diff options
| author | 0cc4m <picard12@live.de> | 2024-02-01 19:25:24 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-02-01 19:25:24 +0100 |
| commit | 4d0924a8902010d31bd737b6f1f594943d120d0f (patch) | |
| tree | 091227c1265488e6a528f280304b6ad92d6e8e17 /ggml-vulkan.cpp | |
| parent | 8ca511cadee2c67f0bd8c7034a2513778ee9a1b7 (diff) | |
Vulkan Phi Fix for AMD Proprietary Drivers (#5260)
* Replace tanh to avoid NaN in gelu shader on AMD proprietary driver
* Fix another Vulkan CPY buffer size bug
Diffstat (limited to 'ggml-vulkan.cpp')
| -rw-r--r-- | ggml-vulkan.cpp | 17 |
1 files changed, 12 insertions, 5 deletions
diff --git a/ggml-vulkan.cpp b/ggml-vulkan.cpp index bccc40bf..b1e0006b 100644 --- a/ggml-vulkan.cpp +++ b/ggml-vulkan.cpp @@ -2876,6 +2876,9 @@ static void ggml_vk_op_f32(vk_context * ctx, const ggml_tensor * src0, const ggm x_sz = ggml_nbytes(src0); d_sz = ggml_nbytes(dst); + if (extra_src0->offset + x_sz >= d_X->size) { + x_sz = VK_WHOLE_SIZE; + } if (extra->offset + d_sz >= d_D->size) { d_sz = VK_WHOLE_SIZE; } @@ -2911,12 +2914,16 @@ static void ggml_vk_op_f32(vk_context * ctx, const ggml_tensor * src0, const ggm break; } - x_sz *= ne02 * ne03; - if (y_sz != VK_WHOLE_SIZE) { - y_sz *= ne12 * ne13; - } if (op != GGML_OP_CPY) { - d_sz *= ne02 * ne03; + if (x_sz != VK_WHOLE_SIZE) { + x_sz *= ne02 * ne03; + } + if (y_sz != VK_WHOLE_SIZE) { + y_sz *= ne12 * ne13; + } + if (d_sz != VK_WHOLE_SIZE) { + d_sz *= ne02 * ne03; + } } if (!use_src1 && op == GGML_OP_SOFT_MAX) { |