ggml-cpu : inspect -march and -mcpu to found the CPU (#16333 )

Signed-off-by: Adrien Gallouët <angt@huggingface.co>
vulkan: check glslc executable string (#17144 )
2026-06-29 17:17:40 +02:00 · 2025-11-10 21:03:36 +02:00 · 2025-11-10 16:59:26 +01:00 · 2025-11-10 16:59:10 +01:00 · 2025-11-10 17:14:23 +02:00
5 changed files with 38 additions and 16 deletions
@@ -126,25 +126,36 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
                )
                if (NOT ARM_MCPU_RESULT)
                    string(REGEX MATCH "-mcpu=[^ ']+" ARM_MCPU_FLAG "${ARM_MCPU}")
+                    string(REGEX MATCH "-march=[^ ']+" ARM_MARCH_FLAG "${ARM_MCPU}")
+
+                    # on some old GCC we need to read -march=
+                    if (ARM_MARCH_FLAG AND NOT "${ARM_MARCH_FLAG}" STREQUAL "-march=native")
+                        set(ARM_NATIVE_FLAG "${ARM_MARCH_FLAG}")
+                    elseif(ARM_MCPU_FLAG AND NOT "${ARM_MCPU_FLAG}" STREQUAL "-mcpu=native")
+                        set(ARM_NATIVE_FLAG "${ARM_MCPU_FLAG}")
+                    endif()
                endif()
-                if ("${ARM_MCPU_FLAG}" STREQUAL "")
-                    set(ARM_MCPU_FLAG -mcpu=native)
-                    message(STATUS "ARM -mcpu not found, -mcpu=native will be used")
+
+                if ("${ARM_NATIVE_FLAG}" STREQUAL "")
+                    set(ARM_NATIVE_FLAG -mcpu=native)
+                    message(WARNING "ARM -march/-mcpu not found, -mcpu=native will be used")
+                else()
+                    message(STATUS "ARM detected flags: ${ARM_NATIVE_FLAG}")
                endif()

                include(CheckCXXSourceRuns)

                function(check_arm_feature tag code)
                    set(CMAKE_REQUIRED_FLAGS_SAVE ${CMAKE_REQUIRED_FLAGS})
-                    set(CMAKE_REQUIRED_FLAGS "${ARM_MCPU_FLAG}+${tag}")
+                    set(CMAKE_REQUIRED_FLAGS "${ARM_NATIVE_FLAG}+${tag}")
                    check_cxx_source_runs("${code}" GGML_MACHINE_SUPPORTS_${tag})
                    if (GGML_MACHINE_SUPPORTS_${tag})
-                        set(ARM_MCPU_FLAG_FIX "${ARM_MCPU_FLAG_FIX}+${tag}" PARENT_SCOPE)
+                        set(ARM_NATIVE_FLAG_FIX "${ARM_NATIVE_FLAG_FIX}+${tag}" PARENT_SCOPE)
                    else()
-                        set(CMAKE_REQUIRED_FLAGS "${ARM_MCPU_FLAG}+no${tag}")
+                        set(CMAKE_REQUIRED_FLAGS "${ARM_NATIVE_FLAG}+no${tag}")
                        check_cxx_source_compiles("int main() { return 0; }" GGML_MACHINE_SUPPORTS_no${tag})
                        if (GGML_MACHINE_SUPPORTS_no${tag})
-                            set(ARM_MCPU_FLAG_FIX "${ARM_MCPU_FLAG_FIX}+no${tag}" PARENT_SCOPE)
+                            set(ARM_NATIVE_FLAG_FIX "${ARM_NATIVE_FLAG_FIX}+no${tag}" PARENT_SCOPE)
                        endif()
                    endif()
                    set(CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS_SAVE})
@@ -155,7 +166,7 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
                check_arm_feature(sve     "#include <arm_sve.h>\nint main()  { svfloat32_t _a, _b; volatile svfloat32_t _c = svadd_f32_z(svptrue_b8(), _a, _b); return 0; }")
                check_arm_feature(sme     "#include <arm_sme.h>\n__arm_locally_streaming int main() { __asm__ volatile(\"smstart; smstop;\"); return 0; }")

-                list(APPEND ARCH_FLAGS "${ARM_MCPU_FLAG}${ARM_MCPU_FLAG_FIX}")
+                list(APPEND ARCH_FLAGS "${ARM_NATIVE_FLAG}${ARM_NATIVE_FLAG_FIX}")
            else()
                if (GGML_CPU_ARM_ARCH)
                    list(APPEND ARCH_FLAGS -march=${GGML_CPU_ARM_ARCH})
@@ -6831,7 +6831,7 @@ static void ggml_vk_mul_mat_vec_q_f16(ggml_backend_vk_context * ctx, vk_context&

    vk_buffer d_B = d_D;
    size_t b_buf_offset = 0;
-    uint64_t b_sz = 0;
+    uint64_t b_sz = 1;

    if (enable_bias) {
        const ggml_tensor * add = cgraph->nodes[node_idx + 1];
@@ -6965,7 +6965,7 @@ static void ggml_vk_mul_mat_vec_p021_f16_f32(ggml_backend_vk_context * ctx, vk_c

    vk_buffer d_B = d_D;
    size_t b_buf_offset = 0;
-    uint64_t b_sz = 0;
+    uint64_t b_sz = 1;

    if (enable_bias) {
        const ggml_tensor * add = cgraph->nodes[node_idx + 1];
@@ -7101,7 +7101,7 @@ static void ggml_vk_mul_mat_vec_nc_f16_f32(ggml_backend_vk_context * ctx, vk_con

    vk_buffer d_B = d_D;
    size_t b_buf_offset = 0;
-    uint64_t b_sz = 0;
+    uint64_t b_sz = 1;

    if (enable_bias) {
        const ggml_tensor * add = cgraph->nodes[node_idx + 1];
@@ -7676,7 +7676,7 @@ static void ggml_vk_mul_mat_vec_id_q_f16(ggml_backend_vk_context * ctx, vk_conte

    vk_buffer d_B = d_D;
    size_t b_buf_offset = 0;
-    uint64_t b_sz = 0;
+    uint64_t b_sz = 1;

    if (enable_bias || enable_scale) {
        const ggml_tensor * bias = cgraph->nodes[node_idx + 1]->src[1];
@@ -18,6 +18,7 @@
 #include <algorithm>
 #include <sys/stat.h>
 #include <sys/types.h>
+#include <filesystem>

