[mlir][llvm] Implement ConstantLike for ZeroOp, UndefOp, PoisonOp #93690
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
guy-david
requested review from
aartbik,
PeimingLiu,
yinying-lisa-li and
matthias-springer
as code owners
|
Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be If you wish to, you can add reviewers by using the "Reviewers" section on this page. If this is not working for you, it is probably because you do not have write If you have received no comments on your PR for a week, you can request a review If you have further questions, they may be answered by the LLVM GitHub User Guide. You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums. |
|
@llvm/pr-subscribers-mlir @llvm/pr-subscribers-mlir-llvm Author: Guy David (guy-david) ChangesThese act as constants and should be propagated whenever possible. It is safe to do so for mlir.undef and mlir.poison because they remain "dirty" through out their lifetime and can be duplicated, merged, etc. per the LangRef. Patch is 28.71 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/93690.diff 8 Files Affected:
diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMAttrDefs.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMAttrDefs.td
index 535cf8dfd2ced..9ac8fd2dba9a7 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMAttrDefs.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMAttrDefs.td
@@ -1037,4 +1037,28 @@ def LLVM_TargetFeaturesAttr : LLVM_Attr<"TargetFeatures", "target_features">
let genVerifyDecl = 1;
}
+//===----------------------------------------------------------------------===//
+// UndefAttr
+//===----------------------------------------------------------------------===//
+
+/// Folded into from LLVM::UndefOp.
+def LLVM_UndefAttr : LLVM_Attr<"Undef", "undef"> {
+}
+
+//===----------------------------------------------------------------------===//
+// PoisonAttr
+//===----------------------------------------------------------------------===//
+
+/// Folded into from LLVM::PoisonOp.
+def LLVM_PoisonAttr : LLVM_Attr<"Poison", "poison"> {
+}
+
+//===----------------------------------------------------------------------===//
+// ZeroAttr
+//===----------------------------------------------------------------------===//
+
+/// Folded into from LLVM::ZeroOp.
+def LLVM_ZeroAttr : LLVM_Attr<"Zero", "zero"> {
+}
+
#endif // LLVMIR_ATTRDEFS
diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
index 84e67d2c11dbd..f6f907f39a4b4 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
@@ -1522,7 +1522,7 @@ def LLVM_NoneTokenOp
let assemblyFormat = "attr-dict `:` type($res)";
}
-def LLVM_UndefOp : LLVM_Op<"mlir.undef", [Pure]>,
+def LLVM_UndefOp : LLVM_Op<"mlir.undef", [Pure, ConstantLike]>,
LLVM_Builder<"$res = llvm::UndefValue::get($_resultType);"> {
let summary = "Creates an undefined value of LLVM dialect type.";
let description = [{
@@ -1541,9 +1541,10 @@ def LLVM_UndefOp : LLVM_Op<"mlir.undef", [Pure]>,
let results = (outs LLVM_Type:$res);
let builders = [LLVM_OneResultOpBuilder];
let assemblyFormat = "attr-dict `:` type($res)";
+ let hasFolder = 1;
}
-def LLVM_PoisonOp : LLVM_Op<"mlir.poison", [Pure]>,
+def LLVM_PoisonOp : LLVM_Op<"mlir.poison", [Pure, ConstantLike]>,
LLVM_Builder<"$res = llvm::PoisonValue::get($_resultType);"> {
let summary = "Creates a poison value of LLVM dialect type.";
let description = [{
@@ -1563,10 +1564,11 @@ def LLVM_PoisonOp : LLVM_Op<"mlir.poison", [Pure]>,
let results = (outs LLVM_Type:$res);
let builders = [LLVM_OneResultOpBuilder];
let assemblyFormat = "attr-dict `:` type($res)";
+ let hasFolder = 1;
}
def LLVM_ZeroOp
- : LLVM_Op<"mlir.zero", [Pure]>,
+ : LLVM_Op<"mlir.zero", [Pure, ConstantLike]>,
LLVM_Builder<"$res = llvm::Constant::getNullValue($_resultType);">
{
let summary = "Creates a zero-initialized value of LLVM dialect type.";
@@ -1588,6 +1590,7 @@ def LLVM_ZeroOp
let builders = [LLVM_OneResultOpBuilder];
let assemblyFormat = "attr-dict `:` type($res)";
let hasVerifier = 1;
+ let hasFolder = 1;
}
def LLVM_ConstantOp
diff --git a/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp b/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp
index dcf3f3b52a606..d91f57f9247ab 100644
--- a/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp
+++ b/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp
@@ -2555,6 +2555,24 @@ Region *LLVMFuncOp::getCallableRegion() {
return &getBody();
}
+//===----------------------------------------------------------------------===//
+// UndefOp.
