| 543 | } |
| 544 | |
| 545 | void cdequant_mm_int32_fp16( |
| 546 | int* A, float* rowStats, float* colStats, half* out, half* bias, int numRows, int numCols, cudaStream_t stream |
| 547 | ) { |
| 548 | dequant_mm_int32_fp16(A, rowStats, colStats, out, bias, numRows, numCols, stream); |
| 549 | } |
| 550 | |
| 551 | void cint8_vector_quant( |
| 552 | half* __restrict__ A, int8_t* out, float* rowStats, float threshold, int rows, int cols, cudaStream_t stream |
nothing calls this directly
no outgoing calls
no test coverage detected