[mlir][tosa] Enhance error_if and verify checks for RESCALE Op #137021
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
Contributor
psunn
commented
Apr 23, 2025
psunn
commented
- add verifier for rank-0 input with per-channel
- add checkErrorIfRescale to tosa validation pass that align with TOSAv1.0
- add LIT tests
|
@llvm/pr-subscribers-mlir-tosa @llvm/pr-subscribers-mlir Author: Peng Sun (psunn) Changes
Full diff: https://github.com/llvm/llvm-project/pull/137021.diff 5 Files Affected:
diff --git a/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp b/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp
index 1ab4ce7d4558b..f1bed1241f971 100644
--- a/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp
+++ b/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp
@@ -3186,6 +3186,12 @@ LogicalResult RescaleOp::verify() {
// otherwise numChannel is dimension in input shape's last axis
int64_t numChannels = 1;
if (getPerChannel()) {
+ if (inputType.getRank() < 1) {
+ emitOpError("requires input to be at least rank 1 when per_channel is "
+ "true, but got rank ")
+ << inputType.getRank();
+ return failure();
+ }
numChannels = inputType.getDimSize(inputType.getRank() - 1);
}
diff --git a/mlir/lib/Dialect/Tosa/Transforms/TosaValidation.cpp b/mlir/lib/Dialect/Tosa/Transforms/TosaValidation.cpp
index baa202833e285..fa337f350197c 100644
--- a/mlir/lib/Dialect/Tosa/Transforms/TosaValidation.cpp
+++ b/mlir/lib/Dialect/Tosa/Transforms/TosaValidation.cpp
@@ -1033,8 +1033,89 @@ bool checkErrorIfTable(Operation *op) {
return true;
}
+bool checkErrorIfRescale(Operation *op) {
+ auto rescale = dyn_cast<tosa::RescaleOp>(op);
+ if (!rescale)
+ return true;
+
+ auto inputType = llvm::dyn_cast<ShapedType>(rescale.getInput().getType());
+ auto outputType = llvm::dyn_cast<ShapedType>(rescale.getOutput().getType());
+ if (!inputType || !outputType || !inputType.getElementType().isInteger() ||
+ !outputType.getElementType().isInteger())
+ return true;
+
+ auto inElemType = inputType.getElementType();
+ auto outElemType = outputType.getElementType();
+ auto inWidth = inElemType.getIntOrFloatBitWidth();
+ auto outWidth = outElemType.getIntOrFloatBitWidth();
+
+ bool inputUnsigned = rescale.getInputUnsigned();
+ bool outputUnsigned = rescale.getOutputUnsigned();
+
+ bool scale32 = rescale.getScale32();
+ auto roundingMode = rescale.getRoundingMode();
+
+
+ // ERROR_IF(scale32 && is_same<in_t,i48_t>())
+ if (scale32 && inWidth == 48) {
+ op->emitOpError() << "scale32 is not allowed with 48-bit input.";
+ return false;
+ }
+
+ // ERROR_IF(!scale32 && (rounding_mode == DOUBLE_ROUND))
+ if (!scale32 && roundingMode == "DOUBLE_ROUND") {
+ op->emitOpError() << "DOUBLE_ROUND is only allowed with scale32=true.";
+ return false;
+ }
+
+ // ERROR_IF(input_unsigned && output_unsigned)
+ if (inputUnsigned && outputUnsigned) {
+ op->emitOpError() << "input and output cannot be both unsigned.";
+ return false;
+ }
+
+ // ERROR_IF(is_same<out_t,i32_t>() && input_unsigned)
+ if (outWidth == 32 && inputUnsigned) {
+ op->emitOpError() << "i32 output type is not allowed with unsigned input.";
+ return false;
+ }
+
+ // ERROR_IF(is_same<in_t,i32_t>() && output_unsigned)
+ if (inWidth == 32 && outputUnsigned) {
+ op->emitOpError() << "i32 input type is not allowed with unsigned output.";
+ return false;
+ }
+
+ // ERROR_IF(is_same<in_t,i48_t>() && output_unsigned)
+ if (inWidth == 48 && outputUnsigned) {
+ op->emitOpError() << "i48 input type is not allowed with unsigned output.";
+ return false;
+ }
+
+ // ERROR_IF(is_same<in_t, i48_t> && input_unsigned)
+ if (inWidth == 48 && inputUnsigned) {
+ op->emitOpError() << "i48 input type cannot be unsigned.";
+ return false;
+ }
+
+ // ERROR_IF(is_same<in_t, i32_t> && input_unsigned)
