From 15a484dee65e8f3390702f6e43d493de26e05eb5 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Sun, 8 Feb 2026 11:06:54 +0200
Subject: [PATCH] metal : fix ACC op

---
 ggml/src/ggml-metal/ggml-metal-device.m |  2 +-
 ggml/src/ggml-metal/ggml-metal-ops.cpp  | 28 +++++++++++++++----------
 2 files changed, 18 insertions(+), 12 deletions(-)

diff --git a/ggml/src/ggml-metal/ggml-metal-device.m b/ggml/src/ggml-metal/ggml-metal-device.m
index 891d70c85a..00c68837a0 100644
--- a/ggml/src/ggml-metal/ggml-metal-device.m
+++ b/ggml/src/ggml-metal/ggml-metal-device.m
@@ -1058,8 +1058,8 @@ bool ggml_metal_device_supports_op(ggml_metal_device_t dev, const struct ggml_te
         case GGML_OP_MUL:
         case GGML_OP_DIV:
         case GGML_OP_ADD_ID:
-            return ggml_is_contiguous_rows(op->src[0]) && ggml_is_contiguous_rows(op->src[1]) && op->src[0]->type == GGML_TYPE_F32;
         case GGML_OP_ACC:
+            return ggml_is_contiguous_rows(op->src[0]) && ggml_is_contiguous_rows(op->src[1]) && op->src[0]->type == GGML_TYPE_F32;
         case GGML_OP_REPEAT:
         case GGML_OP_SCALE:
         case GGML_OP_FILL:
diff --git a/ggml/src/ggml-metal/ggml-metal-ops.cpp b/ggml/src/ggml-metal/ggml-metal-ops.cpp
index dbf25433c2..7bb06921e8 100644
--- a/ggml/src/ggml-metal/ggml-metal-ops.cpp
+++ b/ggml/src/ggml-metal/ggml-metal-ops.cpp
@@ -628,8 +628,8 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
     GGML_ASSERT(op->src[1]->type == GGML_TYPE_F32);
     GGML_ASSERT(op->type         == GGML_TYPE_F32);
 
-    GGML_ASSERT(ggml_is_contiguous(op->src[0]));
-    GGML_ASSERT(ggml_is_contiguous(op->src[1]));
+    GGML_ASSERT(ggml_is_contiguous_rows(op->src[0]));
+    GGML_ASSERT(ggml_is_contiguous_rows(op->src[1]));
 
     const size_t pnb1 = ((const int32_t *) op->op_params)[0];
     const size_t pnb2 = ((const int32_t *) op->op_params)[1];
@@ -679,10 +679,10 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
     }
 
     ggml_metal_kargs_bin args = {
-        /*.ne00 =*/ ne00,
-        /*.ne01 =*/ ne01,
-        /*.ne02 =*/ ne02,
-        /*.ne03 =*/ ne03,
+        /*.ne00 =*/ ne10,
+        /*.ne01 =*/ ne11,
+        /*.ne02 =*/ ne12,
+        /*.ne03 =*/ ne13,
         /*.nb00 =*/ nb00,
         /*.nb01 =*/ pnb1,
         /*.nb02 =*/ pnb2,
@@ -695,10 +695,10 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
         /*.nb11 =*/ nb11,
         /*.nb12 =*/ nb12,
         /*.nb13 =*/ nb13,
-        /*.ne0  =*/ ne0,
-        /*.ne1  =*/ ne1,
-        /*.ne2  =*/ ne2,
-        /*.ne3  =*/ ne3,
+        /*.ne0  =*/ ne10,
+        /*.ne1  =*/ ne11,
+        /*.ne2  =*/ ne12,
+        /*.ne3  =*/ ne13,
         /*.nb0  =*/ nb0,
         /*.nb1  =*/ pnb1,
         /*.nb2  =*/ pnb2,
@@ -715,7 +715,13 @@ int ggml_metal_op_acc(ggml_metal_op_t ctx, int idx) {
     ggml_metal_encoder_set_buffer  (enc, ggml_metal_get_buffer_id(op->src[1]), 2);
     ggml_metal_encoder_set_buffer  (enc, ggml_metal_get_buffer_id(op),         3);
 
-    const int nth = std::min(ggml_metal_pipeline_max_theads_per_threadgroup(pipeline), ne00);
+    const int nth_max = MIN(256, ggml_metal_pipeline_max_theads_per_threadgroup(pipeline));
+
+    int nth = 1;
+
+    while (2*nth < args.ne0 && nth < nth_max) {
+        nth *= 2;
+    }
 
     ggml_metal_encoder_dispatch_threadgroups(enc, ne11, ne12, ne13, nth, 1, 1);