[Mlir-commits] [mlir] 90a0be9 - [mlir][LLVM] Refactor how range() annotations are handled for ROCDL intrinsics (#107658)
llvmlistbot at llvm.org
llvmlistbot at llvm.org
Thu Sep 12 07:46:45 PDT 2024
Author: Krzysztof Drewniak
Date: 2024-09-12T09:46:42-05:00
New Revision: 90a0be9482ad00acc1772440f13656190ec1ac22
URL: https://github.com/llvm/llvm-project/commit/90a0be9482ad00acc1772440f13656190ec1ac22
DIFF: https://github.com/llvm/llvm-project/commit/90a0be9482ad00acc1772440f13656190ec1ac22.diff
LOG: [mlir][LLVM] Refactor how range() annotations are handled for ROCDL intrinsics (#107658)
This commit introduces a ConstantRange attribute to match the
ConstantRange attribute type present in LLVM IR.
It then refactors the LLVM_IntrOpBase so that the basic part of the
intrinsic builder code can be re-used without needing to copy it or
get rid of important context. This, along with adding code for
handling an optional `range` attribute to that same base, allows us to
make the support for range() annotations generic without adding
another bit to IntrOpBase.
This commit then updates the lowering of index intrinsic operations to
use the new ConstantRange attribute and fixes a bug (where we'd be
subtracting 1 from upper bounds instead of adding it on operations
like gpu.block_dim) along the way.
The point of these changes is to enable these range annotations to be
used for the corresponding NVVM operations in a future commit.
Added:
Modified:
mlir/include/mlir/Dialect/LLVMIR/LLVMAttrDefs.td
mlir/include/mlir/Dialect/LLVMIR/LLVMOpBase.td
mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
mlir/lib/Conversion/GPUCommon/IndexIntrinsicsOpLowering.h
mlir/lib/Dialect/LLVMIR/IR/LLVMAttrs.cpp
mlir/lib/Target/LLVMIR/Dialect/ROCDL/ROCDLToLLVMIRTranslation.cpp
mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
mlir/test/Target/LLVMIR/rocdl.mlir
Removed:
################################################################################
diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMAttrDefs.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMAttrDefs.td
index 49e54df3436ff3..2da45eba77655b 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMAttrDefs.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMAttrDefs.td
@@ -1034,6 +1034,40 @@ def LLVM_TBAATagArrayAttr
let constBuilderCall = ?;
}
+//===----------------------------------------------------------------------===//
+// ConstantRangeAttr
+//===----------------------------------------------------------------------===//
+def LLVM_ConstantRangeAttr : LLVM_Attr<"ConstantRange", "constant_range"> {
+ let parameters = (ins
+ "::llvm::APInt":$lower,
+ "::llvm::APInt":$upper
+ );
+ let summary = "A range of two integers, corresponding to LLVM's ConstantRange";
+ let description = [{
+ A pair of two integers, mapping to the ConstantRange structure in LLVM IR,
+ which is allowed to wrap or be empty.
+
+ The range represented is [Lower, Upper), and is either signed or unsigned
+ depending on context.
+
+ `lower` and `upper` must have the same width.
