[MLIR][TOSA] Fix validation for unsigned integer types in RescaleOp #137838
Conversation
This patch fixes a bug in the TOSA RescaleOp verifier that incorrectly rejects unsigned integer types (ui8, ui16), even though they are supported by the TOSA specification. The verifier now properly handles unsigned integer types when the corresponding input_unsigned or output_unsigned attribute is set to true. Added tests for ui8<->i8 and ui16<->i16 rescale operations. Fixes llvm#135699
|
Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be notified. 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 permissions for the repository. In which case you can instead tag reviewers by name in a comment by using If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers. 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-tosa Author: NohHyeon Kwon (swote-git) ChangesThis patch fixes a bug in the TOSA RescaleOp verifier that incorrectly rejects unsigned integer types (ui8, ui16), even though they are supported by the TOSA specification. The verifier now properly handles unsigned integer types when the corresponding input_unsigned or output_unsigned attribute is set to true. Added tests for ui8<->i8 and ui16<->i16 rescale operations. Fixes #135699 Full diff: https://github.com/llvm/llvm-project/pull/137838.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp b/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp
index b2e471f2bba93..980ef18b975f9 100644
--- a/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp
+++ b/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp
@@ -2111,24 +2111,30 @@ static LogicalResult verifyZeroPoint(tosa::RescaleOp op, Value zpVal,
const int64_t &zp,
const std::string &operand) {
bool isInputZp = (operand == "Input");
-
bool tensorUnsigned =
- isInputZp ? op.getInputUnsigned() : op.getOutputUnsigned();
+ isInputZp ? op.getInputUnsigned() : op.getOutputUnsigned();
StringRef tensorName = isInputZp ? "input" : "output";
-
Type zpElemType = getElementTypeOrSelf(zpVal);
if (zp != 0) {
- if (!zpElemType.isInteger(8) &&
- !(zpElemType.isInteger(16) && tensorUnsigned)) {
- return op.emitOpError()
- << "expect " << tensorName << "_zp of 0, got " << zp;
+ bool validType = zpElemType.isInteger(8);
+
+ if (tensorUnsigned && zpElemType.isInteger(8)) {
+ validType = true;
}
- if (zpElemType.isInteger(16) && tensorUnsigned &&
- zp != static_cast<int16_t>(32768)) {
- return op.emitOpError() << "expect " << tensorName
- << "_zp of 0 or 32768 for unsigned int16 "
- << tensorName << ", got " << zp;
+
+ if (zpElemType.isInteger(16) && tensorUnsigned) {
+ validType = true;
+ if (zp != 32768) {
+ return op.emitOpError() << "expect " << tensorName
+ << "_zp of 0 or 32768 for unsigned int16 "
+ << tensorName << ", got " << zp;
+ }
+ }
+
+ if (!validType) {
+ return op.emitOpError()
+ << "expect " << tensorName << "_zp of 0, got " << zp;
}
}
diff --git a/mlir/test/Dialect/Tosa/availability.mlir b/mlir/test/Dialect/Tosa/availability.mlir
index 75126a11ac504..08d2bd30cf971 100644
--- a/mlir/test/Dialect/Tosa/availability.mlir
+++ b/mlir/test/Dialect/Tosa/availability.mlir
@@ -622,6 +622,28 @@ func.func @test_rescale(%arg0: tensor<13x21x3x!quant.uniform<u8:f32, 0.015655439
return %0 : tensor<13x21x3x!quant.uniform<i8:f32, 0.015655439347028732:-1>>
}
+// -----
+// CHECK-LABEL: test_rescale
