[mlir][vector] Add more tests for ConvertVectorToLLVM (6/n) #111121
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Adds tests with scalable vectors for the Vector-To-LLVM conversion pass.
Covers the following Ops:
* `vector.insert_strided_slice`
With this change, for every test with fixed-width vectors, there should
be a corresponding example with scalable vectors (for
`vector.insert_strided_slice`). In addition:
* Test function names are updated to more accurately reflect the case
being exercised (e.g. `@insert_strided_index_slice1` ->
`@insert_strided_index_slice_index_2d_into_3d`)
* For consistency, took the liberty of updating some of the function
names for `vector.extract_strided_slice`
* `@insert_strided_slice_scalable` is effectively replaced with
`@insert_strided_slice_f32_2d_into_3d_scalable`
banach-space
requested review from
dcaballe and
nicolasvasilache
as code owners
|
@llvm/pr-subscribers-mlir Author: Andrzej Warzyński (banach-space) ChangesAdds tests with scalable vectors for the Vector-To-LLVM conversion pass.
With this change, for every test with fixed-width vectors, there should
Full diff: https://github.com/llvm/llvm-project/pull/111121.diff 1 Files Affected:
diff --git a/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir b/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir
index 2fe9ba8fead171..acdc50932697bb 100644
--- a/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir
+++ b/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir
@@ -1740,11 +1740,11 @@ func.func @print_string() {
// -----
-func.func @extract_strided_slice_f32(%arg0: vector<4xf32>) -> vector<2xf32> {
+func.func @extract_strided_slice_f32_1d_from_1d(%arg0: vector<4xf32>) -> vector<2xf32> {
%0 = vector.extract_strided_slice %arg0 {offsets = [2], sizes = [2], strides = [1]} : vector<4xf32> to vector<2xf32>
return %0 : vector<2xf32>
}
-// CHECK-LABEL: @extract_strided_slice_f32(
+// CHECK-LABEL: @extract_strided_slice_f32_1d_from_1d
// CHECK-SAME: %[[A:.*]]: vector<4xf32>)
// CHECK: %[[T0:.*]] = llvm.shufflevector %[[A]], %[[A]] [2, 3] : vector<4xf32>
// CHECK: return %[[T0]] : vector<2xf32>
@@ -1753,11 +1753,11 @@ func.func @extract_strided_slice_f32(%arg0: vector<4xf32>) -> vector<2xf32> {
// -----
-func.func @extract_strided_slice_index(%arg0: vector<4xindex>) -> vector<2xindex> {
+func.func @extract_strided_slice_index_1d_from_1d(%arg0: vector<4xindex>) -> vector<2xindex> {
%0 = vector.extract_strided_slice %arg0 {offsets = [2], sizes = [2], strides = [1]} : vector<4xindex> to vector<2xindex>
return %0 : vector<2xindex>
}
-// CHECK-LABEL: @extract_strided_slice_index(
+// CHECK-LABEL: @extract_strided_slice_index_1d_from_1d
// CHECK-SAME: %[[A:.*]]: vector<4xindex>)
// CHECK: %[[T0:.*]] = builtin.unrealized_conversion_cast %[[A]] : vector<4xindex> to vector<4xi64>
// CHECK: %[[T2:.*]] = llvm.shufflevector %[[T0]], %[[T0]] [2, 3] : vector<4xi64>
@@ -1840,32 +1840,48 @@ func.func @extract_strided_slice_f32_2d_from_2d_scalable(%arg0: vector<4x[8]xf32
// -----
-func.func @insert_strided_slice1(%b: vector<4x4xf32>, %c: vector<4x4x4xf32>) -> vector<4x4x4xf32> {
+func.func @insert_strided_slice_f32_2d_into_3d(%b: vector<4x4xf32>, %c: vector<4x4x4xf32>) -> vector<4x4x4xf32> {
