Change instructions as in llama.cpp

RWKV · Apr 9, 2023 · 69e16e5 · 69e16e5
1 parent 4f31544
commit 69e16e5
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/ggml.c b/ggml.c
@@ -1083,8 +1083,8 @@ static void dequantize_row_q4_1(const void * restrict vx, float * restrict y, in
             const uint8x16_t vq = vcombine_u8(vx_0, vx_1);
 
             // convert to 2x uint16x8_t
-            const uint16x8_t vi_0 = vmovl_s8(vget_low_u8 (vq));
-            const uint16x8_t vi_1 = vmovl_s8(vget_high_u8(vq));
+            const uint16x8_t vi_0 = vmovl_u8(vget_low_u8 (vq));
+            const uint16x8_t vi_1 = vmovl_u8(vget_high_u8(vq));
 
             // convert to 4x float32x4_t
             const float32x4_t vf_0 = vcvtq_f32_u32(vmovl_u16(vget_low_u16 (vi_0)));
@@ -1301,8 +1301,8 @@ static void dequantize_row_q4_1_o(const void * restrict vx, float * restrict y,
             const uint8x16_t vq = vcombine_u8(vx_0, vx_1);
 
             // convert to 2x uint16x8_t
-            const uint16x8_t vi_0 = vmovl_s8(vget_low_u8 (vq));
-            const uint16x8_t vi_1 = vmovl_s8(vget_high_u8(vq));
+            const uint16x8_t vi_0 = vmovl_u8(vget_low_u8 (vq));
+            const uint16x8_t vi_1 = vmovl_u8(vget_high_u8(vq));
 
             // convert to 4x float32x4_t
             const float32x4_t vf_0 = vcvtq_f32_u32(vmovl_u16(vget_low_u16 (vi_0)));