[Mlir-commits] [mlir] [mlir][vector] Add more tests for ConvertVectorToLLVM (5/n) (PR #106510)
Andrzej Warzyński
llvmlistbot at llvm.org
Thu Aug 29 01:46:15 PDT 2024
https://github.com/banach-space created https://github.com/llvm/llvm-project/pull/106510
Adds tests with scalable vectors for the Vector-To-LLVM conversion pass.
Covers the following Ops:
* vector.extract_strided_slice
For consistency with other tests, I have also removed "vector" from test
functions names for:
* vector.print
* vector.type_cast
* vector.extract_strided_slice
The first two Ops precede `vector.extract_strided_slice` in the test
file, i.e. those should be next to be updated in this series of patches.
However,
* For `vector.print`, we don't use vectors in this test file (that
would require running VectorToSCF).
* For `vector.type_cast`, the existing tests assume fixed-width sizes.
We need to write new tests and I am leaving that as a TODO.
Note, I've also updated test function names to be more descriptive and
consistent with other tests, e.g.
* `@extract_strided_slice3` -> `@extract_strided_slice_f32_2d_from_2d`
>From 7d6a1c3a506901e3e2cf6b07b1d7d351c2870ea2 Mon Sep 17 00:00:00 2001
From: Andrzej Warzynski <andrzej.warzynski at arm.com>
Date: Thu, 29 Aug 2024 08:58:56 +0100
Subject: [PATCH] [mlir][vector] Add more tests for ConvertVectorToLLVM (5/n)
Adds tests with scalable vectors for the Vector-To-LLVM conversion pass.
Covers the following Ops:
* vector.extract_strided_slice
For consistency with other tests, I have also removed "vector" from test
functions names for:
* vector.print
* vector.type_cast
* vector.extract_strided_slice
The first two Ops precede `vector.extract_strided_slice` in the test
file, i.e. those should be next to be updated in this series of patches.
However,
* For `vector.print`, we don't use vectors in this test file (that
would require running VectorToSCF).
* For `vector.type_cast`, the existing tests assume fixed-width sizes.
We need to write new tests and I am leaving that as a TODO.
Note, I've also updated test function names to be more descriptive and
consistent with other tests, e.g.
* `@extract_strided_slice3` -> `@extract_strided_slice_f32_2d_from_2d`
---
.../VectorToLLVM/vector-to-llvm.mlir | 153 ++++++++++--------
1 file changed, 88 insertions(+), 65 deletions(-)
diff --git a/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir b/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir
index 63bcecd863e95d..2c8b1a2a6ff1f6 100644
--- a/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir
+++ b/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir
@@ -1,5 +1,7 @@
// RUN: mlir-opt %s -convert-vector-to-llvm -split-input-file | FileCheck %s
+// TODO: Add tests for for vector.type_cast that would cover scalable vectors
+
func.func @bitcast_f32_to_i32_vector_0d(%input: vector<f32>) -> vector<i32> {