+//===----------------------------------------------------------------------===//
+
+/// Fold an undef operation to a dedicated undef attribute.
+OpFoldResult LLVM::UndefOp::fold(FoldAdaptor) {
+ return mlir::LLVM::UndefAttr::get(getContext());
+}
+
+//===----------------------------------------------------------------------===//
+// PoisonOp.
+//===----------------------------------------------------------------------===//
+
+/// Fold a poison operation to a dedicated poison attribute.
+OpFoldResult LLVM::PoisonOp::fold(FoldAdaptor) {
+ return mlir::LLVM::PoisonAttr::get(getContext());
+}
+
//===----------------------------------------------------------------------===//
// ZeroOp.
//===----------------------------------------------------------------------===//
@@ -2568,6 +2586,15 @@ LogicalResult LLVM::ZeroOp::verify() {
return success();
}
+/// Fold a zero operation to a builtin zero attribute when possible and fall
+/// back to a dedicated zero attribute.
+OpFoldResult LLVM::ZeroOp::fold(FoldAdaptor) {
+ OpFoldResult result = Builder(getContext()).getZeroAttr(getType());
+ if (result)
+ return result;
+ return mlir::LLVM::ZeroAttr::get(getContext());
+}
+
//===----------------------------------------------------------------------===//
// ConstantOp.
//===----------------------------------------------------------------------===//
@@ -3271,11 +3298,18 @@ LogicalResult LLVMDialect::verifyRegionResultAttribute(Operation *op,
Operation *LLVMDialect::materializeConstant(OpBuilder &builder, Attribute value,
Type type, Location loc) {
- // If this was folded from an llvm.mlir.addressof operation, it should be
- // materialized as such.
+ // If this was folded from an operation other than llvm.mlir.constant, it
+ // should be materialized as such. Note that an llvm.mlir.zero may fold into
+ // a builtin zero attribute and thus will materialize as a llvm.mlir.constant.
if (auto symbol = dyn_cast<FlatSymbolRefAttr>(value))
if (isa<LLVM::LLVMPointerType>(type))
return builder.create<LLVM::AddressOfOp>(loc, type, symbol);
+ if (isa<LLVM::UndefAttr>(value))
+ return builder.create<LLVM::UndefOp>(loc, type);
+ if (isa<LLVM::PoisonAttr>(value))
+ return builder.create<LLVM::PoisonOp>(loc, type);
+ if (isa<LLVM::ZeroAttr>(value))
+ return builder.create<LLVM::ZeroOp>(loc, type);
// Otherwise try materializing it as a regular llvm.mlir.constant op.
return LLVM::ConstantOp::materialize(builder, value, type, loc);
}
diff --git a/mlir/test/Dialect/GPU/dynamic-shared-memory.mlir b/mlir/test/Dialect/GPU/dynamic-shared-memory.mlir
index fb45faaa712f7..d73125fd763e6 100644
--- a/mlir/test/Dialect/GPU/dynamic-shared-memory.mlir
+++ b/mlir/test/Dialect/GPU/dynamic-shared-memory.mlir
@@ -3,11 +3,11 @@
gpu.module @modules {
// CHECK: llvm.mlir.global internal @__dynamic_shmem__3() {addr_space = 3 : i32, alignment = 16 : i64} : !llvm.array<0 x i8>
llvm.mlir.global internal @__dynamic_shmem__0() {addr_space = 3 : i32, alignment = 4 : i64} : !llvm.array<0 x i8>
- llvm.mlir.global internal @__dynamic_shmem__1() {addr_space = 3 : i32, alignment = 4 : i64} : !llvm.array<0 x i8>
- llvm.mlir.global internal @__dynamic_shmem__2() {alignment = 16 : i64} : !llvm.array<0 x i8>
+ llvm.mlir.global internal @__dynamic_shmem__1() {addr_space = 3 : i32, alignment = 4 : i64} : !llvm.array<0 x i8>
+ llvm.mlir.global internal @__dynamic_shmem__2() {alignment = 16 : i64} : !llvm.array<0 x i8>
// CHECK-LABEL: llvm.func @dynamic_shared_memory_kernel(
// CHECK-SAME: %[[arg0:.+]]: i64)
- gpu.func @dynamic_shared_memory_kernel(%d : index) kernel attributes {gpu.known_block_size = array<i32: 1, 1, 1>, gpu.known_grid_size = array<i32: 1, 1, 1>} {
+ gpu.func @dynamic_shared_memory_kernel(%d : index) kernel attributes {gpu.known_block_size = array<i32: 1, 1, 1>, gpu.known_grid_size = array<i32: 1, 1, 1>} {
%c1 = arith.constant 1 : index
%c8192 = arith.constant 8192 : index
%c16384 = arith.constant 16384 : index
@@ -19,83 +19,83 @@ gpu.module @modules {
