fix cm pipeline compilation failure on adreno

[nihui]

large tile sizes on adreno gpu cause excessive shared memory usage when loops are unrolled. disabling unroll specifically for the adreno to prevent compilation errors.

[tencent-adm]

Thank you for your submission, we really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[codecov-commenter]

Codecov Report

❌ Patch coverage is 17.64706% with 42 lines in your changes missing coverage. Please review.
✅ Project coverage is 93.91%. Comparing base (421de78) to head (f83ba09).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/layer/vulkan/convolution_vulkan.cpp	16.66%	15 Missing ⚠️
src/layer/vulkan/convolution1d_vulkan.cpp	16.66%	10 Missing ⚠️
src/layer/vulkan/sdpa_vulkan.cpp	22.22%	7 Missing ⚠️
src/layer/vulkan/deconvolution_vulkan.cpp	16.66%	5 Missing ⚠️
src/layer/vulkan/gemm_vulkan.cpp	16.66%	5 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##           master    #6720    +/-   ##
========================================
  Coverage   93.90%   93.91%            
========================================
  Files         933      933            
  Lines      310041   309919   -122     
========================================
- Hits       291140   291047    -93     
+ Misses      18901    18872    -29     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[nihui]

HNBKQ:/data/local/tmp $ ./benchncnn 4 1 2 0 0                                                                                                                                                                                                                                                                              
[0] apiVersion = 1.4.295
[0] driverVersion = 512.842.25
[0] vendorID = 5143
[0] deviceID = 44050a31
[0] deviceType = 1
[0] deviceName = Adreno (TM) 840
[0] pipelineCacheUUID = 849392164
[0 Adreno (TM) 840]  queueC=1[1]  queueT=0[4]  rebar=1  r-score=21
[0 Adreno (TM) 840]  fp16-p/s/u/a=1/1/0/1  int8-p/s/u/a=1/1/0/1  bf16-p/s=1/0
[0 Adreno (TM) 840]  subgroup=64(64~64)  ops=1/1/1/1/1/1/1/1/1/1
[0 Adreno (TM) 840]  fp16-cm=0  int8-cm=0  bf16-cm=0  fp8-cm=0
loop_count = 4
num_threads = 1
powersave = 2
gpu_device = 0
cooling_down = 0
          squeezenet  min =    3.02  max =    3.22  avg =    3.11
           mobilenet  min =    4.69  max =    4.76  avg =    4.73
        mobilenet_v2  min =    4.18  max =    4.25  avg =    4.21
        mobilenet_v3  min =    4.30  max =    4.38  avg =    4.33
          shufflenet  min =    3.58  max =    3.91  avg =    3.78
       shufflenet_v2  min =    3.39  max =    3.41  avg =    3.40
             mnasnet  min =    4.15  max =    4.23  avg =    4.19
     proxylessnasnet  min =    4.39  max =    4.48  avg =    4.43
     efficientnet_b0  min =    6.41  max =    6.50  avg =    6.45
   efficientnetv2_b0  min =    9.02  max =    9.22  avg =    9.07
        regnety_400m  min =    5.11  max =    5.29  avg =    5.19
           blazeface  min =    1.70  max =    1.71  avg =    1.70
           googlenet  min =    8.36  max =    8.41  avg =    8.38
            resnet18  min =    5.11  max =    5.18  avg =    5.14
             alexnet  min =    5.83  max =    5.92  avg =    5.88
               vgg16  min =   23.70  max =   24.06  avg =   23.93
            resnet50  min =   14.60  max =   16.90  avg =   15.85
      squeezenet_ssd  min =    7.30  max =    7.42  avg =    7.36
       mobilenet_ssd  min =    9.28  max =    9.43  avg =    9.37
      mobilenet_yolo  min =   15.72  max =   19.20  avg =   17.64
  mobilenetv2_yolov3  min =   13.95  max =   14.29  avg =   14.09
         yolov4-tiny  min =   17.87  max =   19.68  avg =   19.18
           nanodet_m  min =    8.57  max =    8.86  avg =    8.67
    yolo-fastest-1.1  min =    3.44  max =    3.47  avg =    3.46
      yolo-fastestv2  min =    3.23  max =    3.37  avg =    3.29
  vision_transformer  min =   35.77  max =   37.59  avg =   36.26
          FastestDet  min =    3.14  max =    3.30  avg =    3.19
HNBKQ:/data/local/tmp $ ./benchncnn 4 1 2 0 0                                                                                                                                                                                                                                                                              
[0] apiVersion = 1.4.295
[0] driverVersion = 512.842.25
[0] vendorID = 5143
[0] deviceID = 44050a31
[0] deviceType = 1
[0] deviceName = Adreno (TM) 840
[0] pipelineCacheUUID = 1130390036
[0 Adreno (TM) 840]  queueC=1[1]  queueT=0[4]  rebar=1  r-score=21
[0 Adreno (TM) 840]  fp16-p/s/u/a=1/1/0/1  int8-p/s/u/a=1/1/0/1  bf16-p/s=1/0
[0 Adreno (TM) 840]  subgroup=64(64~64)  ops=1/1/1/1/1/1/1/1/1/1
[0 Adreno (TM) 840]  fp16-cm=64x64x16/64x32x16/64x16x16  int8-cm=64x64x32/64x32x32/64x16x32  bf16-cm=0  fp8-cm=0
loop_count = 4
num_threads = 1
powersave = 2
gpu_device = 0
cooling_down = 0

