[Mlir-commits] [mlir] [mlir][tosa] Add NaN Propagation Mode Support (PR #121951)
Jack Frankland
llvmlistbot at llvm.org
Wed Jan 22 07:56:37 PST 2025
https://github.com/FranklandJack updated https://github.com/llvm/llvm-project/pull/121951
>From fd01630c64fe137eca9ec5e82c641d3aeb2dbc7a Mon Sep 17 00:00:00 2001
From: TatWai Chong <tatwai.chong at arm.com>
Date: Thu, 3 Oct 2024 15:18:55 -0700
Subject: [PATCH] [mlir][tosa] Add NaN Propagation Mode Support
The TOSA-V1.0 specification adds "nan propagation" modes as attributes
for several operators. Adjust the ODS definitions of the relevant
operations to include this attribute.
The defined modes are "PROPAGATE" and "IGNORE" and the PROPAGATE mode is
set by default.
MAXIMUM, MINIMUM, REDUCE_MAX, REDUCE_MIN, MAX_POOL, CLAMP, and ARGMAX
support this attribute.
Refactor the clamp + clamp optimization in order to better handle edge
cases such as invalid NaN propagation combinations and disjoint clamp
ranges.
Add lit test including negative tests for valid and invalid
optimizations as part of TOSA canonicalizations.
Signed-off-by: Jack Frankland <jack.frankland at arm.com>
---
mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td | 21 +++--
.../mlir/Dialect/Tosa/IR/TosaTypesBase.td | 8 ++
.../Dialect/Tosa/IR/TosaCanonicalizations.cpp | 85 +++++++++++++++----
mlir/test/Dialect/Tosa/canonicalize.mlir | 52 ++++++++++++
4 files changed, 142 insertions(+), 24 deletions(-)
diff --git a/mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td b/mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td
index e1efa7a3001b9f..cc39a2e51e3a9c 100644
--- a/mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td
+++ b/mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td
@@ -42,7 +42,8 @@ def Tosa_ArgMaxOp : Tosa_InferShapedTypeOp<"argmax"> {
let arguments = (ins
Tosa_Tensor: $input,
- I32Attr: $axis
+ I32Attr: $axis,
+ DefaultValuedAttr<Tosa_NanPropagationAttr, "\"PROPAGATE\"">:$nan_mode
);
let results = (outs
@@ -287,7 +288,8 @@ def Tosa_MaxPool2dOp : Tosa_InferShapedTypeOp<"max_pool2d"> {
Tosa_IntArrayAttr2:$kernel,
Tosa_IntArrayAttr2:$stride,
- Tosa_IntArrayAttr4:$pad
+ Tosa_IntArrayAttr4:$pad,
+ DefaultValuedAttr<Tosa_NanPropagationAttr, "\"PROPAGATE\"">:$nan_mode
);
let results = (outs
@@ -388,7 +390,8 @@ def Tosa_ClampOp : Tosa_ElementwiseUnaryOp<"clamp"> {
I64Attr:$min_int,
I64Attr:$max_int,
Tosa_FloatAttr:$min_fp,
- Tosa_FloatAttr:$max_fp
+ Tosa_FloatAttr:$max_fp,
+ DefaultValuedAttr<Tosa_NanPropagationAttr, "\"PROPAGATE\"">:$nan_mode
);
let results = (outs
@@ -752,7 +755,8 @@ def Tosa_MaximumOp : Tosa_ElementwiseOp<"maximum", [
let arguments = (ins
Tosa_Tensor:$input1,
- Tosa_Tensor:$input2
+ Tosa_Tensor:$input2,
+ DefaultValuedAttr<Tosa_NanPropagationAttr, "\"PROPAGATE\"">:$nan_mode
);
let results = (outs
@@ -775,7 +779,8 @@ def Tosa_MinimumOp : Tosa_ElementwiseOp<"minimum", [
let arguments = (ins
Tosa_Tensor:$input1,
- Tosa_Tensor:$input2
+ Tosa_Tensor:$input2,
+ DefaultValuedAttr<Tosa_NanPropagationAttr, "\"PROPAGATE\"">:$nan_mode
);
let results = (outs
@@ -1382,7 +1387,8 @@ def Tosa_ReduceMaxOp : Tosa_InferTensorTypeOp<"reduce_max"> {
let arguments = (ins
Tosa_Tensor:$input,
- I32Attr:$axis
+ I32Attr:$axis,
+ DefaultValuedAttr<Tosa_NanPropagationAttr, "\"PROPAGATE\"">:$nan_mode
);
let results = (outs
@@ -1417,7 +1423,8 @@ def Tosa_ReduceMinOp : Tosa_InferTensorTypeOp<"reduce_min"> {
let arguments = (ins
Tosa_Tensor:$input,
- I32Attr:$axis
+ I32Attr:$axis,
+ DefaultValuedAttr<Tosa_NanPropagationAttr, "\"PROPAGATE\"">:$nan_mode
);
let results = (outs
diff --git a/mlir/include/mlir/Dialect/Tosa/IR/TosaTypesBase.td b/mlir/include/mlir/Dialect/Tosa/IR/TosaTypesBase.td
index 13325fb0ab9a20..5693acf3a01db4 100644
--- a/mlir/include/mlir/Dialect/Tosa/IR/TosaTypesBase.td
+++ b/mlir/include/mlir/Dialect/Tosa/IR/TosaTypesBase.td
@@ -205,12 +205,20 @@ def Tosa_FloatAttr : Attr<CPred<"::llvm::isa<::mlir::FloatAttr>($_self)">,
//===----------------------------------------------------------------------===//
// Iterable attributes.