+ if (inWidth == 32 && inputUnsigned) {
+ op->emitOpError() << "i32 input type cannot be unsigned.";
+ return false;
+ }
+
+ // ERROR_IF(is_same<out_t, i32_t> && output_unsigned)
+ if (outWidth == 32 && outputUnsigned) {
+ op->emitOpError() << "i32 output type cannot be unsigned.";
+ return false;
+ }
+
+ return true;
+}
+
LogicalResult TosaValidation::applyErrorIfCheck(Operation *op) {
- if (!checkErrorIfResize(op) || !checkErrorIfMul(op) || !checkErrorIfTable(op))
+ if (!checkErrorIfResize(op) || !checkErrorIfMul(op) ||
+ !checkErrorIfTable(op) || !checkErrorIfRescale(op))
return failure();
return success();
}
diff --git a/mlir/test/Dialect/Tosa/error_if_check.mlir b/mlir/test/Dialect/Tosa/error_if_check.mlir
index 65a69be91e0c8..c6a173c92ff9a 100644
--- a/mlir/test/Dialect/Tosa/error_if_check.mlir
+++ b/mlir/test/Dialect/Tosa/error_if_check.mlir
@@ -129,3 +129,111 @@ func.func @test_i8_table_size(%arg0: tensor<2x64xi8>, %arg1: tensor<513xi8>) ->
%0 = tosa.table %arg0, %arg1 : (tensor<2x64xi8>, tensor<513xi8>) -> tensor<2x64xi8>
return %0 : tensor<2x64xi8>
}
+
+// -----
+// CHECK-LABEL: test_error_input_zp_not_allowed
+func.func @test_error_input_zp_not_allowed(%arg0: tensor<1xi48>) -> tensor<1xi8> {
+ %multiplier = "tosa.const"() {values = dense<19689> : tensor<1xi32> } : () -> tensor<1xi32>
+ %shift = "tosa.const"() {values = dense<30> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi48>} : () -> tensor<1xi48>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ // expected-error@+1 {{'tosa.rescale' op scale32 is not allowed with 48-bit input}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = true, rounding_mode = "SINGLE_ROUND", per_channel = false, input_unsigned = false, output_unsigned = false} : (tensor<1xi48>, tensor<1xi32>, tensor<1xi8>, tensor<1xi48>, tensor<1xi8>) -> tensor<1xi8>
+ return %0 : tensor<1xi8>
+}
+
+// -----
+// CHECK-LABEL: test_error_scale32_with_i48
+func.func @test_error_scale32_with_i48(%arg0: tensor<1xi48>) -> tensor<1xi8> {
+ %multiplier = "tosa.const"() {values = dense<19689> : tensor<1xi32> } : () -> tensor<1xi32>
+ %shift = "tosa.const"() {values = dense<30> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi48>} : () -> tensor<1xi48>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ // expected-error@+1 {{'tosa.rescale' op scale32 is not allowed with 48-bit input}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = true, rounding_mode = "SINGLE_ROUND", per_channel = false, input_unsigned = false, output_unsigned = false} : (tensor<1xi48>, tensor<1xi32>, tensor<1xi8>, tensor<1xi48>, tensor<1xi8>) -> tensor<1xi8>
+ return %0 : tensor<1xi8>
+}
+
+// -----
+// CHECK-LABEL: test_error_input_output_unsigned
+func.func @test_error_input_output_unsigned(%arg0: tensor<1xi8>) -> tensor<1xi16> {
+ %multiplier = "tosa.const"() {values = dense<19689> : tensor<1xi16> } : () -> tensor<1xi16>
+ %shift = "tosa.const"() {values = dense<30> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi16>} : () -> tensor<1xi16>
+ // expected-error@+1 {{'tosa.rescale' op input and output cannot be both unsigned}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = false, rounding_mode = "SINGLE_ROUND", per_channel = false, input_unsigned = true, output_unsigned = true} : (tensor<1xi8>, tensor<1xi16>, tensor<1xi8>, tensor<1xi8>, tensor<1xi16>) -> tensor<1xi16>
+ return %0 : tensor<1xi16>
+}
+
+// -----
+// CHECK-LABEL: test_error_i32_output_unsigned_input
+func.func @test_error_i32_output_unsigned_input(%arg0: tensor<1xi8>) -> tensor<1xi32> {
+ %multiplier = "tosa.const"() {values = dense<19689> : tensor<1xi16> } : () -> tensor<1xi16>
+ %shift = "tosa.const"() {values = dense<30> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi32>} : () -> tensor<1xi32>
