vulkan: coopmat2 mul_mat optimizations

[jeffbolznv]

• Increase tile size for k-quants, to match non-k-quants
• Choose more carefully between large and medium tiles, considering how it interacts with split_k
• Allow larger/non-power of two split_k, and make the splits a multiple of 256
• Use split_k==3 to when >1/2 and <=2/3 of the SMs would hae been used

Perf results on 4070 and 5090. The split_k and medium/large selection improvements particularly benefit 5090 where the GPU often wasn't being filled.

before:

Z:\github\jeffbolznv\llama.cpp\build\bin\RelWithDebInfo>llama-bench.exe -fa 1 -n 0 -p 512 -r 50 --prio 1 -m C:\models\DeepSeek-R1-Distill-Qwen-14B-Q4_K_M.gguf -m C:\models\gemma-3n-E4B-it-Q4_K_M.gguf -m C:\models\llama-3.2-3b-instruct-q8_0.gguf -m C:\models\meta-llama-3-8b-instruct.Q4_K_M.gguf -m C:\models\Phi-3-mini-4k-instruct-q4.gguf
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce RTX 4070 (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| qwen2 14B Q4_K - Medium        |   8.37 GiB |    14.77 B | Vulkan     |  99 |  1 |           pp512 |       1920.60 ± 1.79 |
| gemma3n E4B Q4_K - Medium      |   4.22 GiB |     6.87 B | Vulkan     |  99 |  1 |           pp512 |      2609.35 ± 41.34 |
| llama 3B Q8_0                  |   3.18 GiB |     3.21 B | Vulkan     |  99 |  1 |           pp512 |     8254.77 ± 121.91 |
| llama 8B Q4_K - Medium         |   4.58 GiB |     8.03 B | Vulkan     |  99 |  1 |           pp512 |       3765.21 ± 8.59 |
| phi3 3B Q4_K - Medium          |   2.23 GiB |     3.82 B | Vulkan     |  99 |  1 |           pp512 |      6759.22 ± 30.63 |

after:

ggml_vulkan: 0 = NVIDIA GeForce RTX 4070 (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| qwen2 14B Q4_K - Medium        |   8.37 GiB |    14.77 B | Vulkan     |  99 |  1 |           pp512 |       2239.43 ± 3.60 |
| gemma3n E4B Q4_K - Medium      |   4.22 GiB |     6.87 B | Vulkan     |  99 |  1 |           pp512 |      2694.36 ± 46.88 |
| llama 3B Q8_0                  |   3.18 GiB |     3.21 B | Vulkan     |  99 |  1 |           pp512 |     8178.23 ± 134.56 |
| llama 8B Q4_K - Medium         |   4.58 GiB |     8.03 B | Vulkan     |  99 |  1 |           pp512 |       3990.27 ± 5.69 |
| phi3 3B Q4_K - Medium          |   2.23 GiB |     3.82 B | Vulkan     |  99 |  1 |           pp512 |      6636.85 ± 13.82 |

before:

ggml_vulkan: 0 = NVIDIA GeForce RTX 5090 (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| qwen2 14B Q4_K - Medium        |   8.37 GiB |    14.77 B | Vulkan     |  99 |  1 |           pp512 |      5469.64 ± 45.64 |
| gemma3n E4B Q4_K - Medium      |   4.22 GiB |     6.87 B | Vulkan     |  99 |  1 |           pp512 |      4763.36 ± 70.06 |
| llama 3B Q8_0                  |   3.18 GiB |     3.21 B | Vulkan     |  99 |  1 |           pp512 |    24341.22 ± 246.44 |
| llama 8B Q4_K - Medium         |   4.58 GiB |     8.03 B | Vulkan     |  99 |  1 |           pp512 |     9577.41 ± 143.94 |
| phi3 3B Q4_K - Medium          |   2.23 GiB |     3.82 B | Vulkan     |  99 |  1 |           pp512 |    15541.69 ± 134.82 |

after:

ggml_vulkan: 0 = NVIDIA GeForce RTX 5090 (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| qwen2 14B Q4_K - Medium        |   8.37 GiB |    14.77 B | Vulkan     |  99 |  1 |           pp512 |      6351.47 ± 30.99 |
| gemma3n E4B Q4_K - Medium      |   4.22 GiB |     6.87 B | Vulkan     |  99 |  1 |           pp512 |     5170.98 ± 170.61 |
| llama 3B Q8_0                  |   3.18 GiB |     3.21 B | Vulkan     |  99 |  1 |           pp512 |   26591.76 ± 1879.27 |
| llama 8B Q4_K - Medium         |   4.58 GiB |     8.03 B | Vulkan     |  99 |  1 |           pp512 |    11046.12 ± 215.04 |
| phi3 3B Q4_K - Medium          |   2.23 GiB |     3.82 B | Vulkan     |  99 |  1 |           pp512 |    20919.16 ± 788.06 |

[0cc4m]

LGTM