//===----------------------------------------------------------------------===//
+// Defined in `section 3. Enumerations` of the TOSA specification.
+
// Supported regimes for tosa.resize.
def Tosa_ResizeTypeAttr : StringBasedAttr<
CPred<"::llvm::cast<StringAttr>($_self).getValue() == \"BILINEAR\" || " #
"::llvm::cast<StringAttr>($_self).getValue() == \"NEAREST_NEIGHBOR\"">,
"Supported resize/upsampling strategies">;
+// Supported NaN propagation strategies.
+def Tosa_NanPropagationAttr : StringBasedAttr<
+ CPred<"::llvm::cast<StringAttr>($_self).getValue() == \"PROPAGATE\" || " #
+ "::llvm::cast<StringAttr>($_self).getValue() == \"IGNORE\"">,
+ "Supported NaN propagation strategies">;
+
def Tosa_TensorTypeAttr : TypeAttrBase<"TensorType", "Tensor type attribute">;
// Tensor to buffer types.
diff --git a/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp b/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp
index f7a596f1ccb192..5d2b17095b6076 100644
--- a/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp
+++ b/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp
@@ -339,33 +339,84 @@ struct ClampIsNoOp : public OpRewritePattern<tosa::ClampOp> {
}
};
+// Attempts the following transformation:
+//
+// For integers a, b, a', and b' such that [a, b] ∩ [c, d] ≠ ∅ and input
+// tensor X the following identity holds:
+//
+// CLAMP(CLAMP(X, a, b), a', b') = CLAMP(X, max(a, a'), min(b, b'))
+//
+// subject to the following valid NaN propagation semantics:
+// --------------------------------------------
+// | opNanMode | clampNanMode | resultNanMode |
+// |-----------|--------------|---------------|
+// | PROPAGATE | PROPAGATE | PROPAGATE |
+// | PROPAGATE | IGNORE | IGNORE |
+// | IGNORE | PROPAGATE | INVALID |
+// | IGNORE | IGNORE | IGNORE |
+// |------------------------------------------|
+
struct ClampClampOptimization : public OpRewritePattern<tosa::ClampOp> {
using OpRewritePattern<tosa::ClampOp>::OpRewritePattern;
+ // Helper structure to describe the range of a clamp operation.
+ template <typename T>
+ struct ClampRange {
+ ClampRange(const T &start, const T &end) : start(start), end(end) {}
+ T start;
+ T end;
+
+ // Helper function to determine if two Clamp ranges intersect.
+ bool intersects(const ClampRange<T> &otherRange) {
+ return start < otherRange.end && otherRange.start < end;
+ }
+ };
+
LogicalResult matchAndRewrite(tosa::ClampOp op,
PatternRewriter &rewriter) const override {
- Value input = op.getInput();
-
- Operation *definingOp = input.getDefiningOp();
- if (!definingOp)
+ // Check the input to the CLAMP op is itself a CLAMP.
+ auto clampOp =
+ dyn_cast_if_present<tosa::ClampOp>(op.getInput().getDefiningOp());
+ if (!clampOp)
return failure();
- if (tosa::ClampOp clampOp = dyn_cast<tosa::ClampOp>(definingOp)) {
- auto minFp = std::max(op.getMinFp(), clampOp.getMinFp()).convertToFloat();
- auto maxFp = std::min(op.getMaxFp(), clampOp.getMaxFp()).convertToFloat();
+ // Check we have a valid NaN propagation combination.
+ const auto opNanMode = op.getNanMode();
+ const auto clampNanMode = clampOp.getNanMode();
+ if (opNanMode == "IGNORE" && clampNanMode == "PROPAGATE")
+ return failure();
- auto minInt = std::max(op.getMinInt(), clampOp.getMinInt());
- auto maxInt = std::min(op.getMaxInt(), clampOp.getMaxInt());
+ // Check we have intersecting ranges.
