[mlir][arith] Improve extf folder
#80232
Merged
Conversation
This file contains hidden or 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
kuhar
requested review from
antiagainst,
Hardcode84,
zero9178,
Mogball,
dcaballe and
Groverkss
|
@llvm/pr-subscribers-mlir Author: Jakub Kuderski (kuhar) Changes
Full diff: https://github.com/llvm/llvm-project/pull/80232.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Arith/IR/ArithOps.cpp b/mlir/lib/Dialect/Arith/IR/ArithOps.cpp
index 270df3f3f9e99..7b026310ab7d6 100644
--- a/mlir/lib/Dialect/Arith/IR/ArithOps.cpp
+++ b/mlir/lib/Dialect/Arith/IR/ArithOps.cpp
@@ -9,6 +9,7 @@
#include <cassert>
#include <cstdint>
#include <functional>
+#include <optional>
#include <utility>
#include "mlir/Dialect/Arith/IR/Arith.h"
@@ -1258,6 +1259,21 @@ static bool checkWidthChangeCast(TypeRange inputs, TypeRange outputs) {
srcType.getIntOrFloatBitWidth());
}
+/// Attempts to convert `sourceValue` to an APFloat value with
+/// `targetSemantics`, without any information loss or rounding. Return
+/// std::nullopt on failure.
+static std::optional<APFloat>
+convertFloatValue(APFloat sourceValue,
+ const llvm::fltSemantics &targetSemantics) {
+ bool losesInfo = false;
+ auto status = sourceValue.convert(
+ targetSemantics, llvm::RoundingMode::NearestTiesToEven, &losesInfo);
+ if (losesInfo || status != APFloat::opOK)
+ return std::nullopt;
+
+ return sourceValue;
+}
+
//===----------------------------------------------------------------------===//
// ExtUIOp
//===----------------------------------------------------------------------===//
@@ -1321,14 +1337,21 @@ LogicalResult arith::ExtSIOp::verify() {
// ExtFOp
//===----------------------------------------------------------------------===//
-/// Always fold extension of FP constants.
+/// Fold extension of float constants when there is no information loss due the
+/// difference in fp semantics.
OpFoldResult arith::ExtFOp::fold(FoldAdaptor adaptor) {
- auto constOperand = llvm::dyn_cast_if_present<FloatAttr>(adaptor.getIn());
- if (!constOperand)
- return {};
+ auto resElemType = cast<FloatType>(getElementTypeOrSelf(getType()));
+ const llvm::fltSemantics &targetSemantics = resElemType.getFloatSemantics();
+ return constFoldCastOp<FloatAttr, FloatAttr>(
+ adaptor.getOperands(), getType(),
+ [&targetSemantics](const APFloat &a, bool &castStatus) {
+ if (std::optional<APFloat> result =
+ convertFloatValue(a, targetSemantics))
+ return *result;
- // Convert to target type via 'double'.
- return FloatAttr::get(getType(), constOperand.getValue().convertToDouble());
+ castStatus = false;
+ return a;
+ });
}
bool arith::ExtFOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
@@ -1403,11 +1426,12 @@ OpFoldResult arith::TruncFOp::fold(FoldAdaptor adaptor) {
const llvm::fltSemantics &targetSemantics = resElemType.getFloatSemantics();
return constFoldCastOp<FloatAttr, FloatAttr>(
adaptor.getOperands(), getType(),
- [&targetSemantics](APFloat a, bool &castStatus) {
- bool losesInfo = false;
- auto status = a.convert(
- targetSemantics, llvm::RoundingMode::NearestTiesToEven, &losesInfo);
- castStatus = !losesInfo && status == APFloat::opOK;
+ [&targetSemantics](const APFloat &a, bool &castStatus) {
+ if (std::optional<APFloat> result =
+ convertFloatValue(a, targetSemantics))
+ return *result;
+
+ castStatus = false;
return a;
});
}
@@ -1496,6 +1520,7 @@ OpFoldResult arith::SIToFPOp::fold(FoldAdaptor adaptor) {
return apf;
});
}
+
//===----------------------------------------------------------------------===//
// FPToUIOp
//===----------------------------------------------------------------------===//
diff --git a/mlir/test/Dialect/Arith/canonicalize.mlir b/mlir/test/Dialect/Arith/canonicalize.mlir
index 44df11ab2433a..57a71bcc8feeb 100644
--- a/mlir/test/Dialect/Arith/canonicalize.mlir
+++ b/mlir/test/Dialect/Arith/canonicalize.mlir
@@ -701,6 +701,15 @@ func.func @extFPConstant() -> f64 {
return %0 : f64
}
+// CHECK-LABEL: @extFPVectorConstant
+// CHECK: %[[cres:.+]] = arith.constant dense<[0.000000e+00, 1.000000e+00]> : vector<2xf128>
+// CHECK: return %[[cres]]
+func.func @extFPVectorConstant() -> vector<2xf128> {
+ %cst = arith.constant dense<[0.000000e+00, 1.000000e+00]> : vector<2xf80>
+ %0 = arith.extf %cst : vector<2xf80> to vector<2xf128>
+ return %0 : vector<2xf128>
+}
+
// CHECK-LABEL: @truncConstant
// CHECK: %[[cres:.+]] = arith.constant -2 : i16
// CHECK: return %[[cres]]
|
|
@llvm/pr-subscribers-mlir-arith Author: Jakub Kuderski (kuhar) Changes
Full diff: https://github.com/llvm/llvm-project/pull/80232.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Arith/IR/ArithOps.cpp b/mlir/lib/Dialect/Arith/IR/ArithOps.cpp
index 270df3f3f9e99..7b026310ab7d6 100644
--- a/mlir/lib/Dialect/Arith/IR/ArithOps.cpp
+++ b/mlir/lib/Dialect/Arith/IR/ArithOps.cpp
@@ -9,6 +9,7 @@
#include <cassert>
#include <cstdint>
#include <functional>
+#include <optional>
#include <utility>
#include "mlir/Dialect/Arith/IR/Arith.h"
@@ -1258,6 +1259,21 @@ static bool checkWidthChangeCast(TypeRange inputs, TypeRange outputs) {
srcType.getIntOrFloatBitWidth());
}
+/// Attempts to convert `sourceValue` to an APFloat value with
+/// `targetSemantics`, without any information loss or rounding. Return
+/// std::nullopt on failure.