%0 = vector.insert_strided_slice %b, %c {offsets = [2, 0, 0], strides = [1, 1]} : vector<4x4xf32> into vector<4x4x4xf32>
return %0 : vector<4x4x4xf32>
}
-// CHECK-LABEL: @insert_strided_slice1
+// CHECK-LABEL: @insert_strided_slice_f32_2d_into_3d
// CHECK: llvm.extractvalue {{.*}}[2] : !llvm.array<4 x array<4 x vector<4xf32>>>
// CHECK: llvm.insertvalue {{.*}}, {{.*}}[2] : !llvm.array<4 x array<4 x vector<4xf32>>>
+func.func @insert_strided_slice_f32_2d_into_3d_scalable(%b: vector<4x[4]xf32>, %c: vector<4x4x[4]xf32>) -> vector<4x4x[4]xf32> {
+ %0 = vector.insert_strided_slice %b, %c {offsets = [2, 0, 0], strides = [1, 1]} : vector<4x[4]xf32> into vector<4x4x[4]xf32>
+ return %0 : vector<4x4x[4]xf32>
+}
+// CHECK-LABEL: @insert_strided_slice_f32_2d_into_3d_scalable
+// CHECK: llvm.extractvalue {{.*}}[2] : !llvm.array<4 x array<4 x vector<[4]xf32>>>
+// CHECK: llvm.insertvalue {{.*}}, {{.*}}[2] : !llvm.array<4 x array<4 x vector<[4]xf32>>>
+
// -----
-func.func @insert_strided_index_slice1(%b: vector<4x4xindex>, %c: vector<4x4x4xindex>) -> vector<4x4x4xindex> {
+func.func @insert_strided_index_slice_index_2d_into_3d(%b: vector<4x4xindex>, %c: vector<4x4x4xindex>) -> vector<4x4x4xindex> {
%0 = vector.insert_strided_slice %b, %c {offsets = [2, 0, 0], strides = [1, 1]} : vector<4x4xindex> into vector<4x4x4xindex>
return %0 : vector<4x4x4xindex>
}
-// CHECK-LABEL: @insert_strided_index_slice1(
+// CHECK-LABEL: @insert_strided_index_slice_index_2d_into_3d
// CHECK: llvm.extractvalue {{.*}}[2] : !llvm.array<4 x array<4 x vector<4xi64>>>
// CHECK: llvm.insertvalue {{.*}}, {{.*}}[2] : !llvm.array<4 x array<4 x vector<4xi64>>>
+func.func @insert_strided_index_slice_index_2d_into_3d_scalable(%b: vector<4x[4]xindex>, %c: vector<4x4x[4]xindex>) -> vector<4x4x[4]xindex> {
+ %0 = vector.insert_strided_slice %b, %c {offsets = [2, 0, 0], strides = [1, 1]} : vector<4x[4]xindex> into vector<4x4x[4]xindex>
+ return %0 : vector<4x4x[4]xindex>
+}
+// CHECK-LABEL: @insert_strided_index_slice_index_2d_into_3d_scalable
+// CHECK: llvm.extractvalue {{.*}}[2] : !llvm.array<4 x array<4 x vector<[4]xi64>>>
+// CHECK: llvm.insertvalue {{.*}}, {{.*}}[2] : !llvm.array<4 x array<4 x vector<[4]xi64>>>
+
// -----
-func.func @insert_strided_slice2(%a: vector<2x2xf32>, %b: vector<4x4xf32>) -> vector<4x4xf32> {
+func.func @insert_strided_slice_f32_2d_into_2d(%a: vector<2x2xf32>, %b: vector<4x4xf32>) -> vector<4x4xf32> {
%0 = vector.insert_strided_slice %a, %b {offsets = [2, 2], strides = [1, 1]} : vector<2x2xf32> into vector<4x4xf32>
return %0 : vector<4x4xf32>
}
-// CHECK-LABEL: @insert_strided_slice2
+// CHECK-LABEL: @insert_strided_slice_f32_2d
//
// Subvector vector<2xf32> @0 into vector<4xf32> @2
// CHECK: %[[V2_0:.*]] = llvm.extractvalue {{.*}}[0] : !llvm.array<2 x vector<2xf32>>
@@ -1883,14 +1899,33 @@ func.func @insert_strided_slice2(%a: vector<2x2xf32>, %b: vector<4x4xf32>) -> ve
// CHECK: %[[R4_3:.*]] = llvm.shufflevector %[[R4_2]], %[[V4_3]] [4, 5, 0, 1] : vector<4xf32>
// CHECK: llvm.insertvalue %[[R4_3]], {{.*}}[3] : !llvm.array<4 x vector<4xf32>>
+// NOTE: For scalable dimensions, the corresponding "base" size must match
+// (i.e. we can only insert "full" scalable dimensions, e.g. [2] into [2], but
+// not [2] from [4]).