+ // expected-error@+1 {{'tosa.rescale' op i32 output type is not allowed with unsigned input}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = false, rounding_mode = "SINGLE_ROUND", per_channel = false, input_unsigned = true, output_unsigned = false} : (tensor<1xi8>, tensor<1xi16>, tensor<1xi8>, tensor<1xi8>, tensor<1xi32>) -> tensor<1xi32>
+ return %0 : tensor<1xi32>
+}
+
+// -----
+// CHECK-LABEL: test_error_i32_input_unsigned_output
+func.func @test_error_i32_input_unsigned_output(%arg0: tensor<1xi32>) -> tensor<1xi8> {
+ %multiplier = "tosa.const"() {values = dense<19689> : tensor<1xi16> } : () -> tensor<1xi16>
+ %shift = "tosa.const"() {values = dense<30> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi32>} : () -> tensor<1xi32>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ // expected-error@+1 {{'tosa.rescale' op i32 input type is not allowed with unsigned output}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = false, rounding_mode = "SINGLE_ROUND", per_channel = false, input_unsigned = false, output_unsigned = true} : (tensor<1xi32>, tensor<1xi16>, tensor<1xi8>, tensor<1xi32>, tensor<1xi8>) -> tensor<1xi8>
+ return %0 : tensor<1xi8>
+}
+
+// -----
+// CHECK-LABEL: test_error_i48_input_unsigned_output
+func.func @test_error_i48_input_unsigned_output(%arg0: tensor<1xi48>) -> tensor<1xi8> {
+ %multiplier = "tosa.const"() {values = dense<19689> : tensor<1xi16> } : () -> tensor<1xi16>
+ %shift = "tosa.const"() {values = dense<30> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi48>} : () -> tensor<1xi48>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ // expected-error@+1 {{'tosa.rescale' op i48 input type is not allowed with unsigned output}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = false, rounding_mode = "SINGLE_ROUND", per_channel = false, input_unsigned = false, output_unsigned = true} : (tensor<1xi48>, tensor<1xi16>, tensor<1xi8>, tensor<1xi48>, tensor<1xi8>) -> tensor<1xi8>
+ return %0 : tensor<1xi8>
+}
+
+// -----
+// CHECK-LABEL: test_error_i48_unsigned_input
+func.func @test_error_i48_input_unsigned_output(%arg0: tensor<1xi48>) -> tensor<1xi8> {
+ %multiplier = "tosa.const"() {values = dense<19689> : tensor<1xi16> } : () -> tensor<1xi16>
+ %shift = "tosa.const"() {values = dense<30> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi48>} : () -> tensor<1xi48>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ // expected-error@+1 {{'tosa.rescale' op i48 input type cannot be unsigned}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = false, rounding_mode = "SINGLE_ROUND", per_channel = false, input_unsigned = true, output_unsigned = false} : (tensor<1xi48>, tensor<1xi16>, tensor<1xi8>, tensor<1xi48>, tensor<1xi8>) -> tensor<1xi8>
+ return %0 : tensor<1xi8>
+}
+
+// -----
+// CHECK-LABEL: test_error_i32_unsigned_input
+func.func @test_error_i32_input_unsigned_output(%arg0: tensor<1xi32>) -> tensor<1xi8> {
+ %multiplier = "tosa.const"() {values = dense<19689> : tensor<1xi16> } : () -> tensor<1xi16>
+ %shift = "tosa.const"() {values = dense<30> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi32>} : () -> tensor<1xi32>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ // expected-error@+1 {{'tosa.rescale' op i32 input type cannot be unsigned}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = false, rounding_mode = "SINGLE_ROUND", per_channel = false, input_unsigned = true, output_unsigned = false} : (tensor<1xi32>, tensor<1xi16>, tensor<1xi8>, tensor<1xi32>, tensor<1xi8>) -> tensor<1xi8>
+ return %0 : tensor<1xi8>
+}
+
+// -----
+// CHECK-LABEL: test_error_i32_unsigned_output
+func.func @test_error_i32_unsigned_output(%arg0: tensor<1xi8>) -> tensor<1xi32> {
+ %multiplier = "tosa.const"() {values = dense<19689> : tensor<1xi16> } : () -> tensor<1xi16>
