HIP: add gfx1152 and gfx1153 to RDNA3.5

[harkgill-amd]

Overview

Add gfx1152 and gfx1153 definitions to RDNA3.5 macro in ggml/src/ggml-cuda/vendors/hip.h.

Additional information

Resolves ROCm/TheRock#5579 where users report corrupted output with TheRock nightlies + llama.cpp build from source. Patching this change in resolves the issue.

Requirements

• I have read and agree with the contributing guidelines : Done
• AI usage disclosure: To understand where changes were needed to be made.

[Atomic-Germ]

It seems to work; I have a gfx1152 system. It's still a little slower than Vulkan of course, but it isn't a bunch of stuttering nonsense either.

[itn3rd77]

Not related to this PR but why is this labeled Nvidia GPU by the bot? I have seen this obviously wrong labeling several times now.

[soulafein83]

It seems to work; I have a gfx1152 system. It's still a little slower than Vulkan of course, but it isn't a bunch of stuttering nonsense either.

On my laptop HP Omnibook with Ryzen 7 AI 350, I have (on CachyOS) with ROCm 17 TPS, with Vulkan 27 TPS. Model Gemma 4 e4b with 32k

[ckuethe]

Works for me on System76 Pangolin Pro

Gemma4

0.00.153.930 I device_info:
0.00.154.019 I   - ROCm0   : AMD Radeon 860M Graphics (81920 MiB, 75951 MiB free)
0.00.154.023 I   - CPU     : AMD Ryzen AI 7 350 w/ Radeon 860M (94149 MiB, 94149 MiB free)
...
3.05.452.125 I slot print_timing: id  3 | task 0 | prompt eval time =     318.70 ms /    15 tokens (   21.25 ms per token,    47.07 tokens per second)
3.05.452.129 I slot print_timing: id  3 | task 0 |        eval time =   67519.17 ms /  1227 tokens (   55.03 ms per token,    18.17 tokens per second)
3.05.452.130 I slot print_timing: id  3 | task 0 |       total time =   67837.88 ms /  1242 tokens
3.05.452.137 I slot print_timing: id  3 | task 0 |    graphs reused =       1222
3.05.452.168 I slot      release: id  3 | task 0 | stop processing: n_tokens = 1241, truncated = 0

Bonsai-8B

0.00.137.764 I device_info:
0.00.137.843 I   - ROCm0   : AMD Radeon 860M Graphics (81920 MiB, 76868 MiB free)
0.00.137.847 I   - CPU     : AMD Ryzen AI 7 350 w/ Radeon 860M (94149 MiB, 94149 MiB free)
...
1.26.916.161 I slot print_timing: id  3 | task 0 | prompt eval time =     199.98 ms /    19 tokens (   10.53 ms per token,    95.01 tokens per second)
1.26.916.163 I slot print_timing: id  3 | task 0 |        eval time =   47210.86 ms /  1313 tokens (   35.96 ms per token,    27.81 tokens per second)
1.26.916.164 I slot print_timing: id  3 | task 0 |       total time =   47410.84 ms /  1332 tokens
1.26.916.168 I slot print_timing: id  3 | task 0 |    graphs reused =       1307
1.26.916.196 I slot      release: id  3 | task 0 | stop processing: n_tokens = 1331, truncated = 0