vulkan: Introduce driver version check for Windows Intel GPU to mitigate crashing

[rillomas]

Overview

I replaced the crash guard for Intel Windows driver (introduced at #23964 and #23687) to a version checker so we can guarantee fwht usage on drivers equal to or newer than 32.0.101.8860 on Xe2/3 systems. This is to avoid crashing on the user side who may keep using old drivers.

I checked test-backed-ops -o MUL_MAT_HADAMARD not crashing on following systems using df5b988:

Processor / GPU	Driver Version	Architecture
UX7-368H	32.0.101.8860	Xe3
UX7-368H	Mesa 26.1.4	Xe3
U7-268V	32.0.101.8860	Xe2
U7-268V	32.0.101.8826	Xe2
U7-268V	Mesa 25.2.8	Xe2
i7-1270P	32.0.101.7085	Xe
Arc Pro B70	32.0.101.8804	Xe2
Arc B570	Mesa 26.1.4	Xe2
Arc A770	32.0.101.8860	Xe
Arc A770	Mesa 26.1.4	Xe

Additional information

Seeing solid improvements (max 5.4x) with test-backend-ops perf -o MUL_MAT_HADAMARD on PantherLake UX7-368H.

Before

λ build_vk\bin\Release\test-backend-ops.exe perf -o MUL_MAT_HADAMARD
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = Intel(R) Arc(TM) B390 GPU (Intel Corporation) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: KHR_coopmat
Testing 2 devices

Backend 1/2: Vulkan0
  Device description: Intel(R) Arc(TM) B390 GPU
  Device memory: 37099 MB (36330 MB free)

mul_mat_vec_f32_f32_f32
  MUL_MAT_HADAMARD(type_a=f32,type_b=f32,m=128,n=1,k=128,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1):                  131040 runs -     7.77 us/run -  32.77 kFLOP/run -   4.22 GFLOPS
  MUL_MAT_HADAMARD(type_a=f32,type_b=f32,m=64,n=1,k=64,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1):            294840 runs -     3.44 us/run -   8.19 kFLOP/run -   2.38 GFLOPS
  MUL_MAT_HADAMARD(type_a=f32,type_b=f32,m=256,n=1,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1):                  294840 runs -     3.40 us/run - 131.07 kFLOP/run -  38.60 GFLOPS
matmul_f32_f32_aligned_s
  MUL_MAT_HADAMARD(type_a=f32,type_b=f32,m=128,n=32,k=128,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1):                  90090 runs -    11.61 us/run -   1.05 MFLOP/run -  90.34 GFLOPS
  Backend Vulkan0: OK
Backend 2/2: CPU
  Skipping CPU backend
2/2 backends passed
OK

After

λ build_vk\bin\Release\test-backend-ops.exe perf -o MUL_MAT_HADAMARD
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = Intel(R) Arc(TM) B390 GPU (Intel Corporation) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: KHR_coopmat
Testing 2 devices

Backend 1/2: Vulkan0
  Device description: Intel(R) Arc(TM) B390 GPU
  Device memory: 37099 MB (36330 MB free)

fwht_f32
  MUL_MAT_HADAMARD(type_a=f32,type_b=f32,m=128,n=1,k=128,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1):                  204750 runs -     4.93 us/run -  32.77 kFLOP/run -   6.65 GFLOPS
  MUL_MAT_HADAMARD(type_a=f32,type_b=f32,m=64,n=1,k=64,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1):            565110 runs -     1.79 us/run -   8.19 kFLOP/run -   4.58 GFLOPS
  MUL_MAT_HADAMARD(type_a=f32,type_b=f32,m=256,n=1,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1):                  417690 runs -     2.39 us/run - 131.07 kFLOP/run -  54.74 GFLOPS
  MUL_MAT_HADAMARD(type_a=f32,type_b=f32,m=128,n=32,k=128,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1):                 475020 runs -     2.12 us/run -   1.05 MFLOP/run - 494.90 GFLOPS
  Backend Vulkan0: OK
Backend 2/2: CPU
  Skipping CPU backend
2/2 backends passed
OK

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: Yes, used copilot for most of the coding

[rillomas]

Hi @jeffbolznv @0cc4m can you two review this PR?