 #ifdef _WIN32
    #define NOMINMAX
@@ -1080,6 +1081,11 @@ int main(int argc, char** argv) {

    if (args.find("--glslc") != args.end()) {
        GLSLC = args["--glslc"]; // Path to glslc
+
+        if (!std::filesystem::exists(GLSLC) || !std::filesystem::is_regular_file(GLSLC)) {
+            std::cerr << "Error: glslc not found at " << GLSLC << std::endl;
+            return EXIT_FAILURE;
+        }
    }
    if (args.find("--source") != args.end()) {
        input_filepath = args["--source"]; // The shader source file to compile
@@ -151,7 +151,8 @@ bool llama_memory_recurrent::seq_rm(llama_seq_id seq_id, llama_pos p0, llama_pos
        p1 = std::numeric_limits<llama_pos>::max();
    }

-    // models like Mamba or RWKV can't have a state partially erased
+    // models like Mamba or RWKV can't have a state partially erased at the end
+    // of the sequence because their state isn't preserved for previous tokens
    if (seq_id >= (int64_t) size) {
        // could be fatal
        return false;
@@ -160,8 +161,8 @@ bool llama_memory_recurrent::seq_rm(llama_seq_id seq_id, llama_pos p0, llama_pos
        int32_t & tail_id = cells[seq_id].tail;
        if (tail_id >= 0) {
            const auto & cell = cells[tail_id];
-            // partial intersection is invalid
-            if ((0 < p0 && p0 < cell.pos) || (0 < p1 && p1 <= cell.pos)) {
+            // partial intersection is invalid if it includes the final pos
+            if (0 < p0 && p0 <= cell.pos && p1 > cell.pos) {
                //printf("[DEBUG] inside `llama_memory_recurrent::seq_rm`: partial intersection is invalid, so returning false\n");
                return false;
            }
@@ -354,7 +354,11 @@ int main(int argc, char ** argv) {
        }

        // remove any "future" tokens that we might have inherited from the previous session
-        llama_memory_seq_rm(mem, -1, n_matching_session_tokens, -1);
+        if (!llama_memory_seq_rm(mem, -1, n_matching_session_tokens, -1)) {
+            LOG_INF("%s: unable to resuse common prefix\n", __func__);
+            n_matching_session_tokens = 0;
+            llama_memory_seq_rm(mem, -1, -1, -1);
+        }
    }

    LOG_DBG("recalculate the cached logits (check): embd_inp.size() %zu, n_matching_session_tokens %zu, embd_inp.size() %zu, session_tokens.size() %zu\n",
Author	SHA1	Message	Date
Adrien Gallouët	967eb4b2bf	ggml-cpu : inspect -march and -mcpu to found the CPU (#16333 ) Signed-off-by: Adrien Gallouët <angt@huggingface.co>	2025-11-10 21:03:36 +02:00
Ruben Ortlam	f117be185e	vulkan: check glslc executable string (#17144 )	2025-11-10 16:59:26 +01:00
Ruben Ortlam	85234a4b3a	vulkan: fix validation issue introduced by #16868 (#17145 )	2025-11-10 16:59:10 +01:00
Gabe Goodhart	0c74f32632	memory: Hybrid context shift (#17009 ) * feat(memory): Only fail partial erasure of recurrent tail The recurrent state is always assumed to be the state as of the last update from the final token in the sequence. When doing a partial erasure, if the range does not include the final token, the erasure can be considered a success since any memory used for the sequence prior to the final token (which is no memory) has been successfully removed. There is one potential case that this doesn't address which is the pruning of cache to remove sensitive data from the context. This wouldn't work for attention cache partial removal (in the middle) either since the KV state is linearly-dependent and states in later sequence positions would still be based on the state from the sensitive data, even if that data is no longer cached, so I don't think this is relevant, but it is worth noting that the semantics of this change for a partial erasure in the middle of the cache are essentially "my context is already compressed" and not "all trace of the removed tokens has been removed." https://github.com/ggml-org/llama.cpp/issues/16768 Branch: HybridContextShift-16768 Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(main): Check the output of seq_rm for prefix matching This prefix matching is explicitly attempting to remove the tokens at the end of the sequence that don't match. This is the operation that can't be performed on a recurrent cache due to the state being updated in place, so if this removal fails, we need to clear the whole cache. https://github.com/ggml-org/llama.cpp/issues/16768 Branch: HybridContextShift-16768 Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(memory): Fix condition for partial erasure failure if p0 > pos Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: compilade <git@compilade.net> * style: Fix extra parens Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * fix(main.cpp): Set n_matching_session_tokens to 0 on cache clear https://github.com/ggml-org/llama.cpp/issues/16768 Branch: HybridContextShift-16768 Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> --------- Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: compilade <git@compilade.net> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>	2025-11-10 17:14:23 +02:00