[Intel] Counter test with arrays has unexpected results on Intel

[hekota]

This issue is for tracking of inc_counter_array.test unexpected results on Intel.

HLSL:

RWStructuredBuffer<int> Out[4] : register(u0);

[numthreads(4,1,1)]
void main(uint GI : SV_GroupIndex) {
  for (int i = 0; i < GI; i++)
    Out[GI].IncrementCounter();

  Out[GI][0] = Out[GI].IncrementCounter();
}

Expected result:

Counters: [ 1, 2, 3, 4 ]
Data:
- [ 0x0 ]
- [ 0x1 ]
- [ 0x2 ]
- [ 0x3 ]

Actual result:

Counters: [ 4, 6, 6, 4 ] 
Data: 
- [ 0x0 ] 
- [ 0x3 ] 
- [ 0x4 ] 
- [ 0x3 ] 

[s-perron]

I have looked into this for SPIR-V while adding counters. I think the problem is that GI is non uniform, but it does not have NonUniformResourceIndex. I'm thinking of rewriting the test to avoid the use of non-uniform indexes.

[s-perron]

In #461, I had tried to change the test to

// This test verifies handling of resource arrays when the
// resource type has a counter.

RWStructuredBuffer<int> Out[4] : register(u0);

[numthreads(4,1,1)]
void main(uint GI : SV_GroupIndex) {
  for (int i = 0; i < GI; i++)
    Out[i].IncrementCounter();
  switch(GI) {
    case 0:
      Out[0][0] = Out[0].IncrementCounter();
      break;
    case 1:
      Out[1][0] = Out[1].IncrementCounter();
      break;
    case 2:
      Out[2][0] = Out[2].IncrementCounter();
      break;
    case 3:
      Out[3][0] = Out[3].IncrementCounter();
      break;
  }
}

In order to avoid the non-uniform dynamic index into the resource array. I believe this new test was causing the clang-d3d12 tests on QC to hang. So I kept the original test. We should fix this in a separate PR.