diff options
| author | Tim Dettmers <tim.dettmers@gmail.com> | 2022-11-07 18:06:18 -0800 |
|---|---|---|
| committer | Tim Dettmers <tim.dettmers@gmail.com> | 2022-11-07 18:06:18 -0800 |
| commit | 08fa2e7b01dda8959a930295de9829516f8c77bc (patch) | |
| tree | 0c31b3fa012caac459bea4ceda1890c153d81110 /csrc/kernels.cuh | |
| parent | 62a333ac40f157e69c4bb86f30ac06b41ca4ff34 (diff) | |
Fixed bug in cpu quant; faster GPU dequant.
Diffstat (limited to 'csrc/kernels.cuh')
| -rw-r--r-- | csrc/kernels.cuh | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/csrc/kernels.cuh b/csrc/kernels.cuh index bdf61b2..cca983b 100644 --- a/csrc/kernels.cuh +++ b/csrc/kernels.cuh @@ -15,7 +15,7 @@ __global__ void kQuantize(float * code, float * __restrict__ const A, unsigned c __global__ void kDequantize(float *code, unsigned char *A, float *out, const int n); template<typename T, int BLOCK_SIZE, int NUM_PER_TH, int STOCHASTIC> __global__ void kQuantizeBlockwise(float * code, T * __restrict__ const A, float *absmax, unsigned char *out, float * __restrict__ const rand, const int rand_offset, const int n); -template<typename T, int BLOCK_SIZE, int THREADS, int NUM_PER_TH> __global__ void kDequantizeBlockwise(float *code, unsigned char * __restrict__ const A, float * __restrict__ const absmax, T *out, const int n); +template<typename T, int BLOCK_SIZE, int THREADS, int NUM_PER_TH> __global__ void kDequantizeBlockwise(float *code, unsigned char * A, float * absmax, T *out, const int n); template<typename T, int OPTIMIZER, int BLOCK_SIZE, int NUM_VALS> __global__ void kPreconditionOptimizer32bit2State(T* g, T* p, |