+static std::optional<APFloat>
+convertFloatValue(APFloat sourceValue,
+ const llvm::fltSemantics &targetSemantics) {
+ bool losesInfo = false;
+ auto status = sourceValue.convert(
+ targetSemantics, llvm::RoundingMode::NearestTiesToEven, &losesInfo);
+ if (losesInfo || status != APFloat::opOK)
+ return std::nullopt;
+
+ return sourceValue;
+}
+
//===----------------------------------------------------------------------===//
// ExtUIOp
//===----------------------------------------------------------------------===//
@@ -1321,14 +1337,21 @@ LogicalResult arith::ExtSIOp::verify() {
// ExtFOp
//===----------------------------------------------------------------------===//
-/// Always fold extension of FP constants.
+/// Fold extension of float constants when there is no information loss due the
+/// difference in fp semantics.
OpFoldResult arith::ExtFOp::fold(FoldAdaptor adaptor) {
- auto constOperand = llvm::dyn_cast_if_present<FloatAttr>(adaptor.getIn());
- if (!constOperand)
- return {};
+ auto resElemType = cast<FloatType>(getElementTypeOrSelf(getType()));
+ const llvm::fltSemantics &targetSemantics = resElemType.getFloatSemantics();
+ return constFoldCastOp<FloatAttr, FloatAttr>(
+ adaptor.getOperands(), getType(),
+ [&targetSemantics](const APFloat &a, bool &castStatus) {
+ if (std::optional<APFloat> result =
+ convertFloatValue(a, targetSemantics))
+ return *result;
- // Convert to target type via 'double'.
- return FloatAttr::get(getType(), constOperand.getValue().convertToDouble());
+ castStatus = false;
+ return a;
+ });
}
bool arith::ExtFOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
@@ -1403,11 +1426,12 @@ OpFoldResult arith::TruncFOp::fold(FoldAdaptor adaptor) {
const llvm::fltSemantics &targetSemantics = resElemType.getFloatSemantics();
return constFoldCastOp<FloatAttr, FloatAttr>(
adaptor.getOperands(), getType(),
- [&targetSemantics](APFloat a, bool &castStatus) {
- bool losesInfo = false;
- auto status = a.convert(
- targetSemantics, llvm::RoundingMode::NearestTiesToEven, &losesInfo);
- castStatus = !losesInfo && status == APFloat::opOK;
+ [&targetSemantics](const APFloat &a, bool &castStatus) {
+ if (std::optional<APFloat> result =
+ convertFloatValue(a, targetSemantics))
+ return *result;
+
+ castStatus = false;
return a;
});
}
@@ -1496,6 +1520,7 @@ OpFoldResult arith::SIToFPOp::fold(FoldAdaptor adaptor) {
return apf;
});
}
+
//===----------------------------------------------------------------------===//
// FPToUIOp
//===----------------------------------------------------------------------===//
diff --git a/mlir/test/Dialect/Arith/canonicalize.mlir b/mlir/test/Dialect/Arith/canonicalize.mlir
index 44df11ab2433a..57a71bcc8feeb 100644
--- a/mlir/test/Dialect/Arith/canonicalize.mlir
+++ b/mlir/test/Dialect/Arith/canonicalize.mlir
@@ -701,6 +701,15 @@ func.func @extFPConstant() -> f64 {
return %0 : f64
}
+// CHECK-LABEL: @extFPVectorConstant
+// CHECK: %[[cres:.+]] = arith.constant dense<[0.000000e+00, 1.000000e+00]> : vector<2xf128>
+// CHECK: return %[[cres]]
+func.func @extFPVectorConstant() -> vector<2xf128> {
+ %cst = arith.constant dense<[0.000000e+00, 1.000000e+00]> : vector<2xf80>
+ %0 = arith.extf %cst : vector<2xf80> to vector<2xf128>
+ return %0 : vector<2xf128>
+}
+
// CHECK-LABEL: @truncConstant
// CHECK: %[[cres:.+]] = arith.constant -2 : i16
// CHECK: return %[[cres]]
|
Lewuathe
reviewed
Feb 1, 2024
dcaballe
reviewed
Feb 2, 2024
* Use APFloat conversion function to avoid losing information by converting to `double`. This would be the case with large types like `f80` or `f128`. * Check for potential information loss. This is intended for small floating point types that may have values not present in larger ones (e.g., negative zero). * Support folding vector constants.
rsuderman
approved these changes
Feb 2, 2024
dcaballe
approved these changes
Feb 2, 2024
agozillon
pushed a commit
to agozillon/llvm-project
that referenced
this pull request
Feb 5, 2024
* Use APFloat conversion function to avoid losing information by converting to `double`. This would be the case with large types like `f80` or `f128`. * Check for potential information loss. This is intended for small floating point types that may have values not present in larger ones (e.g., f8m2e5fnuz and f16). * Support folding vector constants.
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.
double. This would be the case with large types likef80orf128.