+ %shift = "tosa.const"() {values = dense<30> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi32>} : () -> tensor<1xi32>
+ // expected-error@+1 {{'tosa.rescale' op i32 output type cannot be unsigned}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = false, rounding_mode = "SINGLE_ROUND", per_channel = false, input_unsigned = false, output_unsigned = true} : (tensor<1xi8>, tensor<1xi16>, tensor<1xi8>, tensor<1xi8>, tensor<1xi32>) -> tensor<1xi32>
+ return %0 : tensor<1xi32>
+}
diff --git a/mlir/test/Dialect/Tosa/invalid.mlir b/mlir/test/Dialect/Tosa/invalid.mlir
index 269ed58fdc81c..fe4cc49e89c0d 100644
--- a/mlir/test/Dialect/Tosa/invalid.mlir
+++ b/mlir/test/Dialect/Tosa/invalid.mlir
@@ -1669,6 +1669,18 @@ func.func @test_rescale_invalid_non_perchannel_shift_shape(%arg0: tensor<13x21x3
return %0 : tensor<13x21x3xi16>
}
+// -----
+// CHECK-LABEL: test_error_double_round_without_scale32
+func.func @test_error_double_round_without_scale32(%arg0: tensor<1xi8>) -> tensor<1xi16> {
+ %multiplier = "tosa.const"() {values = dense<19689> : tensor<1xi16> } : () -> tensor<1xi16>
+ %shift = "tosa.const"() {values = dense<30> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi16>} : () -> tensor<1xi16>
+ // expected-error@+1 {{'tosa.rescale' op DOUBLE_ROUND is only allowed with scale32=true}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = false, rounding_mode = "DOUBLE_ROUND", per_channel = false, input_unsigned = false, output_unsigned = false} : (tensor<1xi8>, tensor<1xi16>, tensor<1xi8>, tensor<1xi8>, tensor<1xi16>) -> tensor<1xi16>
+ return %0 : tensor<1xi16>
+}
+
// -----
// CHECK-LABEL: test_matmul_a_zp_same_element_type
func.func @test_matmul_a_zp_same_element_type(%arg0: tensor<1x14x19xf32>, %arg1: tensor<1x19x28xf32>) -> tensor<1x14x28xf32> {
diff --git a/mlir/test/Dialect/Tosa/verifier.mlir b/mlir/test/Dialect/Tosa/verifier.mlir
index fb8726cba1853..a42cf03a0a5cb 100644
--- a/mlir/test/Dialect/Tosa/verifier.mlir
+++ b/mlir/test/Dialect/Tosa/verifier.mlir
@@ -319,3 +319,16 @@ func.func @test_conv3d_wholly_divisible_output_width(%arg0: tensor<1x4x8x21x19xf
: (tensor<1x4x8x21x19xf32>, tensor<34x1x1x1x17xf32>, tensor<21xf32>, tensor<1xf32>, tensor<1xf32>) -> tensor<1x4x8x19x34xf32>
return %0 : tensor<1x4x8x19x34xf32>
}
+
+// -----
+
+// CHECK-LABEL: test_error_scalar_input_with_per_channel
+func.func @test_error_scalar_input_with_per_channel(%arg0: tensor<i8>) -> tensor<i16> {
+ %multiplier = "tosa.const"() {values = dense<4> : tensor<1xi32> } : () -> tensor<1xi32>
+ %shift = "tosa.const"() {values = dense<2> : tensor<1xi8> } : () -> tensor<1xi8>
+ %input_zp = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1xi8>
+ %output_zp = "tosa.const"() {values = dense<0> : tensor<1xi16>} : () -> tensor<1xi16>
+ // expected-error@+1 {{'tosa.rescale' op requires input to be at least rank 1 when per_channel is true, but got rank 0}}
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {scale32 = true, rounding_mode = "SINGLE_ROUND", per_channel = true, input_unsigned = false, output_unsigned = false} : (tensor<i8>, tensor<1xi32>, tensor<1xi8>, tensor<1xi8>, tensor<1xi16>) -> tensor<i16>
+ return %0 : tensor<i16>
+}
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
psunn
force-pushed
the
psunn/rescale_errorif
branch
from
621c120 to
e0d037f
Compare
|
Thanks for the patch. Looks good to me overall. |
lhutton1
approved these changes
Apr 24, 2025
* add verifier for rank-0 input with per-channel * add checkErrorIfRescale to tosa validation pass that align with TOSAv1.0 * add LIT tests Change-Id: Ia07e8c2ee66d8ee4113bea5ad9fa859b5986b009 Signed-off-by: Peng Sun <peng.sun@arm.com>
psunn
force-pushed
the
psunn/rescale_errorif
branch
from
e0d037f to
99ea7d6
Compare
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.