+
+func.func @insert_strided_slice_f32_2d_into_2d_scalable(%a: vector<2x[2]xf32>, %b: vector<4x[2]xf32>) -> vector<4x[2]xf32> {
+ %0 = vector.insert_strided_slice %a, %b {offsets = [2, 0], strides = [1, 1]} : vector<2x[2]xf32> into vector<4x[2]xf32>
+ return %0 : vector<4x[2]xf32>
+}
+
+// CHECK-LABEL: func.func @insert_strided_slice_f32_2d_into_2d_scalable
+// Subvector vector<[2]xf32> @0 into vector<[4]xf32> @2
+// CHECK: %[[A_0:.*]] = llvm.extractvalue {{.*}}[0] : !llvm.array<2 x vector<[2]xf32>>
+// Element @0 -> element @2
+// CHECK: %[[B_UPDATED:.*]] = llvm.insertvalue %[[A_0]], {{.*}}[2] : !llvm.array<4 x vector<[2]xf32>>
+// Subvector vector<[2]xf32> @1 into vector<[4]xf32> @3
+// CHECK: %[[A_1:.*]] = llvm.extractvalue {{.*}}[1] : !llvm.array<2 x vector<[2]xf32>>
+// Element @0 -> element @2
+// CHECK: llvm.insertvalue %[[A_1]], %[[B_UPDATED]][3] : !llvm.array<4 x vector<[2]xf32>>
+
// -----
-func.func @insert_strided_slice3(%arg0: vector<2x4xf32>, %arg1: vector<16x4x8xf32>) -> vector<16x4x8xf32> {
+func.func @insert_strided_slice_f32_2d_into_3d(%arg0: vector<2x4xf32>, %arg1: vector<16x4x8xf32>) -> vector<16x4x8xf32> {
%0 = vector.insert_strided_slice %arg0, %arg1 {offsets = [0, 0, 2], strides = [1, 1]}:
vector<2x4xf32> into vector<16x4x8xf32>
return %0 : vector<16x4x8xf32>
}
-// CHECK-LABEL: func @insert_strided_slice3
+// CHECK-LABEL: func @insert_strided_slice_f32_2d_into_3d
// CHECK: %[[V4_0:.*]] = llvm.extractvalue {{.*}}[0] : !llvm.array<2 x vector<4xf32>>
// CHECK: %[[V4_0_0:.*]] = llvm.extractvalue {{.*}}[0, 0] : !llvm.array<16 x array<4 x vector<8xf32>>>
// CHECK: %[[R8_0:.*]] = llvm.shufflevector %[[V4_0]], %[[V4_0]] [0, 1, 2, 3, 0, 0, 0, 0] : vector<4xf32>
@@ -1903,26 +1938,31 @@ func.func @insert_strided_slice3(%arg0: vector<2x4xf32>, %arg1: vector<16x4x8xf3
// CHECK: %[[R8_3:.*]] = llvm.shufflevector %[[R8_2]], %[[V4_0_1]] [8, 9, 0, 1, 2, 3, 14, 15] : vector<8xf32>
// CHECK: llvm.insertvalue %[[R8_3]], {{.*}}[1] : !llvm.array<4 x vector<8xf32>>
-// -----
+// NOTE: For scalable dimensions, the corresponding "base" size must match
+// (i.e. we can only insert "full" scalable dimensions, e.g. [4] into [4], but
+// not [4] from [8]).
-func.func @insert_strided_slice_scalable(%arg0 : vector<1x1x[4]xi32>, %arg1: vector<1x4x[4]xi32>) -> vector<1x4x[4]xi32> {
- %0 = vector.insert_strided_slice %arg0, %arg1 {offsets = [0, 3, 0], strides = [1, 1, 1]} : vector<1x1x[4]xi32> into vector<1x4x[4]xi32>
- return %0 : vector<1x4x[4]xi32>
+func.func @insert_strided_slice_f32_2d_into_3d_scalable(%arg0: vector<2x[4]xf32>, %arg1: vector<16x4x[4]xf32>) -> vector<16x4x[4]xf32> {
+ %0 = vector.insert_strided_slice %arg0, %arg1 {offsets = [3, 2, 0], strides = [1, 1]}:
+ vector<2x[4]xf32> into vector<16x4x[4]xf32>
+ return %0 : vector<16x4x[4]xf32>
}
-// CHECK-LABEL: func.func @insert_strided_slice_scalable(
-// CHECK-SAME: %[[ARG_0:.*]]: vector<1x1x[4]xi32>,
-// CHECK-SAME: %[[ARG_1:.*]]: vector<1x4x[4]xi32>) -> vector<1x4x[4]xi32> {
-// CHECK-DAG: %[[CAST_1:.*]] = builtin.unrealized_conversion_cast %[[ARG_0]] : vector<1x1x[4]xi32> to !llvm.array<1 x array<1 x vector<[4]xi32>>>
-// CHECK-DAG: %[[CAST_2:.*]] = builtin.unrealized_conversion_cast %[[ARG_1]] : vector<1x4x[4]xi32> to !llvm.array<1 x array<4 x vector<[4]xi32>>>