+func.func @test_rescale_unsigned_i8(%arg0: tensor<13x21x3x!quant.uniform<u8:f32, 0.015655439347028732:127>>, %multiplier : tensor<1xi32>, %shift : tensor<1xi8>) -> tensor<13x21x3x!quant.uniform<i8:f32, 0.015655439347028732:-1>> {
+ %input_zp = "tosa.const"() {values = dense<127> : tensor<1xi8>} : () -> tensor<1xi8>
+ %output_zp = "tosa.const"() {values = dense<-1> : tensor<1xi8>} : () -> tensor<1xi8>
+ // CHECK: tosa.rescale profiles: [ [pro_int] ]
+ // CHECK: tosa.rescale extensions: [ [int16] ]
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {rounding_mode = "SINGLE_ROUND", scale32 = true, per_channel = false, input_unsigned = true, output_unsigned = false} : (tensor<13x21x3x!quant.uniform<u8:f32, 0.015655439347028732:127>>, tensor<1xi32>, tensor<1xi8>, tensor<1xi8>, tensor<1xi8>) -> tensor<13x21x3x!quant.uniform<i8:f32, 0.015655439347028732:-1>>
+ return %0 : tensor<13x21x3x!quant.uniform<i8:f32, 0.015655439347028732:-1>>
+}
+
+// -----
+// CHECK-LABEL: test_rescale
+func.func @test_rescale_to_unsigned_i8(%arg0: tensor<13x21x3x!quant.uniform<i8:f32, 0.015655439347028732:-1>>, %multiplier : tensor<1xi32>, %shift : tensor<1xi8>) -> tensor<13x21x3x!quant.uniform<u8:f32, 0.015655439347028732:127>> {
+ %input_zp = "tosa.const"() {values = dense<-1> : tensor<1xi8>} : () -> tensor<1xi8>
+ %output_zp = "tosa.const"() {values = dense<127> : tensor<1xi8>} : () -> tensor<1xi8>
+ // CHECK: tosa.rescale profiles: [ [pro_int] ]
+ // CHECK: tosa.rescale extensions: [ [int16] ]
+ %0 = tosa.rescale %arg0, %multiplier, %shift, %input_zp, %output_zp {rounding_mode = "SINGLE_ROUND", scale32 = true, per_channel = false, input_unsigned = false, output_unsigned = true} : (tensor<13x21x3x!quant.uniform<i8:f32, 0.015655439347028732:-1>>, tensor<1xi32>, tensor<1xi8>, tensor<1xi8>, tensor<1xi8>) -> tensor<13x21x3x!quant.uniform<u8:f32, 0.015655439347028732:127>>
+ return %0 : tensor<13x21x3x!quant.uniform<u8:f32, 0.015655439347028732:127>>
+}
+
// -----
// CHECK-LABEL: test_const
func.func @test_const(%arg0 : index) -> tensor<4xi32> {
|
|
cc @psunn |
|
Hi, thanks for looking into this issue. The issue you mentioned above concerns whether the TOSA validation pass should allow ui8/16 inputs and outputs — we can continue the discussion from that perspective. However, the changes in this PR modify the zero-point verification logic in the RescaleOp verifier, which seems unrelated to the issue above. |
| @@ -622,6 +622,28 @@ func.func @test_rescale(%arg0: tensor<13x21x3x!quant.uniform<u8:f32, 0.015655439 | |||
| return %0 : tensor<13x21x3x!quant.uniform<i8:f32, 0.015655439347028732:-1>> | |||
There was a problem hiding this comment.
A little unrelated to this PR - I feel this test case should be considered invalid since input_unsigned=false, but the provided input is u8. This is something we should check in the verifier.
|
Thank you for your passionate review. I think I missed an important part of the issue because I misunderstood it. Thank you again for your good comment. Coming back to the topic, what should I do about the PR I patched? |
|
Thanks @swote-git, I think we can abandon this in favour of #138253 |
This patch fixes a bug in the TOSA RescaleOp verifier that incorrectly rejects unsigned integer types (ui8, ui16), even though they are supported by the TOSA specification.
The verifier now properly handles unsigned integer types when the corresponding input_unsigned or output_unsigned attribute is set to true.
Added tests for ui8<->i8 rescale operations.
Fixes #135699