From 41a9ed02f118e20de6a44b9351987ab6fa57f994 Mon Sep 17 00:00:00 2001 From: Julia Longtin Date: Sat, 11 May 2024 11:27:52 +0000 Subject: [PATCH] look at the right final memory location. --- ggml-phi-knc.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/ggml-phi-knc.c b/ggml-phi-knc.c index e9b5352b3..add8be2da 100644 --- a/ggml-phi-knc.c +++ b/ggml-phi-knc.c @@ -100,8 +100,8 @@ inline static void GGML_F32x16_VEC_FMA(const float32x16_t *mvec1, const float32x "vmovaps\t\t64(%%r12),\t%%zmm4\n\t" "vfmadd231ps\t%%zmm3,\t%%zmm4,\t%%zmm0\n\t" // Perform a fused multiply add // No compare. we must be three. - "vmovaps\t\t64(%%r10),\t%%zmm5\n\t" // Load two vectors. - "vmovaps\t\t64(%%r12),\t%%zmm6\n\t" + "vmovaps\t\t128(%%r10),\t%%zmm5\n\t" // Load two vectors. + "vmovaps\t\t128(%%r12),\t%%zmm6\n\t" "vfmadd231ps\t%%zmm5,\t%%zmm6,\t%%zmm0\n\t" // Perform a fused multiply add "2:\n\t" // Label for loop end "vmovnraps\t\t%%zmm0,\t(%[RES])\n\t" // Save our results.