+
+ Syntax:
+ ```
+ `<` `i`(width($lower)) $lower `,` $upper `>`
+ }];
+
+ let builders = [
+ AttrBuilder<(ins "uint32_t":$bitWidth, "int64_t":$lower, "int64_t":$upper), [{
+ return $_get($_ctxt, ::llvm::APInt(bitWidth, lower), ::llvm::APInt(bitWidth, upper));
+ }]>
+ ];
+
+ let hasCustomAssemblyFormat = 1;
+ let genVerifyDecl = 1;
+}
+
+
//===----------------------------------------------------------------------===//
// VScaleRangeAttr
//===----------------------------------------------------------------------===//
diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMOpBase.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMOpBase.td
index 7b9a9cf017c537..c3d352d8d0dd48 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMOpBase.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMOpBase.td
@@ -319,17 +319,19 @@ class LLVM_IntrOpBase<Dialect dialect, string opName, string enumName,
string immArgPositionsCpp = "{" # !interleave(immArgPositions, ", ") # "}";
string immArgAttrNamesCpp = "{" # !interleave(!foreach(name, immArgAttrNames,
"StringLiteral(\"" # name # "\")"), ", ") # "}";
- let llvmBuilder = [{
+ string baseLlvmBuilder = [{
auto *inst = LLVM::detail::createIntrinsicCall(
builder, moduleTranslation, &opInst, llvm::Intrinsic::}] # !interleave([
enumName, "" # numResults, overloadedResultsCpp, overloadedOperandsCpp,
immArgPositionsCpp, immArgAttrNamesCpp], ",") # [{);
(void) inst;
- }] # !if(!gt(requiresAccessGroup, 0), setAccessGroupsMetadataCode, "")
+ }];
+ string baseLlvmBuilderCoda = !if(!gt(numResults, 0), "$res = inst;", "");
+ let llvmBuilder = baseLlvmBuilder # !if(!gt(requiresAccessGroup, 0), setAccessGroupsMetadataCode, "")
# !if(!gt(requiresAliasAnalysis, 0), setAliasAnalysisMetadataCode, "")
- # !if(!gt(numResults, 0), "$res = inst;", "");
+ # baseLlvmBuilderCoda;
- string mlirBuilder = [{
+ string baseMlirBuilder = [{
SmallVector<Value> mlirOperands;
SmallVector<NamedAttribute> mlirAttrs;
if (failed(moduleImport.convertIntrinsicArguments(
@@ -345,9 +347,32 @@ class LLVM_IntrOpBase<Dialect dialect, string opName, string enumName,
}] # !if(!gt(numResults, 0), "{$_resultType};", "{};") # [{
auto op = $_builder.create<$_qualCppClassName>(
$_location, resultTypes, mlirOperands, mlirAttrs);
- }] # !if(!gt(requiresFastmath, 0),
+ }];
+ string baseMlirBuilderCoda = !if(!gt(numResults, 0), "$res = op;", "$_op = op;");
+ let mlirBuilder = baseMlirBuilder # !if(!gt(requiresFastmath, 0),
"moduleImport.setFastmathFlagsAttr(inst, op);", "")
- # !if(!gt(numResults, 0), "$res = op;", "$_op = op;");
+ # baseMlirBuilderCoda;
+
+ // Code for handling a `range` attribute that holds the constant range of the
+ // intrinsic's result (if one is specified at the call site). This is intended
+ // for GPU IDs and other calls where range() is meaningful. It expects
+ // an optional LLVM_ConstantRangeAttr named `range` to be present on the
+ // operation. These are included to abstract out common code in several
+ // dialects.
+ string setRangeRetAttrCode = [{
+ if ($range) {
+ inst->addRangeRetAttr(::llvm::ConstantRange(
+ $range->getLower(), $range->getUpper()));
+ }
+ }];
+ string importRangeRetAttrCode = [{
+ // Note: we don't want to look in to the declaration here.
