[Mlir-commits] [mlir] [mlir][memref] Improve `memref.subview` type inference (PR #96421)
Matthias Springer
llvmlistbot at llvm.org
Sun Jun 23 01:50:15 PDT 2024
https://github.com/matthias-springer created https://github.com/llvm/llvm-project/pull/96421
The `memref.subview` result type inference (`SubViewOp::inferResultType`) sometimes used to produce a dynamic offset when a static offset is possible.
When a dynamic value (stride, size, etc.) is multiplied with zero, the result is always a "static 0". Based on this, the result type inference implementation can be improved to produce more static type information in memref types.
>From fe160460d5d365e719a961ba02d5064ea60b3faf Mon Sep 17 00:00:00 2001
From: Matthias Springer <mspringer at nvidia.com>
Date: Sun, 23 Jun 2024 10:41:37 +0200
Subject: [PATCH] [mlir][memref] Improve `memref.subview` type inference
The `memref.subview` result type inference (`SubViewOp::inferResultType`) sometimes used to produce a dynamic offset when a static offset is possible.
The new result type inference implementation produces memref types with more static type information.
---
.../mlir/Dialect/Utils/StaticValueUtils.h | 6 +++
mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp | 4 +-
mlir/test/Dialect/Linalg/promote.mlir | 42 +++++++++----------
.../Dialect/Linalg/transform-promotion.mlir | 12 +++---
mlir/test/Dialect/MemRef/canonicalize.mlir | 17 ++++----
mlir/test/Dialect/MemRef/subview.mlir | 4 +-
...tor-transfer-collapse-inner-most-dims.mlir | 16 +++----
.../Dialect/Vector/vector-transferop-opt.mlir | 10 ++---
mlir/test/Transforms/canonicalize.mlir | 10 ++---
9 files changed, 63 insertions(+), 58 deletions(-)
diff --git a/mlir/include/mlir/Dialect/Utils/StaticValueUtils.h b/mlir/include/mlir/Dialect/Utils/StaticValueUtils.h
index 594bcf5dbb399..ba4f084d3efd1 100644
--- a/mlir/include/mlir/Dialect/Utils/StaticValueUtils.h
+++ b/mlir/include/mlir/Dialect/Utils/StaticValueUtils.h
@@ -200,6 +200,12 @@ struct SaturatedInteger {
return SaturatedInteger{false, other.v + v};
}
SaturatedInteger operator*(SaturatedInteger other) {
+ // Multiplication with 0 is always 0.
+ if (!other.saturated && other.v == 0)
+ return SaturatedInteger{false, 0};
+ if (!saturated && v == 0)
+ return SaturatedInteger{false, 0};
+ // Otherwise, if this or the other integer is dynamic, so is the result.
if (saturated || other.saturated)
return SaturatedInteger{true, 0};
return SaturatedInteger{false, other.v * v};
diff --git a/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp b/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
index d70e6d0b79cd6..f9e8a27973178 100644
--- a/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
+++ b/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
@@ -2691,10 +2691,10 @@ Type SubViewOp::inferResultType(MemRefType sourceMemRefType,
// `sourceOffset + sum_i(staticOffset_i * sourceStrides_i)`.
int64_t targetOffset = sourceOffset;
for (auto it : llvm::zip(staticOffsets, sourceStrides)) {
- auto staticOffset = std::get<0>(it), targetStride = std::get<1>(it);
+ auto staticOffset = std::get<0>(it), sourceStride = std::get<1>(it);
targetOffset = (SaturatedInteger::wrap(targetOffset) +
SaturatedInteger::wrap(staticOffset) *
- SaturatedInteger::wrap(targetStride))
+ SaturatedInteger::wrap(sourceStride))
.asInteger();
}
diff --git a/mlir/test/Dialect/Linalg/promote.mlir b/mlir/test/Dialect/Linalg/promote.mlir
index 2d640057df340..00b8c649b82c3 100644
--- a/mlir/test/Dialect/Linalg/promote.mlir
+++ b/mlir/test/Dialect/Linalg/promote.mlir
@@ -42,23 +42,23 @@ func.func @matmul_f32(%A: memref<?xi8>, %M: index, %N: index, %K: index) {
///
// CHECK: %[[tmpA:.*]] = memref.alloca() : memref<32xi8>
// CHECK: %[[fullA:.*]] = memref.view %[[tmpA]][{{.*}}][{{.*}}] : memref<32xi8> to memref<?x?xf32>
-// CHECK: %[[partialA:.*]] = memref.subview %[[fullA]]{{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
+// CHECK: %[[partialA:.*]] = memref.subview %[[fullA]]{{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
///
// CHECK: %[[tmpB:.*]] = memref.alloca() : memref<48xi8>
// CHECK: %[[fullB:.*]] = memref.view %[[tmpB]][{{.*}}][{{.*}}] : memref<48xi8> to memref<?x?xf32>
-// CHECK: %[[partialB:.*]] = memref.subview %[[fullB]]{{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
+// CHECK: %[[partialB:.*]] = memref.subview %[[fullB]]{{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
///
// CHECK: %[[tmpC:.*]] = memref.alloca() : memref<24xi8>
// CHECK: %[[fullC:.*]] = memref.view %[[tmpC]][{{.*}}][{{.*}}] : memref<24xi8> to memref<?x?xf32>
-// CHECK: %[[partialC:.*]] = memref.subview %[[fullC]]{{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
+// CHECK: %[[partialC:.*]] = memref.subview %[[fullC]]{{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
-// CHECK: linalg.copy ins(%[[vA]] : memref<?x?xf32, strided<[?, 1], offset: ?>>) outs(%[[partialA]] : memref<?x?xf32, strided<[?, 1], offset: ?>>)
-// CHECK: linalg.copy ins(%[[vB]] : memref<?x?xf32, strided<[?, 1], offset: ?>>) outs(%[[partialB]] : memref<?x?xf32, strided<[?, 1], offset: ?>>)
-// CHECK: linalg.copy ins(%[[vC]] : memref<?x?xf32, strided<[?, 1], offset: ?>>) outs(%[[partialC]] : memref<?x?xf32, strided<[?, 1], offset: ?>>)
+// CHECK: linalg.copy ins(%[[vA]] : memref<?x?xf32, strided<[?, 1], offset: ?>>) outs(%[[partialA]] : memref<?x?xf32, strided<[?, 1]>>)
+// CHECK: linalg.copy ins(%[[vB]] : memref<?x?xf32, strided<[?, 1], offset: ?>>) outs(%[[partialB]] : memref<?x?xf32, strided<[?, 1]>>)
+// CHECK: linalg.copy ins(%[[vC]] : memref<?x?xf32, strided<[?, 1], offset: ?>>) outs(%[[partialC]] : memref<?x?xf32, strided<[?, 1]>>)
//
// CHECK: linalg.matmul ins(%[[partialA]], %[[partialB]]{{.*}} outs(%[[partialC]]
//
-// CHECK: linalg.copy ins(%[[partialC]] : memref<?x?xf32, strided<[?, 1], offset: ?>>) outs(%[[vC]] : memref<?x?xf32, strided<[?, 1], offset: ?>>)
+// CHECK: linalg.copy ins(%[[partialC]] : memref<?x?xf32, strided<[?, 1]>>) outs(%[[vC]] : memref<?x?xf32, strided<[?, 1], offset: ?>>)
//
// CHECK-NOT: memref.dealloc %[[tmpA]] : memref<32xi8>
// CHECK-NOT: memref.dealloc %[[tmpB]] : memref<48xi8>
@@ -112,23 +112,23 @@ func.func @matmul_f64(%A: memref<?xi8>, %M: index, %N: index, %K: index) {
///
// CHECK: %[[tmpA_f64:.*]] = memref.alloc() : memref<64xi8>
// CHECK: %[[fullA_f64:.*]] = memref.view %[[tmpA_f64]][{{.*}}][{{.*}}] : memref<64xi8> to memref<?x?xf64>
-// CHECK: %[[partialA_f64:.*]] = memref.subview %[[fullA_f64]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1], offset: ?>>
+// CHECK: %[[partialA_f64:.*]] = memref.subview %[[fullA_f64]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1]>>
///
// CHECK: %[[tmpB_f64:.*]] = memref.alloc() : memref<96xi8>
// CHECK: %[[fullB_f64:.*]] = memref.view %[[tmpB_f64]][{{.*}}][{{.*}}] : memref<96xi8> to memref<?x?xf64>
-// CHECK: %[[partialB_f64:.*]] = memref.subview %[[fullB_f64]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1], offset: ?>>
+// CHECK: %[[partialB_f64:.*]] = memref.subview %[[fullB_f64]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1]>>
///
// CHECK: %[[tmpC_f64:.*]] = memref.alloc() : memref<48xi8>
// CHECK: %[[fullC_f64:.*]] = memref.view %[[tmpC_f64]][{{.*}}][{{.*}}] : memref<48xi8> to memref<?x?xf64>
-// CHECK: %[[partialC_f64:.*]] = memref.subview %[[fullC_f64]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1], offset: ?>>
+// CHECK: %[[partialC_f64:.*]] = memref.subview %[[fullC_f64]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1]>>
-// CHECK: linalg.copy ins(%[[vA_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>) outs(%[[partialA_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>)
-// CHECK: linalg.copy ins(%[[vB_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>) outs(%[[partialB_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>)
-// CHECK: linalg.copy ins(%[[vC_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>) outs(%[[partialC_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>)
+// CHECK: linalg.copy ins(%[[vA_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>) outs(%[[partialA_f64]] : memref<?x?xf64, strided<[?, 1]>>)
+// CHECK: linalg.copy ins(%[[vB_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>) outs(%[[partialB_f64]] : memref<?x?xf64, strided<[?, 1]>>)
+// CHECK: linalg.copy ins(%[[vC_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>) outs(%[[partialC_f64]] : memref<?x?xf64, strided<[?, 1]>>)
//
// CHECK: linalg.matmul ins(%[[partialA_f64]], %[[partialB_f64]]{{.*}} outs(%[[partialC_f64]]
//
-// CHECK: linalg.copy ins(%[[partialC_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>) outs(%[[vC_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>)
+// CHECK: linalg.copy ins(%[[partialC_f64]] : memref<?x?xf64, strided<[?, 1]>>) outs(%[[vC_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>)
//
// CHECK: memref.dealloc %[[tmpA_f64]] : memref<64xi8>
// CHECK: memref.dealloc %[[tmpB_f64]] : memref<96xi8>
@@ -318,7 +318,7 @@ func.func @linalg_generic_update_all_function_inputs_outputs(%arg0: memref<3x4xf
// CHECK: %[[VAL_21:.*]] = arith.constant 12 : index
// CHECK: %[[VAL_22:.*]] = memref.alloc() : memref<48xi8, #gpu.address_space<workgroup>>
// CHECK: %[[VAL_23:.*]] = memref.view %[[VAL_22]]{{\[}}%[[VAL_18]]]{{\[}}%[[VAL_12]], %[[VAL_15]]] : memref<48xi8, #gpu.address_space<workgroup>> to memref<?x?xf32, #gpu.address_space<workgroup>>
- // CHECK: %[[VAL_24:.*]] = memref.subview %[[VAL_23]][0, 0] {{\[}}%[[VAL_14]], %[[VAL_17]]] [1, 1] : memref<?x?xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[?, 1], offset: ?>, #gpu.address_space<workgroup>>
+ // CHECK: %[[VAL_24:.*]] = memref.subview %[[VAL_23]][0, 0] {{\[}}%[[VAL_14]], %[[VAL_17]]] [1, 1] : memref<?x?xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>
// CHECK: %[[VAL_25:.*]] = arith.constant 0 : index
// CHECK: %[[VAL_26:.*]] = arith.constant 4 : index
// CHECK: %[[VAL_27:.*]] = arith.constant 1 : index
@@ -337,7 +337,7 @@ func.func @linalg_generic_update_all_function_inputs_outputs(%arg0: memref<3x4xf
// CHECK: %[[VAL_40:.*]] = arith.constant 12 : index
// CHECK: %[[VAL_41:.*]] = memref.alloc() : memref<48xi8, #gpu.address_space<workgroup>>
// CHECK: %[[VAL_42:.*]] = memref.view %[[VAL_41]]{{\[}}%[[VAL_37]]]{{\[}}%[[VAL_31]], %[[VAL_34]]] : memref<48xi8, #gpu.address_space<workgroup>> to memref<?x?xf32, #gpu.address_space<workgroup>>
- // CHECK: %[[VAL_43:.*]] = memref.subview %[[VAL_42]][0, 0] {{\[}}%[[VAL_33]], %[[VAL_36]]] [1, 1] : memref<?x?xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[?, 1], offset: ?>, #gpu.address_space<workgroup>>
+ // CHECK: %[[VAL_43:.*]] = memref.subview %[[VAL_42]][0, 0] {{\[}}%[[VAL_33]], %[[VAL_36]]] [1, 1] : memref<?x?xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>
// CHECK: %[[VAL_44:.*]] = arith.constant 0 : index
// CHECK: %[[VAL_45:.*]] = arith.constant 4 : index
// CHECK: %[[VAL_46:.*]] = arith.constant 1 : index
@@ -356,10 +356,10 @@ func.func @linalg_generic_update_all_function_inputs_outputs(%arg0: memref<3x4xf
// CHECK: %[[VAL_59:.*]] = arith.constant 12 : index
// CHECK: %[[VAL_60:.*]] = memref.alloc() : memref<48xi8, #gpu.address_space<workgroup>>
// CHECK: %[[VAL_61:.*]] = memref.view %[[VAL_60]]{{\[}}%[[VAL_56]]]{{\[}}%[[VAL_50]], %[[VAL_53]]] : memref<48xi8, #gpu.address_space<workgroup>> to memref<?x?xf32, #gpu.address_space<workgroup>>
- // CHECK: %[[VAL_62:.*]] = memref.subview %[[VAL_61]][0, 0] {{\[}}%[[VAL_52]], %[[VAL_55]]] [1, 1] : memref<?x?xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[?, 1], offset: ?>, #gpu.address_space<workgroup>>
-// CHECK: linalg.copy ins(%[[VAL_3]] : memref<4x3xf32, strided<[4, 1]>, 1>) outs(%[[VAL_24]] : memref<?x?xf32, strided<[?, 1], offset: ?>, #gpu.address_space<workgroup>>)
-// CHECK: linalg.copy ins(%[[VAL_4]] : memref<4x3xf32, strided<[4, 1]>, 1>) outs(%[[VAL_43]] : memref<?x?xf32, strided<[?, 1], offset: ?>, #gpu.address_space<workgroup>>)
- // CHECK: linalg.generic {doc = "", indexing_maps = [#map, #map, #map], iterator_types = ["parallel", "parallel"], library_call = ""} ins(%[[VAL_24]], %[[VAL_43]] : memref<?x?xf32, strided<[?, 1], offset: ?>, #gpu.address_space<workgroup>>, memref<?x?xf32, strided<[?, 1], offset: ?>, #gpu.address_space<workgroup>>) outs(%[[VAL_62]] : memref<?x?xf32, strided<[?, 1], offset: ?>, #gpu.address_space<workgroup>>) {
+ // CHECK: %[[VAL_62:.*]] = memref.subview %[[VAL_61]][0, 0] {{\[}}%[[VAL_52]], %[[VAL_55]]] [1, 1] : memref<?x?xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>
+// CHECK: linalg.copy ins(%[[VAL_3]] : memref<4x3xf32, strided<[4, 1]>, 1>) outs(%[[VAL_24]] : memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>)
+// CHECK: linalg.copy ins(%[[VAL_4]] : memref<4x3xf32, strided<[4, 1]>, 1>) outs(%[[VAL_43]] : memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>)
+ // CHECK: linalg.generic {doc = "", indexing_maps = [#map, #map, #map], iterator_types = ["parallel", "parallel"], library_call = ""} ins(%[[VAL_24]], %[[VAL_43]] : memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>, memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>) outs(%[[VAL_62]] : memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>) {
// CHECK: ^bb0(%[[VAL_63:.*]]: f32, %[[VAL_64:.*]]: f32, %[[VAL_65:.*]]: f32):
// CHECK: %[[VAL_66:.*]] = arith.addf %[[VAL_63]], %[[VAL_64]] : f32
// CHECK: linalg.yield %[[VAL_66]] : f32
@@ -372,7 +372,7 @@ func.func @linalg_generic_update_all_function_inputs_outputs(%arg0: memref<3x4xf
linalg.yield %1 : f32
}
- // CHECK: linalg.copy ins(%[[VAL_62]] : memref<?x?xf32, strided<[?, 1], offset: ?>, #gpu.address_space<workgroup>>) outs(%[[VAL_5]] : memref<4x3xf32, strided<[4, 1]>, 1>)
+ // CHECK: linalg.copy ins(%[[VAL_62]] : memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>) outs(%[[VAL_5]] : memref<4x3xf32, strided<[4, 1]>, 1>)
// CHECK: memref.dealloc %[[VAL_22]] : memref<48xi8, #gpu.address_space<workgroup>>
// CHECK: memref.dealloc %[[VAL_41]] : memref<48xi8, #gpu.address_space<workgroup>>
// CHECK: memref.dealloc %[[VAL_60]] : memref<48xi8, #gpu.address_space<workgroup>>
diff --git a/mlir/test/Dialect/Linalg/transform-promotion.mlir b/mlir/test/Dialect/Linalg/transform-promotion.mlir
index d6112db0f7772..7c4cd623c742d 100644
--- a/mlir/test/Dialect/Linalg/transform-promotion.mlir
+++ b/mlir/test/Dialect/Linalg/transform-promotion.mlir
@@ -42,15 +42,15 @@ func.func @promote_subview_matmul(%arg0: memref<?x?xf32, strided<[?, 1], offset:
// CHECK: %[[a0:.*]] = memref.alloc() : memref<32000000xi8>
// CHECK: %[[v0:.*]] = memref.view %[[a0]]{{.*}} : memref<32000000xi8> to memref<?x?xf32>
// CHECK: %[[l0:.*]] = memref.subview %[[v0]][0, 0] [%{{.*}}, %{{.*}}] [1, 1]
-// CHECK-SAME: memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
+// CHECK-SAME: memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
// CHECK: %[[a1:.*]] = memref.alloc() : memref<48000000xi8>
// CHECK: %[[v1:.*]] = memref.view %[[a1]]{{.*}} : memref<48000000xi8> to memref<?x?xf32>
// CHECK: %[[l1:.*]] = memref.subview %[[v1]][0, 0] [%{{.*}}, %{{.*}}] [1, 1]
-// CHECK-SAME: memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
+// CHECK-SAME: memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
// CHECK: %[[a2:.*]] = memref.alloc() : memref<24000000xi8>
// CHECK: %[[v2:.*]] = memref.view %[[a2]]{{.*}} : memref<24000000xi8> to memref<?x?xf32>
// CHECK: %[[l2:.*]] = memref.subview %[[v2]][0, 0] [%{{.*}}, %{{.*}}] [1, 1]
-// CHECK-SAME: memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
+// CHECK-SAME: memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
// CHECK: linalg.copy ins(%[[s0]] : memref<?x?xf32, strided{{.*}}>) outs(%[[l0]] : memref<?x?xf32, strided{{.*}}>)
// CHECK: linalg.copy ins(%[[s1]] : memref<?x?xf32, strided{{.*}}>) outs(%[[l1]] : memref<?x?xf32, strided{{.*}}>)
// CHECK: linalg.copy ins(%[[s2]] : memref<?x?xf32, strided{{.*}}>) outs(%[[l2]] : memref<?x?xf32, strided{{.*}}>)
@@ -110,7 +110,7 @@ func.func @promote_first_subview_matmul(%arg0: memref<?x?xf32, strided<[?, 1], o
// CHECK: %[[s2:.*]] = memref.subview {{.*}}: memref<?x?xf32, strided{{.*}}> to memref<?x?xf32, strided{{.*}}>
// CHECK: %[[a0:.*]] = memref.alloc() : memref<32000000xi8>
// CHECK: %[[v0:.*]] = memref.view %[[a0]]{{.*}} : memref<32000000xi8> to memref<?x?xf32>
-// CHECK: %[[l0:.*]] = memref.subview %[[v0]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
+// CHECK: %[[l0:.*]] = memref.subview %[[v0]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
// CHECK-NOT: memref.alloc
// CHECK-NOT: memref.view
// CHECK-NOT: memref.subview
@@ -147,7 +147,7 @@ func.func @aligned_promote_fill(%arg0: memref<?x?xf32, strided<[?, 1], offset: ?
