From: Georgi Gerganov <redacted>
Date: Mon, 23 Sep 2024 08:27:47 +0000 (+0300)
Subject: metal : use F32 prec for K*Q in vec FA (llama/9595)
X-Git-Tag: upstream/0.0.1642~347
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=64f30f362597c427889446484b8f6a9176f14601;p=pkg%2Fggml%2Fsources%2Fggml

metal : use F32 prec for K*Q in vec FA (llama/9595)

ggml-ci
---

diff --git a/src/ggml-metal.metal b/src/ggml-metal.metal
index f323ab5f..2b200032 100644
--- a/src/ggml-metal.metal
+++ b/src/ggml-metal.metal
@@ -2631,11 +2631,11 @@ kernel void kernel_flash_attn_ext_vec_f16(
         const short iv3 = iq3 / rv3;
 
         // load the queries from shared memory into local memory
-        half4 mq[D4];
+        float4 mq[D4];
 
         for (short ii = 0; ii < D4; ii += NW) {
             short i = ii + tiisg;
-            mq[i] = sq4[i];
+            mq[i] = (float4) sq4[i];
         }
 
         // pointer to the mask
@@ -2661,11 +2661,11 @@ kernel void kernel_flash_attn_ext_vec_f16(
                     for (short ii = 0; ii < D4; ii += NW) {
                         const short i = ii + tiisg;
 
-                        half4x4 mk;
-                        mk[0] = pk4[i + 0*(nb11/8)];
-                        mk[1] = pk4[i + 1*(nb11/8)];
-                        mk[2] = pk4[i + 2*(nb11/8)];
-                        mk[3] = pk4[i + 3*(nb11/8)];
+                        float4x4 mk;
+                        mk[0] = (float4) pk4[i + 0*(nb11/8)];
+                        mk[1] = (float4) pk4[i + 1*(nb11/8)];
+                        mk[2] = (float4) pk4[i + 2*(nb11/8)];
+                        mk[3] = (float4) pk4[i + 3*(nb11/8)];
 
                         mqk += (float4) (mq[i] * mk);
                     }