+// CHECK-LABEL: func.func @insert_strided_slice_f32_2d_into_3d_scalable(
+
+// Subvector vector4x[4]xf32> from vector<16x4x[4]xf32> @3
+// CHECK: %[[ARG_1_0:.*]] = llvm.extractvalue {{.*}}[3] : !llvm.array<16 x array<4 x vector<[4]xf32>>>
-// CHECK: %[[EXT_1:.*]] = llvm.extractvalue %[[CAST_2]][0] : !llvm.array<1 x array<4 x vector<[4]xi32>>>
-// CHECK: %[[EXT_2:.*]] = llvm.extractvalue %[[CAST_1]][0, 0] : !llvm.array<1 x array<1 x vector<[4]xi32>>>
+// Subvector vector<[4]xf32> @0 into vector<4x[4]xf32> @2
+// CHECK: %[[ARG_0_0:.*]] = llvm.extractvalue {{.*}}[0] : !llvm.array<2 x vector<[4]xf32>>
+// CHECK: %[[B_UPDATED_0:.*]] = llvm.insertvalue %[[ARG_0_0]], %[[ARG_1_0]][2] : !llvm.array<4 x vector<[4]xf32>>
-// CHECK: %[[INS_1:.*]] = llvm.insertvalue %[[EXT_2]], %[[EXT_1]][3] : !llvm.array<4 x vector<[4]xi32>>
-// CHECK: %[[INS_2:.*]] = llvm.insertvalue %[[INS_1]], %[[CAST_2]][0] : !llvm.array<1 x array<4 x vector<[4]xi32>>>
+// Subvector vector<[4]xf32> @1 into vector<4x[4]xf32> @3
+// CHECK: %[[ARG_0_1:.*]] = llvm.extractvalue {{.*}}[1] : !llvm.array<2 x vector<[4]xf32>>
+// CHECK: %[[B_UPDATED_1:.*]] = llvm.insertvalue %[[ARG_0_1]], %[[B_UPDATED_0]][3] : !llvm.array<4 x vector<[4]xf32>>
-// CHECK: builtin.unrealized_conversion_cast %[[INS_2]] : !llvm.array<1 x array<4 x vector<[4]xi32>>> to vector<1x4x[4]xi32>
+// Subvector vector4x[4]xf32> into vector<16x4x[4]xf32> @3
+// CHECK: llvm.insertvalue %[[B_UPDATED_1]], {{.*}}[3] : !llvm.array<16 x array<4 x vector<[4]xf32>>>
// -----
|
nujaa
reviewed
Oct 4, 2024
| %0 = vector.insert_strided_slice %a, %b {offsets = [2, 2], strides = [1, 1]} : vector<2x2xf32> into vector<4x4xf32> | ||
| return %0 : vector<4x4xf32> | ||
| } | ||
|
|
||
| // CHECK-LABEL: @insert_strided_slice2 | ||
| // CHECK-LABEL: @insert_strided_slice_f32_2d |
There was a problem hiding this comment.
This check does not match the func name.
| // CHECK-DAG: %[[CAST_2:.*]] = builtin.unrealized_conversion_cast %[[ARG_1]] : vector<1x4x[4]xi32> to !llvm.array<1 x array<4 x vector<[4]xi32>>> | ||
| // CHECK-LABEL: func.func @insert_strided_slice_f32_2d_into_3d_scalable( | ||
|
|
||
| // Subvector vector4x[4]xf32> from vector<16x4x[4]xf32> @3 |
There was a problem hiding this comment.
I am not sure about those subvector* lines. Are these supposed to be CHECKs ?
And a tiny NIT
Suggested change
| // Subvector vector4x[4]xf32> from vector<16x4x[4]xf32> @3 | |
| // Subvector vector<4x[4]xf32> from vector<16x4x[4]xf32> @3 |
There was a problem hiding this comment.
The comments? That's just to annotate the tests so that it's clearer what's happening. And it's a format borrowed from elsewhere in the file:
llvm-project/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir
Lines 1868 to 1884 in d9d6233
I find it helpful as we are not really checking the sources (see {{.*}}) - that's primarily to simplify things and to avoid checking builtin_unrealised_cast_conversion 😅
|
|
||
| // CHECK: builtin.unrealized_conversion_cast %[[INS_2]] : !llvm.array<1 x array<4 x vector<[4]xi32>>> to vector<1x4x[4]xi32> | ||
| // Subvector vector4x[4]xf32> into vector<16x4x[4]xf32> @3 |
There was a problem hiding this comment.
