mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-21 14:26:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml update to b7108 (#12992)

Browse Source 
* Revert "vulkan: temporary cary of vulkan fixes (#12971)"

This reverts commit 3a9e8e9fd4.

* ggml update to b7087

* fix argsort on metal

* update to b7108

* fix bakllava regression

This model lacks the metadata for the projector type.

* update to b7209

* fix TopK perf

* only build arm code on arm
This commit is contained in:
Daniel Hiltgen
2025-12-03 19:43:29 -08:00
committed by GitHub
parent 854d40edc5
commit 0cf7794b16
303 changed files with 32711 additions and 23435 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
llama/patches/0028-Add-memory-detection-using-DXGI-PDH.patch
View File

@@ -6,13 +6,13 @@ Subject: [PATCH] Add memory detection using DXGI + PDH
---
ggml/src/CMakeLists.txt | 1 +
ggml/src/ggml-impl.h | 3 +
ggml/src/ggml-vulkan/ggml-vulkan.cpp | 29 ++-
ggml/src/ggml-vulkan/ggml-vulkan.cpp | 26 ++-
ggml/src/mem_dxgi_pdh.cpp | 297 +++++++++++++++++++++++++++
4 files changed, 327 insertions(+), 3 deletions(-)
4 files changed, 325 insertions(+), 2 deletions(-)
create mode 100644 ggml/src/mem_dxgi_pdh.cpp
diff --git a/ggml/src/CMakeLists.txt b/ggml/src/CMakeLists.txt
index 03f359ae9..4b3e5efb5 100644
index ac8f38464..faa1beed2 100644
--- a/ggml/src/CMakeLists.txt
+++ b/ggml/src/CMakeLists.txt
@@ -211,6 +211,7 @@ add_library(ggml-base
@@ -22,9 +22,9 @@ index 03f359ae9..4b3e5efb5 100644
+ mem_dxgi_pdh.cpp
gguf.cpp)
target_include_directories(ggml-base PRIVATE .)
set_target_properties(ggml-base PROPERTIES
diff --git a/ggml/src/ggml-impl.h b/ggml/src/ggml-impl.h
index 44ae76d66..639d551a2 100644
index 1c07e767a..0da3e065b 100644
--- a/ggml/src/ggml-impl.h
+++ b/ggml/src/ggml-impl.h
@@ -684,6 +684,9 @@ GGML_API void ggml_nvml_release();
@@ -38,10 +38,10 @@ index 44ae76d66..639d551a2 100644
#ifdef __cplusplus
}
diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
index ca02ea079..c12b069e5 100644
index a234eda2e..c98f98c73 100644
--- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -73,6 +73,7 @@ DispatchLoaderDynamic & ggml_vk_default_dispatcher();
@@ -74,6 +74,7 @@ DispatchLoaderDynamic & ggml_vk_default_dispatcher();
#define VK_KHR_SHADER_BFLOAT16_EXTENSION_NAME "VK_KHR_shader_bfloat16"
#define VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_BFLOAT16_FEATURES_KHR ((VkStructureType)1000141000)
#define VK_COMPONENT_TYPE_BFLOAT16_KHR ((VkComponentTypeKHR)1000141000)
@@ -49,7 +49,7 @@ index ca02ea079..c12b069e5 100644
typedef struct VkPhysicalDeviceShaderBfloat16FeaturesKHR {
VkStructureType sType;
@@ -12802,6 +12803,7 @@ struct ggml_backend_vk_device_context {
@@ -13655,6 +13656,7 @@ struct ggml_backend_vk_device_context {
std::string pci_id;
std::string id;
std::string uuid;
@@ -57,8 +57,8 @@ index ca02ea079..c12b069e5 100644
int major;
int minor;
int driver_major;
@@ -12817,8 +12819,22 @@ void ggml_backend_vk_get_device_memory(ggml_backend_vk_device_context *ctx, size
vk::PhysicalDeviceMemoryProperties memprops = vkdev.getMemoryProperties();
@@ -13673,6 +13675,20 @@ void ggml_backend_vk_get_device_memory(ggml_backend_vk_device_context *ctx, size
vk::PhysicalDeviceProperties2 props2;
vkdev.getProperties2(&props2);
+ GGML_LOG_DEBUG("ggml_backend_vk_get_device_memory called: uuid %s\n", ctx->uuid.c_str());
@@ -76,22 +76,17 @@ index ca02ea079..c12b069e5 100644
+ ggml_dxgi_pdh_release();
+ }
- if (!ctx->is_integrated_gpu)
+ if (!ctx->is_integrated_gpu)
if (!is_integrated_gpu)
{
// Use vendor specific management libraries for best VRAM reporting if available
switch (props2.properties.vendorID) {
@@ -12846,8 +12862,8 @@ void ggml_backend_vk_get_device_memory(ggml_backend_vk_device_context *ctx, size
break;
}
@@ -13704,7 +13720,6 @@ void ggml_backend_vk_get_device_memory(ggml_backend_vk_device_context *ctx, size
}
- // else fallback to memory budget if supported
// else fallback to memory budget if supported
+ // else fallback to memory budget if supported
*total = 0;
*free = 0;
vk::PhysicalDeviceMemoryBudgetPropertiesEXT mem_budget_props;
@@ -13500,7 +13516,6 @@ static ggml_backend_dev_t ggml_backend_vk_reg_get_device(ggml_backend_reg_t reg,
-
if (membudget_supported) {
memprops.pNext = &budgetprops;
}
@@ -14440,7 +14455,6 @@ static ggml_backend_dev_t ggml_backend_vk_reg_get_device(ggml_backend_reg_t reg,
/* .reg = */ reg,
/* .context = */ ctx,
});
@@ -99,7 +94,7 @@ index ca02ea079..c12b069e5 100644
// Gather additional information about the device
int dev_idx = vk_instance.device_indices[i];
vk::PhysicalDeviceProperties props1;
@@ -13523,6 +13538,14 @@ static ggml_backend_dev_t ggml_backend_vk_reg_get_device(ggml_backend_reg_t reg,
@@ -14463,6 +14477,14 @@ static ggml_backend_dev_t ggml_backend_vk_reg_get_device(ggml_backend_reg_t reg,
}
}
ctx->uuid = oss.str();