// CHECK: %[[s0:.*]] = memref.subview {{.*}}: memref<?x?xf32, strided{{.*}}> to memref<?x?xf32, strided{{.*}}>
// CHECK: %[[a0:.*]] = memref.alloc() {alignment = 32 : i64} : memref<32000000xi8>
// CHECK: %[[v0:.*]] = memref.view %[[a0]]{{.*}} : memref<32000000xi8> to memref<?x?xf32>
-// CHECK: %[[l0:.*]] = memref.subview %[[v0]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
+// CHECK: %[[l0:.*]] = memref.subview %[[v0]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
// CHECK: linalg.fill ins({{.*}} : f32) outs(%[[v0]] : memref<?x?xf32>)
// CHECK: linalg.copy ins(%[[s0]] : memref<?x?xf32, strided{{.*}}>) outs(%[[l0]] : memref<?x?xf32, strided{{.*}}>)
// CHECK: linalg.fill ins(%[[cf]] : f32) outs(%[[v0]] : memref<?x?xf32>)
@@ -180,7 +180,7 @@ func.func @aligned_promote_fill_complex(%arg0: memref<?x?xcomplex<f32>, strided<
// CHECK: %[[s0:.*]] = memref.subview {{.*}}: memref<?x?xcomplex<f32>, strided{{.*}}> to memref<?x?xcomplex<f32>, strided{{.*}}>
// CHECK: %[[a0:.*]] = memref.alloc() {alignment = 32 : i64} : memref<64000000xi8>
// CHECK: %[[v0:.*]] = memref.view %[[a0]]{{.*}} : memref<64000000xi8> to memref<?x?xcomplex<f32>>
-// CHECK: %[[l0:.*]] = memref.subview %[[v0]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xcomplex<f32>> to memref<?x?xcomplex<f32>, strided<[?, 1], offset: ?>>
+// CHECK: %[[l0:.*]] = memref.subview %[[v0]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xcomplex<f32>> to memref<?x?xcomplex<f32>, strided<[?, 1]>>
// CHECK: linalg.fill ins({{.*}} : complex<f32>) outs(%[[v0]] : memref<?x?xcomplex<f32>>)
// CHECK: linalg.copy ins(%[[s0]] : memref<?x?xcomplex<f32>, strided{{.*}}>) outs(%[[l0]] : memref<?x?xcomplex<f32>, strided{{.*}}>)
// CHECK: linalg.fill ins(%[[cc]] : complex<f32>) outs(%[[v0]] : memref<?x?xcomplex<f32>>)
diff --git a/mlir/test/Dialect/MemRef/canonicalize.mlir b/mlir/test/Dialect/MemRef/canonicalize.mlir
index c4ff6480a4ce5..b15af9baca7dc 100644
--- a/mlir/test/Dialect/MemRef/canonicalize.mlir
+++ b/mlir/test/Dialect/MemRef/canonicalize.mlir
@@ -32,15 +32,14 @@ func.func @collapse_expand_rank0_cancel(%arg0 : memref<1x1xi8>) -> memref<1x1xi8
// CHECK-LABEL: func @subview_of_size_memcast
// CHECK-SAME: %[[ARG0:.[a-z0-9A-Z_]+]]: memref<4x6x16x32xi8>
// CHECK: %[[S:.+]] = memref.subview %[[ARG0]][0, 1, 0, 0] [1, 1, 16, 32] [1, 1, 1, 1] : memref<4x6x16x32xi8> to memref<16x32xi8, strided{{.*}}>
-// CHECK: %[[M:.+]] = memref.cast %[[S]] : memref<16x32xi8, {{.*}}> to memref<16x32xi8, strided{{.*}}>
-// CHECK: return %[[M]] : memref<16x32xi8, strided{{.*}}>
+// CHECK: return %[[S]] : memref<16x32xi8, strided{{.*}}>
func.func @subview_of_size_memcast(%arg : memref<4x6x16x32xi8>) ->
- memref<16x32xi8, strided<[32, 1], offset: ?>>{
+ memref<16x32xi8, strided<[32, 1], offset: 512>>{
%0 = memref.cast %arg : memref<4x6x16x32xi8> to memref<?x?x16x32xi8>
%1 = memref.subview %0[0, 1, 0, 0] [1, 1, 16, 32] [1, 1, 1, 1] :
memref<?x?x16x32xi8> to
- memref<16x32xi8, strided<[32, 1], offset: ?>>
- return %1 : memref<16x32xi8, strided<[32, 1], offset: ?>>
+ memref<16x32xi8, strided<[32, 1], offset: 512>>
+ return %1 : memref<16x32xi8, strided<[32, 1], offset: 512>>
}
// -----
@@ -643,13 +642,13 @@ func.func @no_fold_subview_with_non_unit_stride(%arg0 : memref<20x42xf32>) -> me
// -----