%0 = vector.bitcast %input : vector<f32> to vector<i32>
return %0 : vector<i32>
@@ -1467,8 +1469,6 @@ func.func @insert_scalar_into_vec_2d_f32_dynamic_idx(%arg0: vector<1x16xf32>, %a
// CHECK-LABEL: @insert_scalar_into_vec_2d_f32_dynamic_idx(
// CHECK: vector.insert
-// -----
-
func.func @insert_scalar_into_vec_2d_f32_dynamic_idx_scalable(%arg0: vector<1x[16]xf32>, %arg1: f32, %idx: index)
-> vector<1x[16]xf32> {
%0 = vector.insert %arg1, %arg0[0, %idx]: f32 into vector<1x[16]xf32>
@@ -1482,11 +1482,11 @@ func.func @insert_scalar_into_vec_2d_f32_dynamic_idx_scalable(%arg0: vector<1x[1
// -----
-func.func @vector_type_cast(%arg0: memref<8x8x8xf32>) -> memref<vector<8x8x8xf32>> {
+func.func @type_cast_f32(%arg0: memref<8x8x8xf32>) -> memref<vector<8x8x8xf32>> {
%0 = vector.type_cast %arg0: memref<8x8x8xf32> to memref<vector<8x8x8xf32>>
return %0 : memref<vector<8x8x8xf32>>
}
-// CHECK-LABEL: @vector_type_cast
+// CHECK-LABEL: @type_cast_f32
// CHECK: llvm.mlir.undef : !llvm.struct<(ptr, ptr, i64)>
// CHECK: %[[allocated:.*]] = llvm.extractvalue {{.*}}[0] : !llvm.struct<(ptr, ptr, i64, array<3 x i64>, array<3 x i64>)>
// CHECK: llvm.insertvalue %[[allocated]], {{.*}}[0] : !llvm.struct<(ptr, ptr, i64)>
@@ -1495,18 +1495,22 @@ func.func @vector_type_cast(%arg0: memref<8x8x8xf32>) -> memref<vector<8x8x8xf32
// CHECK: llvm.mlir.constant(0 : index
// CHECK: llvm.insertvalue {{.*}}[2] : !llvm.struct<(ptr, ptr, i64)>
+// NOTE: No test for scalable vectors - the input memref is fixed size.
+
// -----
-func.func @vector_index_type_cast(%arg0: memref<8x8x8xindex>) -> memref<vector<8x8x8xindex>> {
+func.func @type_cast_index(%arg0: memref<8x8x8xindex>) -> memref<vector<8x8x8xindex>> {
%0 = vector.type_cast %arg0: memref<8x8x8xindex> to memref<vector<8x8x8xindex>>
return %0 : memref<vector<8x8x8xindex>>
}
-// CHECK-LABEL: @vector_index_type_cast(
+// CHECK-LABEL: @type_cast_index(
// CHECK-SAME: %[[A:.*]]: memref<8x8x8xindex>)
// CHECK: %{{.*}} = builtin.unrealized_conversion_cast %[[A]] : memref<8x8x8xindex> to !llvm.struct<(ptr, ptr, i64, array<3 x i64>, array<3 x i64>)>
// CHECK: %{{.*}} = builtin.unrealized_conversion_cast %{{.*}} : !llvm.struct<(ptr, ptr, i64)> to memref<vector<8x8x8xindex>>
+// NOTE: No test for scalable vectors - the input memref is fixed size.
+
// -----
func.func @vector_type_cast_non_zero_addrspace(%arg0: memref<8x8x8xf32, 3>) -> memref<vector<8x8x8xf32>, 3> {
@@ -1522,16 +1526,18 @@ func.func @vector_type_cast_non_zero_addrspace(%arg0: memref<8x8x8xf32, 3>) -> m
// CHECK: llvm.mlir.constant(0 : index
// CHECK: llvm.insertvalue {{.*}}[2] : !llvm.struct<(ptr<3>, ptr<3>, i64)>
+// NOTE: No test for scalable vectors - the input memref is fixed size.
+
// -----
-func.func @vector_print_scalar_i1(%arg0: i1) {
+func.func @print_scalar_i1(%arg0: i1) {
vector.print %arg0 : i1
return
}
//
// Type "boolean" always uses zero extension.
