Fused FP16 `miopenGcnAsmConv1x1U` verification issue

[Slimakanzer]

• rocm 4.0.0
• gfx906

export MIOPEN_FIND_MODE=1
export MIOPEN_LOG_LEVEL=7
export MIOPEN_DEBUG_CONV_DIRECT=1
export MIOPEN_DEBUG_CONV_WINOGRAD=0
export MIOPEN_DEBUG_AMD_FUSED_WINOGRAD=0
export MIOPEN_DEBUG_CONV_FFT=0
export MIOPEN_DEBUG_CONV_GEMM=0
export MIOPEN_DEBUG_CONV_SCGEMM=0
export MIOPEN_DEBUG_CONV_IMPLICIT_GEMM=0

./MIOpenDriver CBAInferfp16 -W 4 -H 4 -c 512 -n 32 -k 128 -x 1 -y 1 -F 4 -A 1.5 -V 1

...
MIOpen(OpenCL): Info2 [Log] Kernel conv1x1u_bias_activ.s Compile Time, ms: 278.079
MIOpen(OpenCL): Info2 [run] kernel_name = miopenGcnAsmConv1x1U, work_dim = 3, global_work_offset = { 0, 0, 0 }, global_work_dim = { 128, 32, 1 }, local_work_dim = { 64, 1, 1 }
MIOpen(OpenCL): Info [RunForwardCPU] Fusion mode: 4
Running CPU fwd convolution.
Running CPU fwd activation.
Difference in neuron layer: 10.0625 too large at 0 c_v = 113.375 vs g_val = 123.438 tolerance = 0.078125
Forward Activation Failed

miopen-log

[ppanchad-amd]

@Slimakanzer Apologies for the lack of response. Can you please test with latest ROCm 6.0.2 (HIP 6.0.32831)? If resolved, please close ticket. Thanks!

[ppanchad-amd]

@Slimakanzer Closing issue. Please re-open if you still see the issue with latest ROCm 6.1.1 Thanks!