+ auto rangeAttr = inst->getAttributes().getRetAttr(::llvm::Attribute::Range);
+ if (rangeAttr.isValid()) {
+ const ::llvm::ConstantRange& value = rangeAttr.getValueAsConstantRange();
+ op.setRangeAttr(::mlir::LLVM::ConstantRangeAttr::get($_builder.getContext(), value.getLower(), value.getUpper()));
+ }
+ }];
}
// Base class for LLVM intrinsic operations, should not be used directly. Places
diff --git a/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td b/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
index 35fd8270ca6935..de232462556502 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
@@ -98,23 +98,36 @@ class ROCDL_IntrOp<string mnemonic, list<int> overloadedResults,
// ROCDL special register op definitions
//===----------------------------------------------------------------------===//
-class ROCDL_SpecialRegisterOp<string mnemonic,
- list<Trait> traits = []> :
- ROCDL_Op<mnemonic, !listconcat(traits, [Pure])>,
- Results<(outs LLVM_Type:$res)>, Arguments<(ins)> {
- string llvmBuilder = "$res = createIntrinsicCallWithRange(builder,"
- # "llvm::Intrinsic::amdgcn_" # !subst(".","_", mnemonic)
- # ", op->getAttrOfType<::mlir::DenseI32ArrayAttr>(\"range\"));";
- let assemblyFormat = "attr-dict `:` type($res)";
+class ROCDL_SpecialIdRegisterOp<string mnemonic> :
+ ROCDL_IntrPure1Op<mnemonic>,
+ Arguments<(ins OptionalAttr<LLVM_ConstantRangeAttr>:$range)> {
+ string llvmBuilder = baseLlvmBuilder # setRangeRetAttrCode # baseLlvmBuilderCoda;
+ string mlirBuilder = baseMlirBuilder # importRangeRetAttrCode # baseMlirBuilderCoda;
+
+ let assemblyFormat = "(`range` $range^)? attr-dict `:` type($res)";
+
+ // Temporaly builder until Nvidia ops also support range attributes.
+ let builders = [
+ OpBuilder<(ins "Type":$resultType), [{
+ build($_builder, $_state, resultType, ::mlir::LLVM::ConstantRangeAttr{});
+ }]>
+ ];
}
-class ROCDL_DeviceFunctionOp<string mnemonic, string device_function,
+class ROCDL_DimGetterFunctionOp<string mnemonic, string device_function,
int parameter, list<Trait> traits = []> :
ROCDL_Op<mnemonic, !listconcat(traits, [Pure])>,
- Results<(outs LLVM_Type:$res)>, Arguments<(ins)> {
- string llvmBuilder = "$res = createDeviceFunctionCall(builder, \""
+ Results<(outs LLVM_Type:$res)>, Arguments<(ins OptionalAttr<LLVM_ConstantRangeAttr>:$range)> {
+ string llvmBuilder = "$res = createDimGetterFunctionCall(builder, op, \""
# device_function # "\", " # parameter # ");";
- let assemblyFormat = "attr-dict `:` type($res)";
+ let assemblyFormat = "(`range` $range^)? attr-dict `:` type($res)";
+
+ // Temporaly builder until Nvidia ops also support range attributes.
+ let builders = [
+ OpBuilder<(ins "Type":$resultType), [{
+ build($_builder, $_state, resultType, ::mlir::LLVM::ConstantRangeAttr{});
+ }]>
+ ];
}
//===----------------------------------------------------------------------===//
@@ -181,33 +194,33 @@ def ROCDL_BallotOp :
//===----------------------------------------------------------------------===//
// Thread index and Block index
-def ROCDL_ThreadIdXOp : ROCDL_SpecialRegisterOp<"workitem.id.x">;
-def ROCDL_ThreadIdYOp : ROCDL_SpecialRegisterOp<"workitem.id.y">;
-def ROCDL_ThreadIdZOp : ROCDL_SpecialRegisterOp<"workitem.id.z">;
+def ROCDL_ThreadIdXOp : ROCDL_SpecialIdRegisterOp<"workitem.id.x">;
+def ROCDL_ThreadIdYOp : ROCDL_SpecialIdRegisterOp<"workitem.id.y">;
+def ROCDL_ThreadIdZOp : ROCDL_SpecialIdRegisterOp<"workitem.id.z">;
-def ROCDL_BlockIdXOp : ROCDL_SpecialRegisterOp<"workgroup.id.x">;