//
-// CHECK-LABEL: @vector_print_scalar_i1(
+// CHECK-LABEL: @print_scalar_i1(
// CHECK-SAME: %[[A:.*]]: i1)
// CHECK: %[[S:.*]] = arith.extui %[[A]] : i1 to i64
// CHECK: llvm.call @printI64(%[[S]]) : (i64) -> ()
@@ -1539,11 +1545,11 @@ func.func @vector_print_scalar_i1(%arg0: i1) {
// -----
-func.func @vector_print_scalar_i4(%arg0: i4) {
+func.func @print_scalar_i4(%arg0: i4) {
vector.print %arg0 : i4
return
}
-// CHECK-LABEL: @vector_print_scalar_i4(
+// CHECK-LABEL: @print_scalar_i4(
// CHECK-SAME: %[[A:.*]]: i4)
// CHECK: %[[S:.*]] = arith.extsi %[[A]] : i4 to i64
// CHECK: llvm.call @printI64(%[[S]]) : (i64) -> ()
@@ -1551,11 +1557,11 @@ func.func @vector_print_scalar_i4(%arg0: i4) {
// -----
-func.func @vector_print_scalar_si4(%arg0: si4) {
+func.func @print_scalar_si4(%arg0: si4) {
vector.print %arg0 : si4
return
}
-// CHECK-LABEL: @vector_print_scalar_si4(
+// CHECK-LABEL: @print_scalar_si4(
// CHECK-SAME: %[[A:.*]]: si4)
// CHECK: %[[C:.*]] = builtin.unrealized_conversion_cast %[[A]] : si4 to i4
// CHECK: %[[S:.*]] = arith.extsi %[[C]] : i4 to i64
@@ -1564,11 +1570,11 @@ func.func @vector_print_scalar_si4(%arg0: si4) {
// -----
-func.func @vector_print_scalar_ui4(%arg0: ui4) {
+func.func @print_scalar_ui4(%arg0: ui4) {
vector.print %arg0 : ui4
return
}
-// CHECK-LABEL: @vector_print_scalar_ui4(
+// CHECK-LABEL: @print_scalar_ui4(
// CHECK-SAME: %[[A:.*]]: ui4)
// CHECK: %[[C:.*]] = builtin.unrealized_conversion_cast %[[A]] : ui4 to i4
// CHECK: %[[S:.*]] = arith.extui %[[C]] : i4 to i64
@@ -1577,11 +1583,11 @@ func.func @vector_print_scalar_ui4(%arg0: ui4) {
// -----
-func.func @vector_print_scalar_i32(%arg0: i32) {
+func.func @print_scalar_i32(%arg0: i32) {
vector.print %arg0 : i32
return
}
-// CHECK-LABEL: @vector_print_scalar_i32(
+// CHECK-LABEL: @print_scalar_i32(
// CHECK-SAME: %[[A:.*]]: i32)
// CHECK: %[[S:.*]] = arith.extsi %[[A]] : i32 to i64
// CHECK: llvm.call @printI64(%[[S]]) : (i64) -> ()
@@ -1589,11 +1595,11 @@ func.func @vector_print_scalar_i32(%arg0: i32) {
// -----
-func.func @vector_print_scalar_ui32(%arg0: ui32) {
+func.func @print_scalar_ui32(%arg0: ui32) {
vector.print %arg0 : ui32
return
}
-// CHECK-LABEL: @vector_print_scalar_ui32(
+// CHECK-LABEL: @print_scalar_ui32(
// CHECK-SAME: %[[A:.*]]: ui32)
// CHECK: %[[C:.*]] = builtin.unrealized_conversion_cast %[[A]] : ui32 to i32
// CHECK: %[[S:.*]] = arith.extui %[[C]] : i32 to i64
@@ -1601,11 +1607,11 @@ func.func @vector_print_scalar_ui32(%arg0: ui32) {
// -----
-func.func @vector_print_scalar_i40(%arg0: i40) {
+func.func @print_scalar_i40(%arg0: i40) {
vector.print %arg0 : i40
return
}
-// CHECK-LABEL: @vector_print_scalar_i40(
+// CHECK-LABEL: @print_scalar_i40(
