metal : migrate ggml-metal.m to ggml-metal.cpp

ggerganov · ggerganov · commit 0048056bb317 · 2025-09-17T00:10:24.000+03:00
ggml-ci
diff --git a/ggml/src/ggml-metal/CMakeLists.txt b/ggml/src/ggml-metal/CMakeLists.txt
@@ -5,7 +5,7 @@ find_library(METALKIT_FRAMEWORK MetalKit   REQUIRED)
 message(STATUS "Metal framework found")
 
 ggml_add_backend_library(ggml-metal
-                         ggml-metal.m
+                         ggml-metal.cpp
                          ggml-metal-device.m
                          ggml-metal-device.cpp
                          ggml-metal-common.cpp
diff --git a/ggml/src/ggml-metal/ggml-metal-context.m b/ggml/src/ggml-metal/ggml-metal-context.m
@@ -466,11 +466,11 @@ ggml_metal_t ggml_metal_init(ggml_metal_device_t ctx_dev) {
 
     res->ctx_dev = ctx_dev;
 
-    struct ggml_metal_device_props props_dev = ggml_metal_device_get_props(ctx_dev);
+    const struct ggml_metal_device_props * props_dev = ggml_metal_device_get_props(ctx_dev);
 
     res->d_queue = dispatch_queue_create("ggml-metal", DISPATCH_QUEUE_CONCURRENT);
 
-    res->use_bfloat      = props_dev.has_bfloat;
+    res->use_bfloat      = props_dev->has_bfloat;
     res->use_fusion      = getenv("GGML_METAL_FUSION_DISABLE") == nil;
     res->use_concurrency = getenv("GGML_METAL_CONCURRENCY_DISABLE") == nil;
 
@@ -540,9 +540,9 @@ ggml_metal_t ggml_metal_init(ggml_metal_device_t ctx_dev) {
             GGML_LOG_WARN("%s: skipping %-40s (not supported)\n", __func__, "kernel_"#name); \
         }
 
-        const bool has_simdgroup_mm        = props_dev.has_simdgroup_mm;
-        const bool has_simdgroup_reduction = props_dev.has_simdgroup_reduction;
-        const bool has_bfloat              = props_dev.has_bfloat;
+        const bool has_simdgroup_mm        = props_dev->has_simdgroup_mm;
+        const bool has_simdgroup_reduction = props_dev->has_simdgroup_reduction;
+        const bool has_bfloat              = props_dev->has_bfloat;
 
         // simd_sum and simd_max requires MTLGPUFamilyApple7
 
@@ -1258,7 +1258,7 @@ static int ggml_metal_encode_node(struct ggml_metal_encode_context * ctx_enc, in
         }
     }
 
-    struct ggml_metal_device_props props_dev = ggml_metal_device_get_props(ctx->ctx_dev);
+    const struct ggml_metal_device_props * props_dev = ggml_metal_device_get_props(ctx->ctx_dev);
 
     switch (dst->op) {
         case GGML_OP_CONCAT:
@@ -2637,7 +2637,7 @@ static int ggml_metal_encode_node(struct ggml_metal_encode_context * ctx_enc, in
                 } else
                 // for now the matrix-matrix multiplication kernel only works on A14+/M1+ SoCs
                 // AMD GPU and older A-chips will reuse matrix-vector multiplication kernel
-                if (props_dev.supports_gpu_family_apple7 &&
+                if (props_dev->supports_gpu_family_apple7 &&
                     !ggml_is_transposed(src0) &&
                     !ggml_is_transposed(src1) &&
                     src1t == GGML_TYPE_F32 &&
@@ -2975,7 +2975,7 @@ static int ggml_metal_encode_node(struct ggml_metal_encode_context * ctx_enc, in
 
                 // for now the matrix-matrix multiplication kernel only works on A14+/M1+ SoCs
                 // AMD GPU and older A-chips will reuse matrix-vector multiplication kernel
-                if (props_dev.supports_gpu_family_apple7 &&
+                if (props_dev->supports_gpu_family_apple7 &&
                     ne00 % 32 == 0 && ne00 >= 64 &&
                     (ne21 >= ne21_mm_id_min)) {
                     GGML_ASSERT(ne00 % 4 == 0);
@@ -3024,7 +3024,7 @@ static int ggml_metal_encode_node(struct ggml_metal_encode_context * ctx_enc, in
 
