metal : log more info on error (ggerganov#6987)

teleprint-me · Apr 30, 2024 · a68a1e7 · a68a1e7
1 parent 9c67c27
commit a68a1e7
Showing 1 changed file with 39 additions and 0 deletions.
diff --git a/ggml-metal.m b/ggml-metal.m
@@ -2779,6 +2779,45 @@ static enum ggml_status ggml_metal_graph_compute(
         MTLCommandBufferStatus status = [command_buffer status];
         if (status != MTLCommandBufferStatusCompleted) {
             GGML_METAL_LOG_INFO("%s: command buffer %d failed with status %lu\n", __func__, i, status);
+            if (status == MTLCommandBufferStatusError) {
+                MTLCommandBufferError error_code = [command_buffer error].code;
+                switch (error_code) {
+                    case MTLCommandBufferErrorNone:
+                        GGML_METAL_LOG_INFO("no error code reported\n");
+                        break;
+                    case MTLCommandBufferErrorTimeout:
+                        GGML_METAL_LOG_INFO("timeout\n");
+                        break;
+                    case MTLCommandBufferErrorPageFault:
+                        GGML_METAL_LOG_INFO("unserviceable page fault\n");
+                        break;
+                    case MTLCommandBufferErrorOutOfMemory:
+                        GGML_METAL_LOG_INFO("out of memory\n");
+                        break;
+                    case MTLCommandBufferErrorInvalidResource:
+                        GGML_METAL_LOG_INFO("invalid reference to resource\n");
+                        break;
+                    case MTLCommandBufferErrorMemoryless:
+                        GGML_METAL_LOG_INFO("GPU ran out of one or more of its internal resources that support memoryless render pass attachments\n");
+                        break;
+                    case MTLCommandBufferErrorDeviceRemoved:
+                        GGML_METAL_LOG_INFO("device removed\n");
+                        break;
+                    case MTLCommandBufferErrorStackOverflow:
+                        GGML_METAL_LOG_INFO("kernel function of tile shader used too many stack frames\n");
+                        break;
+                    case MTLCommandBufferErrorAccessRevoked:
+                        GGML_METAL_LOG_INFO("access to device revoked by system\n");
+                        break;
+                    case MTLCommandBufferErrorInternal:
+                        GGML_METAL_LOG_INFO("internal error\n");
+                        break;
+                    default:
+                        GGML_METAL_LOG_INFO("unknown error %lu\n", error_code);
+                        break;
+                }
+            }
+
             return GGML_STATUS_FAILED;
         }
     }