-func.func @no_fold_dynamic_no_op_subview(%arg0 : memref<?x?xf32>) -> memref<?x?xf32, strided<[?, 1], offset: ?>> {
+func.func @no_fold_dynamic_no_op_subview(%arg0 : memref<?x?xf32>) -> memref<?x?xf32, strided<[?, 1]>> {
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%0 = memref.dim %arg0, %c0 : memref<?x?xf32>
%1 = memref.dim %arg0, %c1 : memref<?x?xf32>
- %2 = memref.subview %arg0[0, 0] [%0, %1] [1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
- return %2 : memref<?x?xf32, strided<[?, 1], offset: ?>>
+ %2 = memref.subview %arg0[0, 0] [%0, %1] [1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
+ return %2 : memref<?x?xf32, strided<[?, 1]>>
}
// CHECK-LABEL: func @no_fold_dynamic_no_op_subview(
// CHECK: %[[SUBVIEW:.+]] = memref.subview
@@ -969,7 +968,7 @@ func.func @canonicalize_rank_reduced_subview(%arg0 : memref<8x?xf32>,
// CHECK-SAME: %[[ARG0:.+]]: memref<8x?xf32>
// CHECK-SAME: %[[ARG1:.+]]: index
// CHECK: %[[SUBVIEW:.+]] = memref.subview %[[ARG0]][0, 0] [1, %[[ARG1]]] [1, 1]
-// CHECK-SAME: memref<8x?xf32> to memref<?xf32, strided<[1], offset: ?>>
+// CHECK-SAME: memref<8x?xf32> to memref<?xf32, strided<[1]>>
// -----
diff --git a/mlir/test/Dialect/MemRef/subview.mlir b/mlir/test/Dialect/MemRef/subview.mlir
index 2df3e24942d70..135a1124066e4 100644
--- a/mlir/test/Dialect/MemRef/subview.mlir
+++ b/mlir/test/Dialect/MemRef/subview.mlir
@@ -47,9 +47,9 @@ func.func @memref_subview(%arg0 : index, %arg1 : index, %arg2 : index) {
%7 = memref.alloc(%arg1, %arg2) : memref<?x?xf32>
// CHECK: memref.subview {{%.*}}[0, 0] [4, 4] [1, 1] :
// CHECK-SAME: memref<?x?xf32>
- // CHECK-SAME: to memref<4x4xf32, strided<[?, 1], offset: ?>>
+ // CHECK-SAME: to memref<4x4xf32, strided<[?, 1]>>
%8 = memref.subview %7[0, 0][4, 4][1, 1]
- : memref<?x?xf32> to memref<4x4xf32, strided<[?, 1], offset: ?>>
+ : memref<?x?xf32> to memref<4x4xf32, strided<[?, 1]>>
%9 = memref.alloc() : memref<16x4xf32>
// CHECK: memref.subview {{%.*}}[{{%.*}}, {{%.*}}] [4, 4] [{{%.*}}, {{%.*}}] :
diff --git a/mlir/test/Dialect/Vector/vector-transfer-collapse-inner-most-dims.mlir b/mlir/test/Dialect/Vector/vector-transfer-collapse-inner-most-dims.mlir
index 686b4a0b60c2a..bd6845d1c7cda 100644
--- a/mlir/test/Dialect/Vector/vector-transfer-collapse-inner-most-dims.mlir
+++ b/mlir/test/Dialect/Vector/vector-transfer-collapse-inner-most-dims.mlir
@@ -85,9 +85,9 @@ func.func @contiguous_inner_most_dynamic_outer(%a: index, %b: index, %memref: me
// CHECK: %[[D0:.+]] = memref.dim %[[SRC]], %[[C0]]
// CHECK: %[[D1:.+]] = memref.dim %[[SRC]], %[[C1]]
// CHECK: %[[VIEW:.+]] = memref.subview %[[SRC]][0, 0, 0, 0] [%[[D0]], %[[D1]], 8, 1] [1, 1, 1, 1]
-// CHECK-SAME: memref<?x?x8x1xf32> to memref<?x?x8xf32, strided<[?, 8, 1], offset: ?>>
+// CHECK-SAME: memref<?x?x8x1xf32> to memref<?x?x8xf32, strided<[?, 8, 1]>>
// CHECK: %[[VEC:.+]] = vector.transfer_read %[[VIEW]]
-// CHECK-SAME: memref<?x?x8xf32, strided<[?, 8, 1], offset: ?>>, vector<8xf32>
+// CHECK-SAME: memref<?x?x8xf32, strided<[?, 8, 1]>>, vector<8xf32>
// CHECK: %[[RESULT:.+]] = vector.shape_cast %[[VEC]]
// CHECK: return %[[RESULT]]
@@ -105,10 +105,10 @@ func.func @contiguous_inner_most_outer_dim_dyn_scalable_inner_dim(%a: index, %b:
// CHECK-SAME: %[[IDX0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[IDX1:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[SRC:[a-zA-Z0-9]+]]