%1 = memref.view %shmem[%c16384][] : memref<?xi8, #gpu.address_space<workgroup>> to memref<32x64xf32, #gpu.address_space<workgroup>>
"test.use.shared.memory"(%1) : (memref<32x64xf32, #gpu.address_space<workgroup>>) -> ()
-
-// CHECK: %[[S0:.+]] = llvm.mlir.constant(32 : index) : i64
-// CHECK: %[[S1:.+]] = llvm.mlir.constant(64 : index) : i64
-// CHECK: %[[S2:.+]] = llvm.mlir.constant(1 : index) : i64
-// CHECK: %[[S3:.+]] = llvm.mlir.constant(0 : index) : i64
-// CHECK: %[[S4:.+]] = llvm.mlir.addressof @__dynamic_shmem__3 : !llvm.ptr<3>
-// CHECK: %[[S5:.+]] = llvm.mlir.undef : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S6:.+]] = llvm.insertvalue %[[S4]], %[[S5]][0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S7:.+]] = llvm.getelementptr %[[S4]][8192] : (!llvm.ptr<3>) -> !llvm.ptr<3>, i8
-// CHECK: %[[S8:.+]] = llvm.insertvalue %[[S7]], %[[S6]][1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S9:.+]] = llvm.insertvalue %[[S3]], %[[S8]][2] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S10:.+]] = llvm.insertvalue %[[S1]], %[[S9]][3, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S11:.+]] = llvm.insertvalue %[[S2]], %[[S10]][4, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S12:.+]] = llvm.insertvalue %[[S0]], %[[S11]][3, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S13:.+]] = llvm.insertvalue %[[S1]], %[[S12]][4, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S14:.+]] = builtin.unrealized_conversion_cast %[[S13]] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)> to memref<32x64xf32, #gpu.address_space<workgroup>>
-// CHECK: "test.use.shared.memory"(%[[S14]]) : (memref<32x64xf32, #gpu.address_space<workgroup>>) -> ()
-// CHECK: %[[S15:.+]] = llvm.getelementptr %4[16384] : (!llvm.ptr<3>) -> !llvm.ptr<3>, i8
-// CHECK: %[[S16:.+]] = llvm.insertvalue %[[S15]], %[[S6]][1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S17:.+]] = llvm.insertvalue %[[S3]], %[[S16]][2] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S18:.+]] = llvm.insertvalue %[[S1]], %[[S17]][3, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S19:.+]] = llvm.insertvalue %[[S2]], %[[S18]][4, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S20:.+]] = llvm.insertvalue %[[S0]], %[[S19]][3, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S21:.+]] = llvm.insertvalue %[[S1]], %[[S20]][4, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S22:.+]] = builtin.unrealized_conversion_cast %[[S21]] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)> to memref<32x64xf32, #gpu.address_space<workgroup>>
-// CHECK: "test.use.shared.memory"(%[[S22]]) : (memref<32x64xf32, #gpu.address_space<workgroup>>) -> ()
+
+// CHECK-DAG: %[[S0:.+]] = llvm.mlir.constant(32 : index) : i64
+// CHECK-DAG: %[[S1:.+]] = llvm.mlir.constant(64 : index) : i64
+// CHECK-DAG: %[[S2:.+]] = llvm.mlir.undef : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK-DAG: %[[S3:.+]] = llvm.mlir.constant(1 : index) : i64
+// CHECK-DAG: %[[S4:.+]] = llvm.mlir.constant(0 : index) : i64
+// CHECK-DAG: %[[S5:.+]] = llvm.mlir.addressof @__dynamic_shmem__3 : !llvm.ptr<3>
+// CHECK: %[[S6:.+]] = llvm.insertvalue %[[S5]], %[[S2]][0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S7:.+]] = llvm.getelementptr %[[S5]][8192] : (!llvm.ptr<3>) -> !llvm.ptr<3>, i8
+// CHECK: %[[S8:.+]] = llvm.insertvalue %[[S7]], %[[S6]][1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S9:.+]] = llvm.insertvalue %[[S4]], %[[S8]][2] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S10:.+]] = llvm.insertvalue %[[S1]], %[[S9]][3, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S11:.+]] = llvm.insertvalue %[[S3]], %[[S10]][4, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S12:.+]] = llvm.insertvalue %[[S0]], %[[S11]][3, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S13:.+]] = llvm.insertvalue %[[S1]], %[[S12]][4, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S14:.+]] = builtin.unrealized_conversion_cast %[[S13]] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)> to memref<32x64xf32, #gpu.address_space<workgroup>>