-def ROCDL_BlockIdYOp : ROCDL_SpecialRegisterOp<"workgroup.id.y">;
-def ROCDL_BlockIdZOp : ROCDL_SpecialRegisterOp<"workgroup.id.z">;
+def ROCDL_BlockIdXOp : ROCDL_SpecialIdRegisterOp<"workgroup.id.x">;
+def ROCDL_BlockIdYOp : ROCDL_SpecialIdRegisterOp<"workgroup.id.y">;
+def ROCDL_BlockIdZOp : ROCDL_SpecialIdRegisterOp<"workgroup.id.z">;
//===----------------------------------------------------------------------===//
// Thread range and Block range
-def ROCDL_BlockDimXOp : ROCDL_DeviceFunctionOp<"workgroup.dim.x",
+def ROCDL_BlockDimXOp : ROCDL_DimGetterFunctionOp<"workgroup.dim.x",
"__ockl_get_local_size", 0>;
-def ROCDL_BlockDimYOp : ROCDL_DeviceFunctionOp<"workgroup.dim.y",
+def ROCDL_BlockDimYOp : ROCDL_DimGetterFunctionOp<"workgroup.dim.y",
"__ockl_get_local_size", 1>;
-def ROCDL_BlockDimZOp : ROCDL_DeviceFunctionOp<"workgroup.dim.z",
+def ROCDL_BlockDimZOp : ROCDL_DimGetterFunctionOp<"workgroup.dim.z",
"__ockl_get_local_size", 2>;
-def ROCDL_GridDimXOp : ROCDL_DeviceFunctionOp<"grid.dim.x",
+def ROCDL_GridDimXOp : ROCDL_DimGetterFunctionOp<"grid.dim.x",
"__ockl_get_num_groups", 0>;
-def ROCDL_GridDimYOp : ROCDL_DeviceFunctionOp<"grid.dim.y",
+def ROCDL_GridDimYOp : ROCDL_DimGetterFunctionOp<"grid.dim.y",
"__ockl_get_num_groups", 1>;
-def ROCDL_GridDimZOp : ROCDL_DeviceFunctionOp<"grid.dim.z",
+def ROCDL_GridDimZOp : ROCDL_DimGetterFunctionOp<"grid.dim.z",
"__ockl_get_num_groups", 2>;
//===----------------------------------------------------------------------===//
diff --git a/mlir/lib/Conversion/GPUCommon/IndexIntrinsicsOpLowering.h b/mlir/lib/Conversion/GPUCommon/IndexIntrinsicsOpLowering.h
index e4cd24e0380e71..eaf1554a83f89a 100644
--- a/mlir/lib/Conversion/GPUCommon/IndexIntrinsicsOpLowering.h
+++ b/mlir/lib/Conversion/GPUCommon/IndexIntrinsicsOpLowering.h
@@ -114,9 +114,9 @@ struct OpLowering : public ConvertOpToLLVMPattern<Op> {
if (upperBound && intrType != IntrType::None) {
int32_t min = (intrType == IntrType::Dim ? 1 : 0);
- int32_t max = *upperBound - (intrType == IntrType::Id ? 0 : 1);
- newOp->setAttr(
- "range", DenseI32ArrayAttr::get(op.getContext(), ArrayRef{min, max}));
+ int32_t max = *upperBound + (intrType == IntrType::Id ? 0 : 1);
+ newOp->setAttr("range", LLVM::ConstantRangeAttr::get(
+ rewriter.getContext(), 32, min, max));
}
if (indexBitwidth > 32) {
newOp = rewriter.create<LLVM::SExtOp>(
diff --git a/mlir/lib/Dialect/LLVMIR/IR/LLVMAttrs.cpp b/mlir/lib/Dialect/LLVMIR/IR/LLVMAttrs.cpp
index 491dcc7f01e73d..6047c4a7ef5150 100644
--- a/mlir/lib/Dialect/LLVMIR/IR/LLVMAttrs.cpp
+++ b/mlir/lib/Dialect/LLVMIR/IR/LLVMAttrs.cpp
@@ -232,6 +232,47 @@ DIRecursiveTypeAttrInterface DISubprogramAttr::getRecSelf(DistinctAttr recId) {
{}, {}, {}, {}, {}, 0, 0, {}, {}, {}, {});
}
+//===----------------------------------------------------------------------===//
+// ConstantRangeAttr
+//===----------------------------------------------------------------------===//
+
+Attribute ConstantRangeAttr::parse(AsmParser &parser, Type odsType) {
+ llvm::SMLoc loc = parser.getCurrentLocation();
+ IntegerType widthType;
+ if (parser.parseLess() || parser.parseType(widthType) ||
+ parser.parseComma()) {
+ return Attribute{};
+ }
+ unsigned bitWidth = widthType.getWidth();
+ APInt lower(bitWidth, 0);
+ APInt upper(bitWidth, 0);
+ if (parser.parseInteger(lower) || parser.parseComma() ||
+ parser.parseInteger(upper) || parser.parseGreater())
+ return Attribute{};
+ // For some reason, 0 is always parsed as 64-bits, fix that if needed.