+ const auto opMinInt = op.getMinInt();
+ const auto opMaxInt = op.getMaxInt();
+ const auto clampOpMinInt = clampOp.getMinInt();
+ const auto clampOpMaxInt = clampOp.getMaxInt();
+ ClampRange<std::int64_t> opRangeIntRange(opMinInt, opMaxInt);
+ ClampRange<std::int64_t> clampRangeIntRange(clampOpMinInt, clampOpMaxInt);
+ if (!opRangeIntRange.intersects(clampRangeIntRange))
+ return failure();
- rewriter.replaceOpWithNewOp<tosa::ClampOp>(
- op, op.getType(), clampOp.getInput(),
- rewriter.getI64IntegerAttr(minInt),
- rewriter.getI64IntegerAttr(maxInt), rewriter.getF32FloatAttr(minFp),
- rewriter.getF32FloatAttr(maxFp));
- return success();
- }
+ const auto opMinFloat = op.getMinFp();
+ const auto opMaxFloat = op.getMaxFp();
+ const auto clampOpMinFloat = clampOp.getMinFp();
+ const auto clampOpMaxFloat = clampOp.getMaxFp();
+ ClampRange opRangeFloatRange(opMinFloat, opMaxFloat);
+ ClampRange clampRangeFloatRange(clampOpMinFloat, clampOpMaxFloat);
+ if (!opRangeFloatRange.intersects(clampRangeFloatRange))
+ return failure();
- return failure();
+ // Run the transformation.
+ const auto minFp = std::max(opMinFloat, clampOpMinFloat).convertToFloat();
+ const auto maxFp = std::min(opMaxFloat, clampOpMaxFloat).convertToFloat();
+ const auto minInt = std::max(opMinInt, clampOpMinInt);
+ const auto maxInt = std::min(opMaxInt, clampOpMaxInt);
+ rewriter.replaceOpWithNewOp<tosa::ClampOp>(
+ op, op.getType(), clampOp.getInput(),
+ rewriter.getI64IntegerAttr(minInt), rewriter.getI64IntegerAttr(maxInt),
+ rewriter.getF32FloatAttr(minFp), rewriter.getF32FloatAttr(maxFp),
+ rewriter.getStringAttr((opNanMode != clampNanMode) ? "IGNORE"
+ : opNanMode));
+ return success();
}
};
diff --git a/mlir/test/Dialect/Tosa/canonicalize.mlir b/mlir/test/Dialect/Tosa/canonicalize.mlir
index 889e2eda9e5b84..52e121e511b70b 100644
--- a/mlir/test/Dialect/Tosa/canonicalize.mlir
+++ b/mlir/test/Dialect/Tosa/canonicalize.mlir
@@ -138,6 +138,58 @@ func.func @clamp_twice_is_single_clamp(%arg0: tensor<4xi8>) -> tensor<4xi8> {
// -----
+// CHECK: @disjoint_clamp_twice_is_not_single_clamp(%[[INPUT:.*]]: tensor<4xi8>)
+func.func @disjoint_clamp_twice_is_not_single_clamp(%arg0: tensor<4xi8>) -> tensor<4xi8> {
+ // CHECK: %[[CLAMP_1:.*]] = tosa.clamp %[[INPUT]] {max_fp = -5.000000e+00 : f32, max_int = -5 : i64, min_fp = -1.000000e+00 : f32, min_int = -10 : i64} : (tensor<4xi8>) -> tensor<4xi8>
+ // CHECK-NEXT: tosa.clamp %[[CLAMP_1]] {max_fp = 5.000000e+00 : f32, max_int = 5 : i64, min_fp = 1.000000e+00 : f32, min_int = 1 : i64} : (tensor<4xi8>) -> tensor<4xi8>
+ %0 = tosa.clamp %arg0 {max_fp = -5.0 : f32, max_int = -5 : i64, min_fp = -1.0 : f32, min_int = -10 : i64} : (tensor<4xi8>) -> tensor<4xi8>
+ %1 = tosa.clamp %0 {max_fp = 5.0 : f32, max_int = 5 : i64, min_fp = 1.0 : f32, min_int = 1 : i64} : (tensor<4xi8>) -> tensor<4xi8>
+ return %1 : tensor<4xi8>
+}
+
+// -----
+
+// CHECK-LABEL: @clamp_twice_with_nan_propagate_is_single_clamp
+func.func @clamp_twice_with_nan_propagate_is_single_clamp(%arg0: tensor<4xi8>) -> tensor<4xi8> {