Suggested change
| // Subvector vector4x[4]xf32> into vector<16x4x[4]xf32> @3 | |
| // Subvector vector<4x[4]xf32> into vector<16x4x[4]xf32> @3 |
banach-space
changed the title
Kyvangka1610
added a commit
to Kyvangka1610/llvm-project
that referenced
this pull request
Oct 7, 2024
* commit 'FETCH_HEAD': [X86] getIntImmCostInst - pull out repeated Imm.getBitWidth() calls. NFC. [X86] Add test coverage for llvm#111323 [Driver] Use empty multilib file in another test (llvm#111352) [clang][OpenMP][test] Use x86_64-linux-gnu triple for test referencing avx512f feature (llvm#111337) [doc] Fix Kaleidoscope tutorial chapter 3 code snippet and full listing discrepancies (llvm#111289) [Flang][OpenMP] Improve entry block argument creation and binding (llvm#110267) [x86] combineMul - handle 0/-1 KnownBits cases before MUL_IMM logic (REAPPLIED) [llvm-dis] Fix non-deterministic disassembly across multiple inputs (llvm#110988) [lldb][test] TestDataFormatterLibcxxOptionalSimulator.py: change order of ifdefs [lldb][test] Add libcxx-simulators test for std::optional (llvm#111133) [x86] combineMul - use computeKnownBits directly to find MUL_IMM constant splat. (REAPPLIED) Reland "[lldb][test] TestDataFormatterLibcxxStringSimulator.py: add new padding layout" (llvm#111123) Revert "[x86] combineMul - use computeKnownBits directly to find MUL_IMM constant splat." update_test_checks: fix a simple regression (llvm#111347) [LegalizeVectorTypes] Always widen fabs (llvm#111298) [lsan] Make ReportUnsuspendedThreads return bool also for Fuchsia [mlir][vector] Add more tests for ConvertVectorToLLVM (6/n) (llvm#111121) [bazel] port 9144fed [SystemZ] Remove inlining threshold multiplier. (llvm#106058) [LegalizeVectorTypes] When widening don't check for libcalls if promoted (llvm#111297) [clang][Driver] Improve multilib custom error reporting (llvm#110804) [clang][Driver] Rename "FatalError" key to "Error" in multilib.yaml (llvm#110804) [LLVM][Maintainers] Update release managers (llvm#111164) [Clang][Driver] Add option to provide path for multilib's YAML config file (llvm#109640) [LoopVectorize] Remove redundant code in emitSCEVChecks (llvm#111132) [AMDGPU] Only emit SCOPE_SYS global_wb (llvm#110636) [ELF] Change Ctx::target to unique_ptr (llvm#111260) [ELF] Pass Ctx & to some free functions [RISCV] Only disassemble fcvtmod.w.d if the rounding mode is rtz. (llvm#111308) [Clang] Remove the special-casing for RequiresExprBodyDecl in BuildResolvedCallExpr() after fd87d76 (llvm#111277) [ELF] Pass Ctx & to InputFile [clang-format] Add AlignFunctionDeclarations to AlignConsecutiveDeclarations (llvm#108241) [AMDGPU] Support preloading hidden kernel arguments (llvm#98861) [ELF] Move static nextGroupId isInGroup to LinkerDriver [clangd] Add ArgumentLists config option under Completion (llvm#111322) [ELF] Pass Ctx & to SyntheticSections [ELF] Pass Ctx & to Symbols [ELF] Pass Ctx & to Symbols [ELF] getRelocTargetVA: pass Ctx and Relocation. NFC [clang-tidy] Avoid capturing a local variable in a static lambda in UseRangesCheck (llvm#111282) [VPlan] Use pointer to member 0 as VPInterleaveRecipe's pointer arg. (llvm#106431) [clangd] Simplify ternary expressions with std::optional::value_or (NFC) (llvm#111309) [libc++][format][2/3] Optimizes c-string arguments. (llvm#101805) [RISCV] Combine RVBUnary and RVKUnary into classes that are more similar to ALU(W)_r(r/i). NFC (llvm#111279) [ELF] Pass Ctx & to InputFiles [libc] GPU RPC interface: add return value to `rpc_host_call` (llvm#111288) Signed-off-by: kyvangka1610 <kyvangka2002@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Adds tests with scalable vectors for the Vector-To-LLVM conversion pass.
Covers the following Ops:
vector.insert_strided_sliceWith this change, for every test with fixed-width vectors, there should
be a corresponding example with scalable vectors (for
vector.insert_strided_slice). In addition:being exercised (e.g.
@insert_strided_index_slice1->@insert_strided_index_slice_index_2d_into_3d)names for
vector.extract_strided_slice@insert_strided_slice_scalableis effectively replaced with@insert_strided_slice_f32_2d_into_3d_scalable