-// CHECK: %[[VIEW:.+]] = memref.subview %[[SRC]]{{.*}} memref<?x?x8x1xf32> to memref<?x?x8xf32, strided<[?, 8, 1], offset: ?>>
+// CHECK: %[[VIEW:.+]] = memref.subview %[[SRC]]{{.*}} memref<?x?x8x1xf32> to memref<?x?x8xf32, strided<[?, 8, 1]>>
// CHECK: %[[VEC_READ:.+]] = vector.transfer_read %[[VIEW]]
// CHECK-SAME: {in_bounds = [true]}
-// CHECK-SAME: memref<?x?x8xf32, strided<[?, 8, 1], offset: ?>>, vector<[8]xf32>
+// CHECK-SAME: memref<?x?x8xf32, strided<[?, 8, 1]>>, vector<[8]xf32>
// CHECK: vector.shape_cast %[[VEC_READ]]
// -----
@@ -340,9 +340,9 @@ func.func @contiguous_inner_most_dynamic_outer(%a: index, %b: index, %arg0: memr
// CHECK: %[[C0:.*]] = arith.constant 0 : index
// CHECK: %[[DIM0:.*]] = memref.dim %[[MEM]], %[[C0]] : memref<?x?x16x1xf32>
// CHECK: %[[DIM1:.*]] = memref.dim %[[MEM]], %[[C1]] : memref<?x?x16x1xf32>
-// CHECK: %[[SV:.*]] = memref.subview %[[MEM]][0, 0, 0, 0] {{\[}}%[[DIM0]], %[[DIM1]], 16, 1] [1, 1, 1, 1] : memref<?x?x16x1xf32> to memref<?x?x16xf32, strided<[?, 16, 1], offset: ?>>
+// CHECK: %[[SV:.*]] = memref.subview %[[MEM]][0, 0, 0, 0] {{\[}}%[[DIM0]], %[[DIM1]], 16, 1] [1, 1, 1, 1] : memref<?x?x16x1xf32> to memref<?x?x16xf32, strided<[?, 16, 1]>>
// CHECK: %[[SC:.*]] = vector.shape_cast %[[VEC]] : vector<8x1xf32> to vector<8xf32>
-// CHECK: vector.transfer_write %[[SC]], %[[SV]]{{\[}}%[[IDX_0]], %[[IDX_1]], %[[C0]]] {in_bounds = [true]} : vector<8xf32>, memref<?x?x16xf32, strided<[?, 16, 1], offset: ?>>
+// CHECK: vector.transfer_write %[[SC]], %[[SV]]{{\[}}%[[IDX_0]], %[[IDX_1]], %[[C0]]] {in_bounds = [true]} : vector<8xf32>, memref<?x?x16xf32, strided<[?, 16, 1]>>
// Same as the top example within this split, but with the outer vector
// dim scalable. Note that this example only makes sense when "8 = [8]" (i.e.
@@ -361,9 +361,9 @@ func.func @contiguous_inner_most_dynamic_outer_scalable_inner_dim(%a: index, %b:
// CHECK: %[[C0:.*]] = arith.constant 0 : index
// CHECK: %[[DIM0:.*]] = memref.dim %[[MEM]], %[[C0]] : memref<?x?x16x1xf32>
// CHECK: %[[DIM1:.*]] = memref.dim %[[MEM]], %[[C1]] : memref<?x?x16x1xf32>
-// CHECK: %[[SV:.*]] = memref.subview %[[MEM]][0, 0, 0, 0] {{\[}}%[[DIM0]], %[[DIM1]], 16, 1] [1, 1, 1, 1] : memref<?x?x16x1xf32> to memref<?x?x16xf32, strided<[?, 16, 1], offset: ?>>
+// CHECK: %[[SV:.*]] = memref.subview %[[MEM]][0, 0, 0, 0] {{\[}}%[[DIM0]], %[[DIM1]], 16, 1] [1, 1, 1, 1] : memref<?x?x16x1xf32> to memref<?x?x16xf32, strided<[?, 16, 1]>>
// CHECK: %[[SC:.*]] = vector.shape_cast %[[VEC]] : vector<[8]x1xf32> to vector<[8]xf32>
-// CHECK: vector.transfer_write %[[SC]], %[[SV]]{{\[}}%[[IDX_0]], %[[IDX_1]], %[[C0]]] {in_bounds = [true]} : vector<[8]xf32>, memref<?x?x16xf32, strided<[?, 16, 1], offset: ?>>
+// CHECK: vector.transfer_write %[[SC]], %[[SV]]{{\[}}%[[IDX_0]], %[[IDX_1]], %[[C0]]] {in_bounds = [true]} : vector<[8]xf32>, memref<?x?x16xf32, strided<[?, 16, 1]>>
// -----
diff --git a/mlir/test/Dialect/Vector/vector-transferop-opt.mlir b/mlir/test/Dialect/Vector/vector-transferop-opt.mlir
index 3ddfacf40cf64..07e6647533c6f 100644
--- a/mlir/test/Dialect/Vector/vector-transferop-opt.mlir
+++ b/mlir/test/Dialect/Vector/vector-transferop-opt.mlir
@@ -504,14 +504,14 @@ func.func @forward_and_eliminate_stores_through_trivial_aliases(
%c32 = arith.constant 32 : index
%cst = arith.constant 0.0 : f32
