From: slaren <redacted>
Date: Thu, 21 Mar 2024 12:59:53 +0000 (+0100)
Subject: cuda : fix LLAMA_CUDA_F16 build (#6197)
X-Git-Tag: upstream/0.0.4488~2001
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=03a8f8fafec2e21009be9fe7297d92e5d4538964;p=pkg%2Fggml%2Fsources%2Fllama.cpp

cuda : fix LLAMA_CUDA_F16 build (#6197)
---

diff --git a/ggml-cuda.cu b/ggml-cuda.cu
index 280839ea..04c6f5d0 100644
--- a/ggml-cuda.cu
+++ b/ggml-cuda.cu
@@ -9453,7 +9453,7 @@ static void ggml_cuda_op_dequantize_mul_mat_vec(
 
     // on some GPUs it is faster to convert src1 to half and to use half precision intrinsics
 #ifdef GGML_CUDA_F16
-    cuda_pool_alloc<half> src1_dfloat_a;
+    ggml_cuda_pool_alloc<half> src1_dfloat_a(ctx.pool());
     half * src1_dfloat = nullptr; // dfloat == half
 
     bool src1_convert_f16 =