From 684da25926e5c505f725b4f10b5485b218fa1fc7 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Mon, 10 Apr 2023 19:29:48 +0300 Subject: [PATCH] ggml : fix quantize_row_q4_1() ARM_NEON (close #876) --- ggml.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-) diff --git a/ggml.c b/ggml.c index 326b8e8..9616eb9 100644 --- a/ggml.c +++ b/ggml.c @@ -599,10 +599,7 @@ static void quantize_row_q4_0(const float * restrict x, void * restrict vy, int for (int l = 0; l < 2; l++) amaxv[4*l] = vmaxq_f32(amaxv[4*l], amaxv[4*l+2]); for (int l = 0; l < 1; l++) amaxv[8*l] = vmaxq_f32(amaxv[8*l], amaxv[8*l+4]); - // absolute max - const float amax = MAX( - MAX(vgetq_lane_f32(amaxv[0], 0), vgetq_lane_f32(amaxv[0], 1)), - MAX(vgetq_lane_f32(amaxv[0], 2), vgetq_lane_f32(amaxv[0], 3))); + const float amax = vmaxvq_f32(amaxv[0]); const float d = amax / ((1 << 3) - 1); const float id = d ? 1.0f/d : 0.0f; @@ -924,7 +921,7 @@ static void quantize_row_q4_1(const float * restrict x, void * restrict vy, int float32x4_t minv[8]; float32x4_t maxv[8]; - for (int l = 0; l < 8; l++) srcv[l] = vld1q_f32(x + i*32 + 4*l); + for (int l = 0; l < 8; l++) srcv[l] = vld1q_f32(x + i*QK + 4*l); for (int l = 0; l < 4; l++) minv[2*l] = vminq_f32(srcv[2*l], srcv[2*l + 1]); for (int l = 0; l < 2; l++) minv[4*l] = vminq_f32(minv[4*l], minv[4*l + 2]); @@ -947,7 +944,8 @@ static void quantize_row_q4_1(const float * restrict x, void * restrict vy, int for (int l = 0; l < 8; l++) { const float32x4_t v = vmulq_n_f32(vsubq_f32(srcv[l], minv0), id); - const int32x4_t vi = vcvtq_s32_f32(v); + const float32x4_t vf = vaddq_f32(v, vdupq_n_f32(0.5f)); // needed to round to nearest + const int32x4_t vi = vcvtq_s32_f32(vf); y[i].qs[2*l + 0] = vgetq_lane_s32(vi, 0) | (vgetq_lane_s32(vi, 1) << 4); y[i].qs[2*l + 1] = vgetq_lane_s32(vi, 2) | (vgetq_lane_s32(vi, 3) << 4);