Just installed Catalyst 11.11, and all my OpenCL kernels with #pragma unroll when built on a CPU device abort in clBuildProgram -- an Access Violation in AMD's DLL somewhere. APP KernelAnalyzer also crashes generating x86 code. Kernel is below.
__kernel void MatrixMultiplySimple( __global float* C, __global const float* A, __global const float* B, const int width ) { const int x = get_global_id( 0 ); const int y = get_global_id( 1 ); float sum = 0.0f; #pragma unroll 4 for( int k = 0; k < width; ++k ) sum += A[ y * width + k ] * B[ k * width + x ]; C[ y * width + x ] = sum; }