+ if (lower.isZero())
+ lower = lower.sextOrTrunc(bitWidth);
+ if (upper.isZero())
+ upper = upper.sextOrTrunc(bitWidth);
+ return parser.getChecked<ConstantRangeAttr>(loc, parser.getContext(), lower,
+ upper);
+}
+
+void ConstantRangeAttr::print(AsmPrinter &printer) const {
+ printer << "<i" << getLower().getBitWidth() << ", " << getLower() << ", "
+ << getUpper() << ">";
+}
+
+LogicalResult
+ConstantRangeAttr::verify(llvm::function_ref<InFlightDiagnostic()> emitError,
+ APInt lower, APInt upper) {
+ if (lower.getBitWidth() != upper.getBitWidth())
+ return emitError()
+ << "expected lower and upper to have matching bitwidths but got "
+ << lower.getBitWidth() << " vs. " << upper.getBitWidth();
+ return success();
+}
+
//===----------------------------------------------------------------------===//
// TargetFeaturesAttr
//===----------------------------------------------------------------------===//
diff --git a/mlir/lib/Target/LLVMIR/Dialect/ROCDL/ROCDLToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/ROCDL/ROCDLToLLVMIRTranslation.cpp
index c1ee6507763566..ec21fbf714c24a 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/ROCDL/ROCDLToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/ROCDL/ROCDLToLLVMIRTranslation.cpp
@@ -26,25 +26,13 @@ using namespace mlir;
using namespace mlir::LLVM;
using mlir::LLVM::detail::createIntrinsicCall;
-static llvm::Value *createIntrinsicCallWithRange(llvm::IRBuilderBase &builder,
- llvm::Intrinsic::ID intrinsic,
- DenseI32ArrayAttr maybeRange) {
- auto *inst = llvm::cast<llvm::CallInst>(
- createIntrinsicCall(builder, intrinsic, {}, {}));
- if (maybeRange) {
- llvm::ConstantRange Range(APInt(32, maybeRange[0]),
- APInt(32, maybeRange[1]));
- inst->addRangeRetAttr(Range);
- }
- return inst;
-}
-
-// Create a call to ROCm-Device-Library function
-// Currently this routine will work only for calling ROCDL functions that
-// take a single int32 argument. It is likely that the interface of this
-// function will change to make it more generic.
-static llvm::Value *createDeviceFunctionCall(llvm::IRBuilderBase &builder,
- StringRef fnName, int parameter) {
+// Create a call to ROCm-Device-Library function that returns an ID.
+// This is intended to specifically call device functions that fetch things like
+// block or grid dimensions, and so is limited to functions that take one
+// integer parameter.
+static llvm::Value *createDimGetterFunctionCall(llvm::IRBuilderBase &builder,
+ Operation *op, StringRef fnName,
+ int parameter) {
llvm::Module *module = builder.GetInsertBlock()->getModule();
llvm::FunctionType *functionType = llvm::FunctionType::get(
llvm::Type::getInt64Ty(module->getContext()), // return type.
