Nvidia RTX 40/50 Vulkan runtime error on benchncnn_llm

[DickyQi]

OS: Ubuntu 26.04

codewin@codewin-XRStudio:/Gitcodes/ncnn/build/benchmark$ ./benchncnn_llm 64 1 0 0 0
[0 NVIDIA GeForce RTX 4060 Ti] queueC=2[8] queueT=1[2] rebar=0 r-score=74
[0 NVIDIA GeForce RTX 4060 Ti] fp16-p/s/u/a=1/1/1/1 int8-p/s/u/a=1/1/1/1 bf16-p/s=1/1
[0 NVIDIA GeForce RTX 4060 Ti] subgroup=32(3232) ops=1/1/1/1/1/1/1/1/1/1
[0 NVIDIA GeForce RTX 4060 Ti] fp16-cm=16x16x16/16x8x16/16x8x8 int8-cm=16x16x32/16x8x32 bf16-cm=16x16x16 fp8-cm=0
[1 llvmpipe (LLVM 21.1.8, 256 bits)] queueC=0[1] queueT=0[1] rebar=1 r-score=14
[1 llvmpipe (LLVM 21.1.8, 256 bits)] fp16-p/s/u/a=1/1/1/1 int8-p/s/u/a=1/1/1/1 bf16-p/s=1/0
[1 llvmpipe (LLVM 21.1.8, 256 bits)] subgroup=8(8~8) ops=1/1/1/1/1/1/1/1/1/1
[1 llvmpipe (LLVM 21.1.8, 256 bits)] fp16-cm=8x8x8 int8-cm=8x8x8 bf16-cm=0 fp8-cm=0
loop_count = 64
num_threads = 1
powersave = 0
gpu_device = 0
cooling_down = 0
compile spir-v module failed
ERROR: 3:163: 'assign' : cannot convert from ' temp coopmat<3, 1, 1, 2> bfloat16_t' to ' temp coopmat<3, 1, 1, 2> float'
ERROR: 3:163: '' : missing #endif
ERROR: 3:163: '' : compilation terminated
ERROR: 3 compilation errors. No code generated.

[DickyQi]

compilation errors on file gemm_cm.comp

[nihui]

update your glslang or use the submodule one