vector.transfer_write %vec, %buffer[%c0, %c0] {in_bounds = [true, true]} : vector<[8]x[8]xf32>, memref<?x?xf32>
- %direct_subview = memref.subview %buffer[0, 0] [%a_size, %a_size] [1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
- %cast = memref.cast %direct_subview : memref<?x?xf32, strided<[?, 1], offset: ?>> to memref<?x?xf32>
- %subview_of_cast = memref.subview %cast[0, 0] [%another_size, %another_size] [1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
- %21 = vector.transfer_read %direct_subview[%c0, %c0], %cst {in_bounds = [true, true]} : memref<?x?xf32, strided<[?, 1], offset: ?>>, vector<[8]x[8]xf32>
+ %direct_subview = memref.subview %buffer[0, 0] [%a_size, %a_size] [1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
+ %cast = memref.cast %direct_subview : memref<?x?xf32, strided<[?, 1]>> to memref<?x?xf32>
+ %subview_of_cast = memref.subview %cast[0, 0] [%another_size, %another_size] [1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>
+ %21 = vector.transfer_read %direct_subview[%c0, %c0], %cst {in_bounds = [true, true]} : memref<?x?xf32, strided<[?, 1]>>, vector<[8]x[8]xf32>
%23 = scf.for %arg2 = %c0 to %c32 step %c1 iter_args(%arg3 = %21) -> (vector<[8]x[8]xf32>) {
%24 = arith.addf %arg3, %arg3 : vector<[8]x[8]xf32>
scf.yield %24 : vector<[8]x[8]xf32>
}
- vector.transfer_write %23, %subview_of_cast[%c0, %c0] {in_bounds = [true, true]} : vector<[8]x[8]xf32>, memref<?x?xf32, strided<[?, 1], offset: ?>>
+ vector.transfer_write %23, %subview_of_cast[%c0, %c0] {in_bounds = [true, true]} : vector<[8]x[8]xf32>, memref<?x?xf32, strided<[?, 1]>>
return
}
diff --git a/mlir/test/Transforms/canonicalize.mlir b/mlir/test/Transforms/canonicalize.mlir
index 5e0d78fafb25c..9b74362b6ee75 100644
--- a/mlir/test/Transforms/canonicalize.mlir
+++ b/mlir/test/Transforms/canonicalize.mlir
@@ -505,9 +505,9 @@ func.func @dim_op_fold(%arg0: index, %arg1: index, %arg2: index, %BUF: memref<?x
affine.for %arg5 = %l to %u {
"foo"() : () -> ()
}
- %sv2 = memref.subview %0[0, 0][17, %arg4][1, 1] : memref<?x?xf32> to memref<17x?xf32, strided<[?, 1], offset: ?>>
+ %sv2 = memref.subview %0[0, 0][17, %arg4][1, 1] : memref<?x?xf32> to memref<17x?xf32, strided<[?, 1]>>
%l2 = memref.dim %v, %c1 : memref<?x?xf32>
- %u2 = memref.dim %sv2, %c1 : memref<17x?xf32, strided<[?, 1], offset: ?>>
+ %u2 = memref.dim %sv2, %c1 : memref<17x?xf32, strided<[?, 1]>>
scf.for %arg5 = %l2 to %u2 step %c1 {
"foo"() : () -> ()
}
@@ -1057,13 +1057,13 @@ func.func @memref_cast_folding_subview(%arg0: memref<4x5xf32>, %i: index) -> (me
// CHECK-LABEL: func @memref_cast_folding_subview_static(
func.func @memref_cast_folding_subview_static(%V: memref<16x16xf32>, %a: index, %b: index)
- -> memref<3x4xf32, strided<[?, 1], offset: ?>>
+ -> memref<3x4xf32, strided<[?, 1]>>
{
%0 = memref.cast %V : memref<16x16xf32> to memref<?x?xf32>
- %1 = memref.subview %0[0, 0][3, 4][1, 1] : memref<?x?xf32> to memref<3x4xf32, strided<[?, 1], offset: ?>>
+ %1 = memref.subview %0[0, 0][3, 4][1, 1] : memref<?x?xf32> to memref<3x4xf32, strided<[?, 1]>>
// CHECK: memref.subview{{.*}}: memref<16x16xf32> to memref<3x4xf32, strided<[16, 1]>>
- return %1: memref<3x4xf32, strided<[?, 1], offset: ?>>
+ return %1: memref<3x4xf32, strided<[?, 1]>>
}
// -----
More information about the Mlir-commits
mailing list