+// CHECK: "test.use.shared.memory"(%[[S14]]) : (memref<32x64xf32, #gpu.address_space<workgroup>>) -> ()
+// CHECK: %[[S15:.+]] = llvm.getelementptr %[[S5]][16384] : (!llvm.ptr<3>) -> !llvm.ptr<3>, i8
+// CHECK: %[[S16:.+]] = llvm.insertvalue %[[S15]], %[[S6]][1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S17:.+]] = llvm.insertvalue %[[S4]], %[[S16]][2] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S18:.+]] = llvm.insertvalue %[[S1]], %[[S17]][3, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S19:.+]] = llvm.insertvalue %[[S3]], %[[S18]][4, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S20:.+]] = llvm.insertvalue %[[S0]], %[[S19]][3, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S21:.+]] = llvm.insertvalue %[[S1]], %[[S20]][4, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S22:.+]] = builtin.unrealized_conversion_cast %[[S21]] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)> to memref<32x64xf32, #gpu.address_space<workgroup>>
+// CHECK: "test.use.shared.memory"(%[[S22]]) : (memref<32x64xf32, #gpu.address_space<workgroup>>) -> ()
gpu.return
}
// CHECK-LABEL: llvm.func @gpu_device_function
- gpu.func @gpu_device_function() {
+ gpu.func @gpu_device_function() {
%c8192 = arith.constant 8192 : index
%shmem = gpu.dynamic_shared_memory : memref<?xi8, #gpu.address_space<workgroup>>
%0 = memref.view %shmem[%c8192][] : memref<?xi8, #gpu.address_space<workgroup>> to memref<32x64xf32, #gpu.address_space<workgroup>>
"test.use.shared.memory"(%0) : (memref<32x64xf32, #gpu.address_space<workgroup>>) -> ()
-// CHECK: %[[S0:.+]] = llvm.mlir.constant(32 : index) : i64
-// CHECK: %[[S1:.+]] = llvm.mlir.constant(64 : index) : i64
-// CHECK: %[[S2:.+]] = llvm.mlir.constant(1 : index) : i64
-// CHECK: %[[S3:.+]] = llvm.mlir.constant(0 : index) : i64
-// CHECK: %[[S4:.+]] = llvm.mlir.addressof @__dynamic_shmem__3 : !llvm.ptr<3>
-// CHECK: %[[S5:.+]] = llvm.mlir.undef : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S6:.+]] = llvm.insertvalue %[[S4]], %[[S5]][0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S7:.+]] = llvm.getelementptr %[[S4]][8192] : (!llvm.ptr<3>) -> !llvm.ptr<3>, i8
-// CHECK: %[[S8:.+]] = llvm.insertvalue %[[S7]], %[[S6]][1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S9:.+]] = llvm.insertvalue %[[S3]], %[[S8]][2] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S10:.+]] = llvm.insertvalue %[[S1]], %[[S9]][3, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S11:.+]] = llvm.insertvalue %[[S2]], %[[S10]][4, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S12:.+]] = llvm.insertvalue %[[S0]], %[[S11]][3, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S13:.+]] = llvm.insertvalue %[[S1]], %[[S12]][4, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S14:.+]] = builtin.unrealized_conversion_cast %13 : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)> to memref<32x64xf32, #gpu.address_space<workgroup>>
-// CHECK: "test.use.shared.memory"(%[[S14]]) : (memref<32x64xf32, #gpu.address_space<workgroup>>) -> ()
+// CHECK-DAG: %[[S0:.+]] = llvm.mlir.constant(32 : index) : i64
+// CHECK-DAG: %[[S1:.+]] = llvm.mlir.constant(64 : index) : i64
+// CHECK-DAG: %[[S2:.+]] = llvm.mlir.undef : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK-DAG: %[[S3:.+]] = llvm.mlir.constant(1 : index) : i64
+// CHECK-DAG: %[[S4:.+]] = llvm.mlir.constant(0 : index) : i64
+// CHECK-DAG: %[[S5:.+]] = llvm.mlir.addressof @__dynamic_shmem__3 : !llvm.ptr<3>
+// CHECK: %[[S6:.+]] = llvm.insertvalue %[[S5]], %[[S2]][0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S7:.+]] = llvm.getelementptr %[[S5]][8192] : (!llvm.ptr<3>) -> !llvm.ptr<3>, i8
+// CHECK: %[[S8:.+]] = llvm.insertvalue %[[S7]], %[[S6]][1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S9:.+]] = llvm.insertvalue %[[S4]], %[[S8]][2] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S10:.+]] = llvm.insertvalue %[[S1]], %[[S9]][3, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S11:.+]] = llvm.insertvalue %[[S3]], %[[S10]][4, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S12:.+]] = llvm.insertvalue %[[S0]], %[[S11]][3, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S13:.+]] = llvm.insertvalue %[[S1]], %[[S12]][4, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
