[Mlir-commits] [mlir] Sub-channel quantized type implementation (PR #120172)

[Sandeep Dasgupta]

sdasgup3 wrote:

> I'm seeing a test regression on 32-bit x86 with this change:
> 
> ```
> FAIL: MLIR :: CAPI/quant.c (38 of 2812)
> ******************** TEST 'MLIR :: CAPI/quant.c' FAILED ********************
> Exit Code: 1
> 
> Command Output (stdout):
> --
> # RUN: at line 10
> /var/tmp/portage/llvm-core/mlir-21.0.0.9999/work/mlir_build-abi_x86_32.x86/bin/mlir-capi-quant-test 2>&1 | /usr/lib/llvm/21/bin/FileCheck /var/tmp/portage/llvm-core/mlir-21.0.0.9999/work/mlir/test/CAPI/quant.c
> # executed command: /var/tmp/portage/llvm-core/mlir-21.0.0.9999/work/mlir_build-abi_x86_32.x86/bin/mlir-capi-quant-test
> # executed command: /usr/lib/llvm/21/bin/FileCheck /var/tmp/portage/llvm-core/mlir-21.0.0.9999/work/mlir/test/CAPI/quant.c
> # .---command stderr------------
> # | /var/tmp/portage/llvm-core/mlir-21.0.0.9999/work/mlir/test/CAPI/quant.c:270:12: error: CHECK: expected string not found in input
> # |  // CHECK: num dims: 2
> # |            ^
> # | <stdin>:51:28: note: scanning from here
> # | is null sub-channel type: 1
> # |                            ^
> # | <stdin>:52:1: note: possible intended match here
> # | num dims: 6292612763541831682
> # | ^
> # | 
> # | Input file: <stdin>
> # | Check file: /var/tmp/portage/llvm-core/mlir-21.0.0.9999/work/mlir/test/CAPI/quant.c
> # | 
> # | -dump-input=help explains the following input dump.
> # | 
> # | Input was:
> # | <<<<<<
> # |              .
> # |              .
> # |              .
> # |             46: equal: 1 
> # |             47: !quant.uniform<i8:f32:1, {2.000000e+02,9.987200e-01:120}> 
> # |             48:  
> # |             49:  
> # |             50: testUniformSubChannelType 
> # |             51: is null sub-channel type: 1 
> # | check:270'0                                X error: no match found
> # |             52: num dims: 6292612763541831682 
> # | check:270'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> # | check:270'1     ?                              possible intended match
> # |             53: axis-block-size-pair[0]: 0:1 
> # | check:270'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> # |             54: axis-block-size-pair[1]: 1:2 
> # | check:270'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> # |             55: tensor<2x2xf32> 
> # | check:270'0     ~~~~~~~~~~~~~~~~
> # |             56: tensor<2x2xi8> 
> # | check:270'0     ~~~~~~~~~~~~~~~
> # |             57: number of quantization parameters: 4 
> # | check:270'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> # |              .
> # |              .
> # |              .
> # | >>>>>>
> # `-----------------------------
> # error: command failed with exit status: 1
> 
> --
> 
> ********************
> ```

@mgorny Please find the fix @ https://github.com/llvm/llvm-project/pull/133763 and let me know if this looks good.

https://github.com/llvm/llvm-project/pull/120172