From ccd25927829feda05045702be1843e5d529d4da1 Mon Sep 17 00:00:00 2001 From: 0cc4m Date: Sun, 6 Aug 2023 05:25:33 +0200 Subject: [PATCH] Add further missing barrier --- ggml-vulkan.cpp | 1 + 1 file changed, 1 insertion(+) diff --git a/ggml-vulkan.cpp b/ggml-vulkan.cpp index 89be9a57e..83cbe42e2 100644 --- a/ggml-vulkan.cpp +++ b/ggml-vulkan.cpp @@ -2346,6 +2346,7 @@ static void ggml_vk_mul_f32(const ggml_tensor * src0, const ggml_tensor * src1, ggml_vk_sync_buffers(s.buffer, { ggml_vk_subbuffer(d_X), ggml_vk_subbuffer(d_Y) }, vk_device.compute_queue, vk::AccessFlagBits::eTransferWrite, vk::AccessFlagBits::eShaderRead, true); ggml_vk_sync_buffers(s.buffer, { ggml_vk_subbuffer(d_D) }, vk_device.compute_queue, vk::AccessFlagBits::eTransferRead, vk::AccessFlagBits::eShaderWrite, false); ggml_vk_dispatch_pipeline(s, vk_pipeline_mul_f32, { { d_X, 0, buf_sz }, { d_Y, 0, (uint32_t) d_Y.size }, { d_D, 0, buf_sz } }, sizeof(int) * pc.size(), pc.data(), { (uint32_t)ne00, (uint32_t)ne01, 1}); + ggml_vk_sync_buffers(s.buffer, { ggml_vk_subbuffer(d_D) }, vk_device.compute_queue, vk::AccessFlagBits::eShaderWrite, vk::AccessFlagBits::eTransferRead, true); // copy dst to host float * d = (float *) ((char *) dst->data + i02*nb2 + i03*nb3);