+// CHECK: %[[S14:.+]] = builtin.unrealized_conversion_cast %13 : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)> to memref<32x64xf32, #gpu.address_space<workgroup>>
+// CHECK: "test.use.shared.memory"(%[[S14]]) : (memref<32x64xf32, #gpu.address_space<workgroup>>) -> ()
gpu.return
}
// CHECK-LABEL: llvm.func @func_device_function
- func.func @func_device_function() {
+ func.func @func_device_function() {
%c8192 = arith.constant 8192 : index
%shmem = gpu.dynamic_shared_memory : memref<?xi8, #gpu.address_space<workgroup>>
%0 = memref.view %shmem[%c8192][] : memref<?xi8, #gpu.address_space<workgroup>> to memref<32x64xf32, #gpu.address_space<workgroup>>
"test.use.shared.memory"(%0) : (memref<32x64xf32, #gpu.address_space<workgroup>>) -> ()
-// CHECK: %[[S0:.+]] = llvm.mlir.constant(32 : index) : i64
-// CHECK: %[[S1:.+]] = llvm.mlir.constant(64 : index) : i64
-// CHECK: %[[S2:.+]] = llvm.mlir.constant(1 : index) : i64
-// CHECK: %[[S3:.+]] = llvm.mlir.constant(0 : index) : i64
-// CHECK: %[[S4:.+]] = llvm.mlir.addressof @__dynamic_shmem__3 : !llvm.ptr<3>
-// CHECK: %[[S5:.+]] = llvm.mlir.undef : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S6:.+]] = llvm.insertvalue %[[S4]], %[[S5]][0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S7:.+]] = llvm.getelementptr %[[S4]][8192] : (!llvm.ptr<3>) -> !llvm.ptr<3>, i8
-// CHECK: %[[S8:.+]] = llvm.insertvalue %[[S7]], %[[S6]][1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S9:.+]] = llvm.insertvalue %[[S3]], %[[S8]][2] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S10:.+]] = llvm.insertvalue %[[S1]], %[[S9]][3, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S11:.+]] = llvm.insertvalue %[[S2]], %[[S10]][4, 1] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S12:.+]] = llvm.insertvalue %[[S0]], %[[S11]][3, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S13:.+]] = llvm.insertvalue %[[S1]], %[[S12]][4, 0] : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)>
-// CHECK: %[[S14:.+]] = builtin.unrealized_conversion_cast %13 : !llvm.struct<(ptr<3>, ptr<3>, i64, array<2 x i64>, array<2 x i64>)> to memref<32x64xf32, #gpu.address_space<workgroup>>
-// CHECK: "test.use.shared.memory"(%[[S14]]) : (memref<32x64xf32, #gpu.address_space<workgroup>>) -> ()
+// CHECK-DAG: %[[S0:.+]] = llvm.mlir.constant(32 : index) : i64
+// CHECK-DAG: %[[S1:.+]] = llvm.mlir.constant(64 : index) : i64
+// CHECK-DAG: %[[S2:.+]] = llvm.mlir.undef : !llvm.struct<(p...
[truncated]
|
guy-david
force-pushed
the
llvm-constant-like
branch
from
92b51bc
to
dab7be7
Compare
These act as constants and should be propagated whenever possible. It is safe to do so for mlir.undef and mlir.poison because they remain "dirty" through out their lifetime and can be duplicated, merged, etc. per the LangRef. Signed-off-by: Guy David <guy.david@nextsilicon.com>
guy-david
force-pushed
the
llvm-constant-like
branch
from
dab7be7
to
8fefca1
Compare
|
@guy-david Congratulations on having your first Pull Request (PR) merged into the LLVM Project! Your changes will be combined with recent changes from other authors, then tested Please check whether problems have been caused by your change specifically, as How to do this, and the rest of the post-merge process, is covered in detail here. If your change does cause a problem, it may be reverted, or you can revert it yourself. If you don't get any reports, no action is required from you. Your changes are working as expected, well done! |
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.
These act as constants and should be propagated whenever possible. It is safe to do so for mlir.undef and mlir.poison because they remain "dirty" through out their lifetime and can be duplicated, merged, etc. per the LangRef.