// CHECK-SAME: %[[A:.*]]: i40)
// CHECK: %[[S:.*]] = arith.extsi %[[A]] : i40 to i64
// CHECK: llvm.call @printI64(%[[S]]) : (i64) -> ()
@@ -1613,11 +1619,11 @@ func.func @vector_print_scalar_i40(%arg0: i40) {
// -----
-func.func @vector_print_scalar_si40(%arg0: si40) {
+func.func @print_scalar_si40(%arg0: si40) {
vector.print %arg0 : si40
return
}
-// CHECK-LABEL: @vector_print_scalar_si40(
+// CHECK-LABEL: @print_scalar_si40(
// CHECK-SAME: %[[A:.*]]: si40)
// CHECK: %[[C:.*]] = builtin.unrealized_conversion_cast %[[A]] : si40 to i40
// CHECK: %[[S:.*]] = arith.extsi %[[C]] : i40 to i64
@@ -1626,11 +1632,11 @@ func.func @vector_print_scalar_si40(%arg0: si40) {
// -----
-func.func @vector_print_scalar_ui40(%arg0: ui40) {
+func.func @print_scalar_ui40(%arg0: ui40) {
vector.print %arg0 : ui40
return
}
-// CHECK-LABEL: @vector_print_scalar_ui40(
+// CHECK-LABEL: @print_scalar_ui40(
// CHECK-SAME: %[[A:.*]]: ui40)
// CHECK: %[[C:.*]] = builtin.unrealized_conversion_cast %[[A]] : ui40 to i40
// CHECK: %[[S:.*]] = arith.extui %[[C]] : i40 to i64
@@ -1639,22 +1645,22 @@ func.func @vector_print_scalar_ui40(%arg0: ui40) {
// -----
-func.func @vector_print_scalar_i64(%arg0: i64) {
+func.func @print_scalar_i64(%arg0: i64) {
vector.print %arg0 : i64
return
}
-// CHECK-LABEL: @vector_print_scalar_i64(
+// CHECK-LABEL: @print_scalar_i64(
// CHECK-SAME: %[[A:.*]]: i64)
// CHECK: llvm.call @printI64(%[[A]]) : (i64) -> ()
// CHECK: llvm.call @printNewline() : () -> ()
// -----
-func.func @vector_print_scalar_ui64(%arg0: ui64) {
+func.func @print_scalar_ui64(%arg0: ui64) {
vector.print %arg0 : ui64
return
}
-// CHECK-LABEL: @vector_print_scalar_ui64(
+// CHECK-LABEL: @print_scalar_ui64(
// CHECK-SAME: %[[A:.*]]: ui64)
// CHECK: %[[C:.*]] = builtin.unrealized_conversion_cast %[[A]] : ui64 to i64
// CHECK: llvm.call @printU64(%[[C]]) : (i64) -> ()
@@ -1662,11 +1668,11 @@ func.func @vector_print_scalar_ui64(%arg0: ui64) {
// -----
-func.func @vector_print_scalar_index(%arg0: index) {
+func.func @print_scalar_index(%arg0: index) {
vector.print %arg0 : index
return
}
-// CHECK-LABEL: @vector_print_scalar_index(
+// CHECK-LABEL: @print_scalar_index(
// CHECK-SAME: %[[A:.*]]: index)
// CHECK: %[[C:.*]] = builtin.unrealized_conversion_cast %[[A]] : index to i64
// CHECK: llvm.call @printU64(%[[C]]) : (i64) -> ()
@@ -1674,22 +1680,22 @@ func.func @vector_print_scalar_index(%arg0: index) {
// -----
-func.func @vector_print_scalar_f32(%arg0: f32) {
+func.func @print_scalar_f32(%arg0: f32) {
vector.print %arg0 : f32
return
}
-// CHECK-LABEL: @vector_print_scalar_f32(
+// CHECK-LABEL: @print_scalar_f32(
// CHECK-SAME: %[[A:.*]]: f32)
// CHECK: llvm.call @printF32(%[[A]]) : (f32) -> ()
