vulkan: Check maxStorageBufferRange in supports_op (llama/18709)

* vulkan: Check maxStorageBufferRange in supports_op

* skip maxStorageBufferRange check when shader64BitIndexing is enabled

diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
index deed5055d5463742770a0def84be32c013418a67..0fabbcec31d416f1b4ce55c763fd4488a54cefb3 100644 (file)
--- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -14413,13 +14413,29 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
     ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)dev->context;
     const vk_device& device = ggml_vk_get_device(ctx->device);
 
+    const bool uses_bda = (op->op == GGML_OP_IM2COL || op->op == GGML_OP_IM2COL_3D) &&
+                          device->shader_int64 && device->buffer_device_address;
+
+    auto const & tensor_size_supported = [&](size_t tensor_size) {
+        if (tensor_size > device->max_buffer_size) {
+            return false;
+        }
+        // For im2col shaders using BDA, maxStorageBufferRange limit doesn't apply.
+        // If shader64BitIndexing is enabled, maxStorageBufferRange limit doesn't apply.
+        if (!uses_bda && !device->shader_64b_indexing) {
+            if (tensor_size > device->properties.limits.maxStorageBufferRange) {
+                return false;
+            }
+        }
+        return true;
+    };
     // reject any tensors larger than the max buffer size
     for (int i = 0; i < GGML_MAX_SRC; i++) {
-        if (op->src[i] && ggml_nbytes(op->src[i]) > device->max_buffer_size) {
+        if (op->src[i] && !tensor_size_supported(ggml_nbytes(op->src[i]))) {
             return false;
         }
     }
-    if (ggml_nbytes(op) > device->max_buffer_size) {
+    if (!tensor_size_supported(ggml_nbytes(op))) {
         return false;
     }