+ // CHECK: tosa.clamp %arg0 {max_fp = 3.000000e+00 : f32, max_int = 2 : i64, min_fp = -3.000000e+00 : f32, min_int = -2 : i64}
+ %0 = tosa.clamp %arg0 {max_fp = 3.0 : f32, max_int = 4 : i64, min_fp = -5.0 : f32, min_int = -2 : i64, nan_mode = "PROPAGATE"} : (tensor<4xi8>) -> tensor<4xi8>
+ %1 = tosa.clamp %0 {max_fp = 5.0 : f32, max_int = 2 : i64, min_fp = -3.0 : f32, min_int = -4 : i64, nan_mode = "PROPAGATE"} : (tensor<4xi8>) -> tensor<4xi8>
+ return %1 : tensor<4xi8>
+}
+
+// -----
+
+// CHECK-LABEL: @clamp_twice_with_nan_ignore_is_single_clamp
+func.func @clamp_twice_with_nan_ignore_is_single_clamp(%arg0: tensor<4xi8>) -> tensor<4xi8> {
+ // CHECK: tosa.clamp %arg0 {max_fp = 3.000000e+00 : f32, max_int = 2 : i64, min_fp = -3.000000e+00 : f32, min_int = -2 : i64, nan_mode = "IGNORE"}
+ %0 = tosa.clamp %arg0 {max_fp = 3.0 : f32, max_int = 4 : i64, min_fp = -5.0 : f32, min_int = -2 : i64, nan_mode = "IGNORE"} : (tensor<4xi8>) -> tensor<4xi8>
+ %1 = tosa.clamp %0 {max_fp = 5.0 : f32, max_int = 2 : i64, min_fp = -3.0 : f32, min_int = -4 : i64, nan_mode = "IGNORE"} : (tensor<4xi8>) -> tensor<4xi8>
+ return %1 : tensor<4xi8>
+}
+
+// -----
+
+// CHECK-LABEL: @clamp_twice_with_nan_ignore_propagate_is_single_clamp
+func.func @clamp_twice_with_nan_ignore_propagate_is_single_clamp(%arg0: tensor<4xi8>) -> tensor<4xi8> {
+ // CHECK: tosa.clamp %arg0 {max_fp = 3.000000e+00 : f32, max_int = 2 : i64, min_fp = -3.000000e+00 : f32, min_int = -2 : i64, nan_mode = "IGNORE"}
+ %0 = tosa.clamp %arg0 {max_fp = 3.0 : f32, max_int = 4 : i64, min_fp = -5.0 : f32, min_int = -2 : i64, nan_mode = "IGNORE"} : (tensor<4xi8>) -> tensor<4xi8>
+ %1 = tosa.clamp %0 {max_fp = 5.0 : f32, max_int = 2 : i64, min_fp = -3.0 : f32, min_int = -4 : i64, nan_mode = "PROPAGATE"} : (tensor<4xi8>) -> tensor<4xi8>
+ return %1 : tensor<4xi8>
+}
+
+// -----
+
+// CHECK: @clamp_twice_with_nan_propagate_ignore_is_not_single_clamp(%[[INPUT:.*]]: tensor<4xi8>)
+func.func @clamp_twice_with_nan_propagate_ignore_is_not_single_clamp(%arg0: tensor<4xi8>) -> tensor<4xi8> {
+ // CHECK: %[[CLAMP_1:.*]] = tosa.clamp %[[INPUT]] {max_fp = 3.000000e+00 : f32, max_int = 4 : i64, min_fp = -5.000000e+00 : f32, min_int = -2 : i64} : (tensor<4xi8>) -> tensor<4xi8>
+ // CHECK-NEXT: tosa.clamp %[[CLAMP_1]] {max_fp = 5.000000e+00 : f32, max_int = 2 : i64, min_fp = -3.000000e+00 : f32, min_int = -4 : i64, nan_mode = "IGNORE"} : (tensor<4xi8>) -> tensor<4xi8>
+ %0 = tosa.clamp %arg0 {max_fp = 3.0 : f32, max_int = 4 : i64, min_fp = -5.0 : f32, min_int = -2 : i64, nan_mode = "PROPAGATE"} : (tensor<4xi8>) -> tensor<4xi8>
+ %1 = tosa.clamp %0 {max_fp = 5.0 : f32, max_int = 2 : i64, min_fp = -3.0 : f32, min_int = -4 : i64, nan_mode = "IGNORE"} : (tensor<4xi8>) -> tensor<4xi8>
+ return %1 : tensor<4xi8>
+}
+
+// -----
+
// CHECK-LABEL: @concat_fold
func.func @concat_fold(%arg0: tensor<?x1xf32>) -> tensor<?x1xf32> {
// CHECK: return %arg0
More information about the Mlir-commits
mailing list