                         const size_t smem = ne02*ne20*sizeof(uint16_t);
 
-                        GGML_ASSERT(smem <= props_dev.max_theadgroup_memory_size);
+                        GGML_ASSERT(smem <= props_dev->max_theadgroup_memory_size);
 
                         [encoder setComputePipelineState:pipeline];
                         [encoder setBytes:&args    length:sizeof(args) atIndex:0];
@@ -4209,7 +4209,7 @@ static int ggml_metal_encode_node(struct ggml_metal_encode_context * ctx_enc, in
                     //    nsgmax = 2;
                     //    while (true) {
                     //        const size_t smem = FATTN_SMEM(nsgmax);
-                    //        if (smem > props_dev.max_theadgroup_memory_size) {
+                    //        if (smem > props_dev->max_theadgroup_memory_size) {
                     //            break;
                     //        }
                     //        nsgmax *= 2;
@@ -4277,8 +4277,8 @@ static int ggml_metal_encode_node(struct ggml_metal_encode_context * ctx_enc, in
 
                     [encoder setBuffer:id_dst offset:offs_dst atIndex:6];
 
-                    //printf("smem: %zu, max: %zu, nsg = %d, ne02 = %d, ne12 = %d\n", smem, props_dev.max_theadgroup_memory_size, (int) nsg, ne02, ne12);
-                    GGML_ASSERT(smem <= props_dev.max_theadgroup_memory_size);
+                    //printf("smem: %zu, max: %zu, nsg = %d, ne02 = %d, ne12 = %d\n", smem, props_dev->max_theadgroup_memory_size, (int) nsg, ne02, ne12);
+                    GGML_ASSERT(smem <= props_dev->max_theadgroup_memory_size);
                     [encoder setThreadgroupMemoryLength:smem atIndex:0];
                     [encoder dispatchThreadgroups:MTLSizeMake((ne01 + nqptg - 1)/nqptg, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(32, nsg, 1)];
 #undef FATTN_SMEM
@@ -4305,7 +4305,7 @@ static int ggml_metal_encode_node(struct ggml_metal_encode_context * ctx_enc, in
                     while (true) {
                         const size_t smem = FATTN_SMEM(nsgmax);
                         // avoid using more than half of the threadgroup memory - can cause slow downs especially for large head sizes
-                        if (smem > props_dev.max_theadgroup_memory_size/2) {
+                        if (smem > props_dev->max_theadgroup_memory_size/2) {
                             break;
                         }
                         nsgmax *= 2;
@@ -4394,8 +4394,8 @@ static int ggml_metal_encode_node(struct ggml_metal_encode_context * ctx_enc, in
 
                     const size_t smem = FATTN_SMEM(nsg);
 
-                    //printf("smem: %zu, max: %zu, nsg = %d, nsgmax = %d\n", smem, props_dev.max_theadgroup_memory_size, (int) nsg, (int) nsgmax);
-                    GGML_ASSERT(smem <= props_dev.max_theadgroup_memory_size);
+                    //printf("smem: %zu, max: %zu, nsg = %d, nsgmax = %d\n", smem, props_dev->max_theadgroup_memory_size, (int) nsg, (int) nsgmax);
+                    GGML_ASSERT(smem <= props_dev->max_theadgroup_memory_size);
 
                     if (nwg == 1) {
                         // using 1 workgroup -> write the result directly into dst
diff --git a/ggml/src/ggml-metal/ggml-metal-device.h b/ggml/src/ggml-metal/ggml-metal-device.h
@@ -42,7 +42,7 @@ void * ggml_metal_device_get_queue  (ggml_metal_device_t ctx); // id<MTLCommandQ
 void ggml_metal_device_get_memory(ggml_metal_device_t ctx, size_t * free, size_t * total);
 bool ggml_metal_device_supports_op(ggml_metal_device_t ctx, const struct ggml_tensor * op);
 
-struct ggml_metal_device_props ggml_metal_device_get_props(ggml_metal_device_t ctx);
+const struct ggml_metal_device_props * ggml_metal_device_get_props(ggml_metal_device_t ctx);
 
 //
 // device buffers
diff --git a/ggml/src/ggml-metal/ggml-metal-device.m b/ggml/src/ggml-metal/ggml-metal-device.m
@@ -538,8 +538,8 @@ bool ggml_metal_device_supports_op(ggml_metal_device_t ctx, const struct ggml_te
     }
 }
 