// CHECK: llvm.call @printNewline() : () -> ()
// -----
-func.func @vector_print_scalar_f64(%arg0: f64) {
+func.func @print_scalar_f64(%arg0: f64) {
vector.print %arg0 : f64
return
}
-// CHECK-LABEL: @vector_print_scalar_f64(
+// CHECK-LABEL: @print_scalar_f64(
// CHECK-SAME: %[[A:.*]]: f64)
// CHECK: llvm.call @printF64(%[[A]]) : (f64) -> ()
// CHECK: llvm.call @printNewline() : () -> ()
@@ -1699,46 +1705,50 @@ func.func @vector_print_scalar_f64(%arg0: f64) {
// CHECK-LABEL: module {
// CHECK: llvm.func @printString(!llvm.ptr)
// CHECK: llvm.mlir.global private constant @[[GLOBAL_STR:.*]]({{.*}})
-// CHECK: @vector_print_string
+// CHECK: @print_string
// CHECK-NEXT: %[[GLOBAL_ADDR:.*]] = llvm.mlir.addressof @[[GLOBAL_STR]] : !llvm.ptr
// CHECK-NEXT: %[[STR_PTR:.*]] = llvm.getelementptr %[[GLOBAL_ADDR]][0] : (!llvm.ptr) -> !llvm.ptr
// CHECK-NEXT: llvm.call @printString(%[[STR_PTR]]) : (!llvm.ptr) -> ()
-func.func @vector_print_string() {
+func.func @print_string() {
vector.print str "Hello, World!"
return
}
// -----
-func.func @extract_strided_slice1(%arg0: vector<4xf32>) -> vector<2xf32> {
+func.func @extract_strided_slice_f32(%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_slice1(
+// CHECK-LABEL: @extract_strided_slice_f32(
// CHECK-SAME: %[[A:.*]]: vector<4xf32>)
// CHECK: %[[T0:.*]] = llvm.shufflevector %[[A]], %[[A]] [2, 3] : vector<4xf32>
// CHECK: return %[[T0]] : vector<2xf32>
+// NOTE: For scalable vectors we could only extract vector<[4]xf32> from vector<[4]xf32>, but that would be a NOP.
+
// -----
-func.func @extract_strided_index_slice1(%arg0: vector<4xindex>) -> vector<2xindex> {
+func.func @extract_strided_index_slice_index(%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_index_slice1(
+// CHECK-LABEL: @extract_strided_index_slice_index(
// 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>
// CHECK: %[[T3:.*]] = builtin.unrealized_conversion_cast %[[T2]] : vector<2xi64> to vector<2xindex>
// CHECK: return %[[T3]] : vector<2xindex>
+// NOTE: For scalable vectors we could only extract vector<[4]xindex> from vector<[4]xindex>, but that would be a NOP.
+
// -----
-func.func @extract_strided_slice2(%arg0: vector<4x8xf32>) -> vector<2x8xf32> {
+func.func @extract_strided_slice_f32_1d_from_2d(%arg0: vector<4x8xf32>) -> vector<2x8xf32> {
%0 = vector.extract_strided_slice %arg0 {offsets = [2], sizes = [2], strides = [1]} : vector<4x8xf32> to vector<2x8xf32>
return %0 : vector<2x8xf32>
}
-// CHECK-LABEL: @extract_strided_slice2(
+// CHECK-LABEL: @extract_strided_slice_f32_1d_from_2d(
// CHECK-SAME: %[[ARG:.*]]: vector<4x8xf32>)
// CHECK: %[[A:.*]] = builtin.unrealized_conversion_cast %[[ARG]] : vector<4x8xf32> to !llvm.array<4 x vector<8xf32>>
