From a4539e1636d94f30bf1c968e33c9c4fc858a423d Mon Sep 17 00:00:00 2001
From: Slaren <2141330+slaren@users.noreply.github.com>
Date: Sat, 8 Apr 2023 13:41:57 +0200
Subject: [PATCH] Use the work buffer instead to fix MSVC build

---
 ggml.c | 17 +++++++++++++----
 1 file changed, 13 insertions(+), 4 deletions(-)

diff --git a/ggml.c b/ggml.c
index 3b32aff9b689b7..dd089b2a78439a 100644
--- a/ggml.c
+++ b/ggml.c
@@ -5350,6 +5350,8 @@ static void ggml_compute_forward_add_q_f32(
     const int ir0 = dr*ith;
     const int ir1 = MIN(ir0 + dr, nr);
 
+    float * wdata = (float*) params->wdata + ne00 * ith;
+
     for (int ir = ir0; ir < ir1; ++ir) {
         // src0 indices
         const int i03 = ir/(ne02*ne01);
@@ -5372,12 +5374,11 @@ static void ggml_compute_forward_add_q_f32(
         assert(ne00 % 32 == 0);
 
         // unquantize row from src0 to temp buffer
-        float tmp[ne00];
-        dequantize_row_q(src0_row, tmp, ne00);
+        dequantize_row_q(src0_row, wdata, ne00);
         // add src1
-        ggml_vec_acc_f32(ne00, tmp, src1_row);
+        ggml_vec_acc_f32(ne00, wdata, src1_row);
         // quantize row to dst
-        quantize_row_q(tmp, dst_row, ne00);
+        quantize_row_q(wdata, dst_row, ne00);
     }
 }
 
@@ -9566,6 +9567,14 @@ void ggml_graph_compute(struct ggml_context * ctx, struct ggml_cgraph * cgraph)
                 case GGML_OP_ADD:
                     {
                         node->n_tasks = n_threads;
+
+                        size_t cur = 0;
+
+                        if (node->src0->type == GGML_TYPE_Q4_0 || node->src0->type == GGML_TYPE_Q4_1) {
+                            cur = GGML_TYPE_SIZE[GGML_TYPE_F32] * node->src0->ne[0] * n_threads;
+                        }
+
+                        work_size = MAX(work_size, cur);
                     } break;
                 case GGML_OP_SUB:
                 case GGML_OP_MUL: