) outs(%8 : tensor<32x4x2x32xf16>) {
^bb0(%arg0: f32):
iree_linalg_ext.yield %arg0 : f32
} -> tensor<32x4x2x32xf16>
check.expect_almost_eq_const(%9, dense<1.500000e+00> : tensor<32x4x2x32xf16>) : tensor<32x4x2x32xf16>
return
}
Loading