@@ -54,7 +42,14 @@ static llvm::Value *createDeviceFunctionCall(llvm::IRBuilderBase &builder,
module->getOrInsertFunction(fnName, functionType).getCallee());
llvm::Value *fnOp0 = llvm::ConstantInt::get(
llvm::Type::getInt32Ty(module->getContext()), parameter);
- return builder.CreateCall(fn, ArrayRef<llvm::Value *>(fnOp0));
+ auto *call = builder.CreateCall(fn, ArrayRef<llvm::Value *>(fnOp0));
+ if (auto rangeAttr = op->getAttrOfType<LLVM::ConstantRangeAttr>("range")) {
+ // Zero-extend to 64 bits because the GPU dialect uses 32-bit bounds but
+ // these ockl functions are defined to be 64-bits
+ call->addRangeRetAttr(llvm::ConstantRange(rangeAttr.getLower().zext(64),
+ rangeAttr.getUpper().zext(64)));
+ }
+ return call;
}
namespace {
diff --git a/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir b/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
index b6fb08522ae1f3..56b65beb036954 100644
--- a/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
+++ b/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
@@ -77,18 +77,18 @@ gpu.module @test_module {
{known_block_size = array<i32: 8, 12, 16>,
known_grid_size = array<i32: 20, 24, 28>} {
- // CHECK: rocdl.workitem.id.x {range = array<i32: 0, 8>} : i32
+ // CHECK: rocdl.workitem.id.x range <i32, 0, 8> : i32
%tIdX = gpu.thread_id x
- // CHECK: rocdl.workitem.id.y {range = array<i32: 0, 12>} : i32
+ // CHECK: rocdl.workitem.id.y range <i32, 0, 12> : i32
%tIdY = gpu.thread_id y
- // CHECK: rocdl.workitem.id.z {range = array<i32: 0, 16>} : i32
+ // CHECK: rocdl.workitem.id.z range <i32, 0, 16> : i32
%tIdZ = gpu.thread_id z
- // CHECK: rocdl.workgroup.id.x {range = array<i32: 0, 20>} : i32
+ // CHECK: rocdl.workgroup.id.x range <i32, 0, 20> : i32
%bIdX = gpu.block_id x
- // CHECK: rocdl.workgroup.id.y {range = array<i32: 0, 24>} : i32
+ // CHECK: rocdl.workgroup.id.y range <i32, 0, 24> : i32
%bIdY = gpu.block_id y
- // CHECK: rocdl.workgroup.id.z {range = array<i32: 0, 28>} : i32
+ // CHECK: rocdl.workgroup.id.z range <i32, 0, 28> : i32
%bIdZ = gpu.block_id z
// "Usage" to make the ID calls not die
diff --git a/mlir/test/Target/LLVMIR/rocdl.mlir b/mlir/test/Target/LLVMIR/rocdl.mlir
index 97b505746fc751..0f0c2412e5ec2a 100644
--- a/mlir/test/Target/LLVMIR/rocdl.mlir
+++ b/mlir/test/Target/LLVMIR/rocdl.mlir
@@ -28,8 +28,10 @@ llvm.func @rocdl_special_regs() -> i32 {
%12 = rocdl.grid.dim.z : i64
// CHECK: call range(i32 0, 64) i32 @llvm.amdgcn.workitem.id.x()
- %13 = rocdl.workitem.id.x {range = array<i32: 0, 64>} : i32
+ %13 = rocdl.workitem.id.x range <i32, 0, 64> : i32
+ // CHECK: call range(i64 1, 65) i64 @__ockl_get_local_size(i32 0)
+ %14 = rocdl.workgroup.dim.x range <i32, 1, 65> : i64
llvm.return %1 : i32
}
More information about the Mlir-commits
mailing list