// CHECK: %[[T0:.*]] = llvm.mlir.undef : !llvm.array<2 x vector<8xf32>>
@@ -1749,13 +1759,28 @@ func.func @extract_strided_slice2(%arg0: vector<4x8xf32>) -> vector<2x8xf32> {
// CHECK: %[[T5:.*]] = builtin.unrealized_conversion_cast %[[T4]] : !llvm.array<2 x vector<8xf32>> to vector<2x8xf32>
// CHECK: return %[[T5]]
+func.func @extract_strided_slice_f32_1d_from_2d_scalable(%arg0: vector<4x[8]xf32>) -> vector<2x[8]xf32> {
+ %0 = vector.extract_strided_slice %arg0 {offsets = [2], sizes = [2], strides = [1]} : vector<4x[8]xf32> to vector<2x[8]xf32>
+ return %0 : vector<2x[8]xf32>
+}
+// CHECK-LABEL: func.func @extract_strided_slice_f32_1d_from_2d_scalable(
+// CHECK-SAME: %[[ARG:.*]]: vector<4x[8]xf32>)
+// CHECK: %[[A:.*]] = builtin.unrealized_conversion_cast %[[ARG]] : vector<4x[8]xf32> to !llvm.array<4 x vector<[8]xf32>>
+// CHECK: %[[T0:.*]] = llvm.mlir.undef : !llvm.array<2 x vector<[8]xf32>>
+// CHECK: %[[T1:.*]] = llvm.extractvalue %[[A]][2] : !llvm.array<4 x vector<[8]xf32>>
+// CHECK: %[[T2:.*]] = llvm.insertvalue %[[T1]], %[[T0]][0] : !llvm.array<2 x vector<[8]xf32>>
+// CHECK: %[[T3:.*]] = llvm.extractvalue %[[A]][3] : !llvm.array<4 x vector<[8]xf32>>
+// CHECK: %[[T4:.*]] = llvm.insertvalue %[[T3]], %[[T2]][1] : !llvm.array<2 x vector<[8]xf32>>
+// CHECK: %[[T5:.*]] = builtin.unrealized_conversion_cast %[[T4]] : !llvm.array<2 x vector<[8]xf32>> to vector<2x[8]xf32>
+// CHECK: return %[[T5]]
+
// -----
-func.func @extract_strided_slice3(%arg0: vector<4x8xf32>) -> vector<2x2xf32> {
+func.func @extract_strided_slice_f32_2d_from_2d(%arg0: vector<4x8xf32>) -> vector<2x2xf32> {
%0 = vector.extract_strided_slice %arg0 {offsets = [2, 2], sizes = [2, 2], strides = [1, 1]} : vector<4x8xf32> to vector<2x2xf32>
return %0 : vector<2x2xf32>
}
-// CHECK-LABEL: @extract_strided_slice3(
+// CHECK-LABEL: @extract_strided_slice_f32_2d_from_2d(
// CHECK-SAME: %[[ARG:.*]]: vector<4x8xf32>)
// CHECK: %[[A:.*]] = builtin.unrealized_conversion_cast %[[ARG]] : vector<4x8xf32> to !llvm.array<4 x vector<8xf32>>
// CHECK: %[[VAL_2:.*]] = arith.constant dense<0.000000e+00> : vector<2x2xf32>
@@ -1769,27 +1794,25 @@ func.func @extract_strided_slice3(%arg0: vector<4x8xf32>) -> vector<2x2xf32> {
// CHECK: %[[VAL_12:.*]] = builtin.unrealized_conversion_cast %[[T7]] : !llvm.array<2 x vector<2xf32>> to vector<2x2xf32>
// CHECK: return %[[VAL_12]] : vector<2x2xf32>
-// -----
-
-func.func @extract_strided_slice_scalable(%arg0 : vector<1x4x[4]xi32>) -> vector<1x1x[4]xi32> {
- %0 = vector.extract_strided_slice %arg0 {offsets = [0, 3, 0], sizes = [1, 1, 4], strides = [1, 1, 1]} : vector<1x4x[4]xi32> to vector<1x1x[4]xi32>
