Vulkan: Support fp32 accumulator in quantized matmul to fix GLM4-32B incoherence#13607

0cc4m merged 1 commit intomasterggml-org/llama.cpp:masterfrom

0cc4m/vulkan-mmq-fp32-acc-glm4ggml-org/llama.cpp:0cc4m/vulkan-mmq-fp32-acc-glm4Copy head branch name to clipboard