^C
130|HNBKQ:/data/local/tmp $ ./benchncnn 4 1 2 0 0                                                                                                                                                                                                                                                                          
[0] apiVersion = 1.4.295
[0] driverVersion = 512.842.25
[0] vendorID = 5143
[0] deviceID = 44050a31
[0] deviceType = 1
[0] deviceName = Adreno (TM) 840
[0] pipelineCacheUUID = 862003764
[0 Adreno (TM) 840]  queueC=1[1]  queueT=0[4]  rebar=1  r-score=21
[0 Adreno (TM) 840]  fp16-p/s/u/a=1/1/0/1  int8-p/s/u/a=1/1/0/1  bf16-p/s=1/0
[0 Adreno (TM) 840]  subgroup=64(64~64)  ops=1/1/1/1/1/1/1/1/1/1
[0 Adreno (TM) 840]  fp16-cm=64x64x16/64x32x16/64x16x16  int8-cm=64x64x32/64x32x32/64x16x32  bf16-cm=0  fp8-cm=0
loop_count = 4
num_threads = 1
powersave = 2
gpu_device = 0
cooling_down = 0
          squeezenet  min =    1.74  max =    1.87  avg =    1.78
           mobilenet  min =    2.41  max =    2.62  avg =    2.53
        mobilenet_v2  min =    3.50  max =    3.68  avg =    3.62
        mobilenet_v3  min =    3.17  max =    3.38  avg =    3.28
          shufflenet  min =    2.16  max =    2.22  avg =    2.19
       shufflenet_v2  min =    2.56  max =    2.74  avg =    2.62
             mnasnet  min =    3.16  max =    3.91  avg =    3.47
     proxylessnasnet  min =    3.12  max =    3.67  avg =    3.50
     efficientnet_b0  min =    5.61  max =    5.82  avg =    5.68
   efficientnetv2_b0  min =    7.68  max =    8.88  avg =    8.05
        regnety_400m  min =    4.43  max =    4.66  avg =    4.50
           blazeface  min =    1.19  max =    1.22  avg =    1.21
           googlenet  min =    6.47  max =    6.65  avg =    6.58
            resnet18  min =    3.93  max =    3.98  avg =    3.96
             alexnet  min =    5.59  max =    5.79  avg =    5.69
               vgg16  min =   15.00  max =   18.21  avg =   16.73
            resnet50  min =    8.79  max =    9.30  avg =    8.96
      squeezenet_ssd  min =    6.05  max =    6.12  avg =    6.08
       mobilenet_ssd  min =    5.59  max =    5.70  avg =    5.62
      mobilenet_yolo  min =    7.05  max =    7.16  avg =    7.11
  mobilenetv2_yolov3  min =   11.86  max =   12.29  avg =   12.07
         yolov4-tiny  min =    5.39  max =    5.51  avg =    5.44
           nanodet_m  min =    7.70  max =    7.79  avg =    7.76
    yolo-fastest-1.1  min =    3.10  max =    3.30  avg =    3.17
      yolo-fastestv2  min =    2.52  max =    2.79  avg =    2.68
  vision_transformer  min =   29.49  max =   30.22  avg =   29.84
          FastestDet  min =    2.19  max =    2.41  avg =    2.24

[nihui]

coopMatLoad/coopMatStore layout that depends on push constant breaks adreno driver

vkCreateComputePipelines failed -13

if (p.elempack == 4)
{
    coopMatStore(sum, tmp, 0, stride, gl_CooperativeMatrixLayoutColumnMajor);
}
else
{
    coopMatStore(sum, tmp, 0, stride, gl_CooperativeMatrixLayoutRowMajor);
}