Misc. bug: test-backend-ops grad crash by GGML_ASSERT error

[masamaru-san]

Name and Version

.\llama-cli.exe --version
version: 4942 (fbdfefe)
built with MSVC 19.43.34808.0 for x64

Operating systems

Windows

Which llama.cpp modules do you know to be affected?

Test code

Command line

> .\test-backend-ops.exe grad -o CPY
or
> .\test-backend-ops.exe grad

Problem description & steps to reproduce

description

Commit #12310 crashes test-backend-ops grad.
It doesn't seem to matter which backend.

steps to reproduce

Run test-backend-ops as grad mode.

First Bad Commit

Commit #12310 : SHA ba932df

Relevant log output

[3/23 08:24:26] PS E:\AI\llama.cpp\b4942\llama-b4942-bin-win-vulkan-x64
> .\test-backend-ops.exe grad -o CPY
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = AMD Radeon(TM) Graphics (AMD proprietary driver) | uma: 1 | fp16: 1 | warp size: 64 | shared memory: 32768 | matrix cores: none
Testing 2 devices

Backend 1/2: Vulkan0
  Device description: AMD Radeon(TM) Graphics
  Device memory: 256 MB (256 MB free)

  CPY(type_src=f32,type_dst=f32,ne=[1,2,3,4],permute_src=[0,0,0,0],permute_dst=[0,0,0,0]): OK
  CPY(type_src=f32,type_dst=f32,ne=[1,2,3,4],permute_src=[0,2,1,3],permute_dst=[0,0,0,0]): OK
  CPY(type_src=f32,type_dst=f32,ne=[1,2,3,4],permute_src=[0,3,1,2],permute_dst=[0,2,1,3]): D:\a\llama.cpp\llama.cpp\ggml\src\ggml.c:5816: GGML_ASSERT(!src0_needs_grads || ggml_are_same_shape(src0, cgraph->grads[isrc0])) failed
[3/23 08:24:39] PS E:\AI\llama.cpp\b4942\llama-b4942-bin-win-vulkan-x64
> cd ..\llama-b4942-bin-win-avx2-x64\
[3/23 08:24:55] PS E:\AI\llama.cpp\b4942\llama-b4942-bin-win-avx2-x64
> .\test-backend-ops.exe grad -o CPY
Testing 1 devices

Backend 1/1: CPU
  Device description: AMD Ryzen 7 5700U with Radeon Graphics
  Device memory: 0 MB (0 MB free)

  CPY(type_src=f32,type_dst=f32,ne=[1,2,3,4],permute_src=[0,0,0,0],permute_dst=[0,0,0,0]): OK
  CPY(type_src=f32,type_dst=f32,ne=[1,2,3,4],permute_src=[0,2,1,3],permute_dst=[0,0,0,0]): OK
  CPY(type_src=f32,type_dst=f32,ne=[1,2,3,4],permute_src=[0,3,1,2],permute_dst=[0,2,1,3]): D:\a\llama.cpp\llama.cpp\ggml\src\ggml.c:5816: GGML_ASSERT(!src0_needs_grads || ggml_are_same_shape(src0, cgraph->grads[isrc0])) failed

[masamaru-san]

GGML_ASSERT line here.

llama.cpp/ggml/src/ggml.c

Lines 5814 to 5818 in fbdfefe

	}
	GGML_ASSERT(!src0_needs_grads || ggml_are_same_shape(src0, cgraph->grads[isrc0]));
	GGML_ASSERT(!src1_needs_grads || ggml_are_same_shape(src1, cgraph->grads[isrc1]));
	GGML_ASSERT(!src2_needs_grads || ggml_are_same_shape(src2, cgraph->grads[isrc2]));