-struct ggml_metal_device_props ggml_metal_device_get_props(ggml_metal_device_t ctx) {
-    return ctx->props;
+const struct ggml_metal_device_props * ggml_metal_device_get_props(ggml_metal_device_t ctx) {
+    return &ctx->props;
 }
 
 //
@@ -686,9 +686,9 @@ ggml_metal_buffer_t ggml_metal_buffer_init(ggml_metal_device_t ctx, size_t size,
         size_aligned += (size_page - (size_aligned % size_page));
     }
 
-    const struct ggml_metal_device_props props_dev = ggml_metal_device_get_props(ctx);
+    const struct ggml_metal_device_props * props_dev = ggml_metal_device_get_props(ctx);
 
-    shared = shared && props_dev.use_shared_buffers;
+    shared = shared && props_dev->use_shared_buffers;
 
     // allocate shared buffer if the device supports it and it is required by the buffer type
     if (shared) {
@@ -711,7 +711,7 @@ ggml_metal_buffer_t ggml_metal_buffer_init(ggml_metal_device_t ctx, size_t size,
         res->buffers[0].metal = nil;
 
         if (size_aligned > 0) {
-            if (props_dev.use_shared_buffers &&shared) {
+            if (props_dev->use_shared_buffers &&shared) {
                 res->buffers[0].metal = [res->device newBufferWithBytesNoCopy:res->all_data
                                                                   length:size_aligned
                                                                  options:MTLResourceStorageModeShared
@@ -732,7 +732,7 @@ ggml_metal_buffer_t ggml_metal_buffer_init(ggml_metal_device_t ctx, size_t size,
         return NULL;
     }
 
-    res->use_residency_sets = props_dev.use_residency_sets;
+    res->use_residency_sets = props_dev->use_residency_sets;
 
     if (!ggml_metal_buffer_rset_init(res)) {
         GGML_LOG_ERROR("%s: error: failed to initialize residency set\n", __func__);
@@ -772,10 +772,10 @@ ggml_metal_buffer_t ggml_metal_buffer_map(ggml_metal_device_t ctx, void * ptr, s
     res->device = ggml_metal_device_get_device(ctx);
     res->queue  = ggml_metal_device_get_queue (ctx);
 
-    const struct ggml_metal_device_props props_dev = ggml_metal_device_get_props(ctx);
+    const struct ggml_metal_device_props * props_dev = ggml_metal_device_get_props(ctx);
 
     // the buffer fits into the max buffer size allowed by the device
-    if (size_aligned <= props_dev.max_buffer_size) {
+    if (size_aligned <= props_dev->max_buffer_size) {
         res->buffers[res->n_buffers].data  = ptr;
         res->buffers[res->n_buffers].size  = size;
         res->buffers[res->n_buffers].metal = nil;
@@ -797,8 +797,8 @@ ggml_metal_buffer_t ggml_metal_buffer_map(ggml_metal_device_t ctx, void * ptr, s
         // this overlap between the views will guarantee that the tensor with the maximum size will fully fit into
         // one of the views
         const size_t size_ovlp = ((max_tensor_size + size_page - 1) / size_page + 1) * size_page; // round-up 2 pages just in case
-        const size_t size_step = props_dev.max_buffer_size - size_ovlp;
-        const size_t size_view = props_dev.max_buffer_size;
+        const size_t size_step = props_dev->max_buffer_size - size_ovlp;
+        const size_t size_view = props_dev->max_buffer_size;
 
         for (size_t i = 0; i < size; i += size_step) {
             const size_t size_step_aligned = (i + size_view <= size) ? size_view : (size_aligned - i);
@@ -827,7 +827,7 @@ ggml_metal_buffer_t ggml_metal_buffer_map(ggml_metal_device_t ctx, void * ptr, s
         }
     }
 
-    res->use_residency_sets = props_dev.use_residency_sets;
+    res->use_residency_sets = props_dev->use_residency_sets;
 
     if (!ggml_metal_buffer_rset_init(res)) {
         GGML_LOG_ERROR("%s: error: failed to initialize residency set\n", __func__);
diff --git a/ggml/src/ggml-metal/ggml-metal.cpp b/ggml/src/ggml-metal/ggml-metal.cpp
