From 08fa2e7b01dda8959a930295de9829516f8c77bc Mon Sep 17 00:00:00 2001
From: Tim Dettmers <tim.dettmers@gmail.com>
Date: Mon, 7 Nov 2022 18:06:18 -0800
Subject: Fixed bug in cpu quant; faster GPU dequant.

---
 csrc/kernels.cuh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'csrc/kernels.cuh')

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, 
-- 
cgit v1.2.3