- return %0 : vector<1x1x[4]xi32>
-}
-
-// CHECK-LABEL: func.func @extract_strided_slice_scalable(
-// CHECK-SAME: %[[ARG_0:.*]]: vector<1x4x[4]xi32>) -> vector<1x1x[4]xi32> {
-
-// CHECK: %[[CAST_1:.*]] = builtin.unrealized_conversion_cast %[[ARG_0]] : vector<1x4x[4]xi32> to !llvm.array<1 x array<4 x vector<[4]xi32>>>
-// CHECK: %[[CST:.*]] = arith.constant dense<0> : vector<1x1x[4]xi32>
-// CHECK: %[[CAST_2:.*]] = builtin.unrealized_conversion_cast %[[CST]] : vector<1x1x[4]xi32> to !llvm.array<1 x array<1 x vector<[4]xi32>>>
-// CHECK: %[[CST_1:.*]] = arith.constant dense<0> : vector<1x[4]xi32>
-// CHECK: %[[CAST_3:.*]] = builtin.unrealized_conversion_cast %[[CST_1]] : vector<1x[4]xi32> to !llvm.array<1 x vector<[4]xi32>>
-
-// CHECK: %[[EXT:.*]] = llvm.extractvalue %[[CAST_1]][0, 3] : !llvm.array<1 x array<4 x vector<[4]xi32>>>
-// CHECK: %[[INS_1:.*]] = llvm.insertvalue %[[EXT]], %[[CAST_3]][0] : !llvm.array<1 x vector<[4]xi32>>
-// CHECK: %[[INS_2:.*]] = llvm.insertvalue %[[INS_1]], %[[CAST_2]][0] : !llvm.array<1 x array<1 x vector<[4]xi32>>>
-
-// CHECK: builtin.unrealized_conversion_cast %[[INS_2]] : !llvm.array<1 x array<1 x vector<[4]xi32>>> to vector<1x1x[4]xi32>
+// NOTE: For scalable vectors, we can only extract "full" scalable dimensions
+// (e.g. [8] from [8], but not [4] from [8]).
+
+func.func @extract_strided_slice_f32_2d_from_2d_scalable(%arg0: vector<4x[8]xf32>) -> vector<2x[8]xf32> {
+ %0 = vector.extract_strided_slice %arg0 {offsets = [2, 0], sizes = [2, 8], strides = [1, 1]} : vector<4x[8]xf32> to vector<2x[8]xf32>
+ return %0 : vector<2x[8]xf32>
+}
+// CHECK-LABEL: @extract_strided_slice_f32_2d_from_2d_scalable(
+// CHECK-SAME: %[[ARG:.*]]: vector<4x[8]xf32>)
+// CHECK: %[[T1:.*]] = builtin.unrealized_conversion_cast %[[ARG]] : vector<4x[8]xf32> to !llvm.array<4 x vector<[8]xf32>>
+// CHECK: %[[T2:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[T3:.*]] = arith.constant dense<0.000000e+00> : vector<2x[8]xf32>
+// CHECK: %[[T4:.*]] = builtin.unrealized_conversion_cast %[[T3]] : vector<2x[8]xf32> to !llvm.array<2 x vector<[8]xf32>>
+// CHECK: %[[T5:.*]] = llvm.extractvalue %[[T1]][2] : !llvm.array<4 x vector<[8]xf32>>
+// CHECK: %[[T6:.*]] = llvm.insertvalue %[[T5]], %[[T4]][0] : !llvm.array<2 x vector<[8]xf32>>
+// CHECK: %[[T7:.*]] = llvm.extractvalue %[[T1]][3] : !llvm.array<4 x vector<[8]xf32>>
+// CHECK: %[[T8:.*]] = llvm.insertvalue %[[T7]], %[[T6]][1] : !llvm.array<2 x vector<[8]xf32>>
+// CHECK: %[[T9:.*]] = builtin.unrealized_conversion_cast %[[T8]] : !llvm.array<2 x vector<[8]xf32>> to vector<2x[8]xf32>
+// CHECK: return %[[T9]] : vector<2x[8]xf32>
// -----
More information about the Mlir-commits
mailing list