d47cccc157
+ Microarchitecture optimizations + 64-bit support + Xilinx FPGA support + LLVM-16 support + Refactoring and quality control fixes
21 lines
471 B
Common Lisp
21 lines
471 B
Common Lisp
#include "common.h"
|
|
|
|
__kernel void sgemm (__global const TYPE *A,
|
|
__global const TYPE *B,
|
|
__global TYPE *C,
|
|
int N)
|
|
{
|
|
// Thread identifiers
|
|
const int r = get_global_id(0); // Row ID
|
|
const int c = get_global_id(1); // Col ID
|
|
|
|
// Compute a single element (loop a K)
|
|
TYPE acc = 0;
|
|
for (int k = 0; k < N; k++) {
|
|
acc += A[k * N + r] * B[c * N + k];
|
|
}
|
|
|
|
// Store the result
|
|
C[c * N + r] = acc;
|
|
}
|