vram profiling

src/common.h

@@ -687,6 +687,16 @@ INLINE void __prof_d3d11_zone_cleanup_func(TracyCD3D11ZoneCtx *ctx) { ___tracy_d
 #define __prof_d3d11_ctx_release(ctx) ___tracy_d3d11_context_terminate(ctx)
 #define __prof_d3d11_collect(ctx) ___tracy_d3d11_context_collect(ctx)
 
+enum __prof_plot_type {
+    __prof_plot_type_number = TracyPlotFormatNumber,
+    __prof_plot_type_memory = TracyPlotFormatMemory,
+    __prof_plot_type_percentage = TracyPlotFormatPercentage,
+    __prof_plot_type_watt = TracyPlotFormatWatt
+};
+#define __prof_plot_init(name, type, step, fill, color) TracyCPlotConfig(name, type, step, fill, color)
+#define __prof_plot(name, val) TracyCPlot(name, val)
+#define __prof_plot_i(name, val) TracyCPlotI(name, val)
+
 #if PROFILING_CAPTURE_FRAME_IMAGE
 # define __profframeimage(image, width, height, offset, flipped)    TracyCFrameImage((image), (width), (height), (offset), (flipped));
 #else
@@ -723,6 +733,9 @@ INLINE void __prof_d3d11_zone_cleanup_func(TracyCD3D11ZoneCtx *ctx) { ___tracy_d
 #define __prof_d3d11_ctx_alloc(ctx, device, devicectx, name, name_len)
 #define __prof_d3d11_ctx_release(ctx)
 #define __prof_d3d11_collect(ctx)
+#define __prof_plot_init(name, type, step, fill, color)
+#define __prof_plot(name, val)
+#define __prof_plot_i(name, val)
 
 #endif  /* PROFILING */
 

src/gstat.h

@@ -10,6 +10,8 @@
 struct _gstats {
     struct atomic_u64 GSTAT_SOCK_BYTES_SENT;
     struct atomic_u64 GSTAT_SOCK_BYTES_RECEIVED;
+    struct atomic_u64 GSTAT_VRAM_USAGE;
+    struct atomic_u64 GSTAT_VRAM_BUDGET;
     struct atomic_u64 GSTAT_MEMORY_COMMITTED;
     struct atomic_u64 GSTAT_MEMORY_RESERVED;
     struct atomic_u64 GSTAT_NUM_ARENAS;
@@ -17,12 +19,14 @@ struct _gstats {
 
 extern struct _gstats _g_gstats;
 
+#define gstat_set(name, v) atomic_u64_eval_exchange(&_g_gstats.name, (v))
 #define gstat_add(name, v) atomic_u64_eval_add_u64(&_g_gstats.name, (v))
 #define gstat_sub(name, v) atomic_u64_eval_add_i64(&_g_gstats.name, -((i64)(v)))
 #define gstat_get(name) atomic_u64_eval(&_g_gstats.name)
 
 #else
 
+#define gstat_set(name, v)
 #define gstat_add(name, v)
 #define gstat_sub(name, v)
 #define gstat_get(name) 0

src/renderer_d3d11.c

@@ -9,6 +9,7 @@
 #include "inc.h"
 #include "sprite.h"
 #include "log.h"
+#include "gstat.h"
 
 #pragma warning(push, 0)
 # define UNICODE
@@ -17,7 +18,7 @@
 # include <d3d11.h>
 # include <d3dcompiler.h>
 # include <dxgidebug.h>
-# include <dxgi1_3.h>
+# include <dxgi1_4.h>
 #pragma warning(pop)
 
 #pragma comment(lib, "d3d11")
@@ -935,6 +936,42 @@ struct v2i32 renderer_texture_get_size(struct renderer_texture texture)
  * Backbuffer
  * ========================== */
 
+#if GSTAT_ENABLED || PROFILING
+INTERNAL struct DXGI_QUERY_VIDEO_MEMORY_INFO get_memory_info(void)
+{
+    __prof;
+
+    /* Get DXGI device from D3D11 device */
+    IDXGIDevice *dxgiDevice;
+    HRESULT hr = ID3D11Device_QueryInterface(G.dev, &IID_IDXGIDevice, (void **)&dxgiDevice);
+    (UNUSED)hr;
+    ASSERT(SUCCEEDED(hr));
+
+    /* Get DXGI adapter from DXGI device */
+    IDXGIAdapter *dxgiAdapter;
+    hr = IDXGIDevice_GetAdapter(dxgiDevice, &dxgiAdapter);
+    ASSERT(SUCCEEDED(hr));
+
+    IDXGIAdapter3 *dxgiAdapter3 = NULL;
+    hr = IDXGIAdapter_QueryInterface(dxgiAdapter, &IID_IDXGIAdapter3, (void **)&dxgiAdapter3);
+    ASSERT(SUCCEEDED(hr));
+
+    struct DXGI_QUERY_VIDEO_MEMORY_INFO info = ZI;
+    IDXGIAdapter3_QueryVideoMemoryInfo(
+        dxgiAdapter3,
+        0,
+        DXGI_MEMORY_SEGMENT_GROUP_LOCAL,
+        &info
+    );
+
+    IDXGIAdapter_Release(dxgiAdapter3);
+    IDXGIAdapter_Release(dxgiAdapter);
+    IDXGIDevice_Release(dxgiDevice);
+
+    return info;
+}
+#endif
+
 struct renderer_texture renderer_backbuffer_recreate(struct v2i32 size)
 {
     struct renderer_texture res = ZI;
@@ -962,6 +999,28 @@ void renderer_backbuffer_present(i32 vsync)
     }
 #endif
 
+#if GSTAT_ENABLED || PROFILING
+    {
+        struct DXGI_QUERY_VIDEO_MEMORY_INFO info = get_memory_info();
+        u64 vram = info.CurrentUsage;
+        u64 budget = info.Budget;
+        gstat_set(GSTAT_VRAM_USAGE, vram);
+        gstat_set(GSTAT_VRAM_BUDGET, budget);
+# if PROFILING
+        static char *plot_name = NULL;
+        static u64 prev_vram = 0;
+        if (!plot_name) {
+            plot_name = "Video memory usage";
+            __prof_plot_init(plot_name, __prof_plot_type_memory, 1, 1, 0);
+        }
+        if (vram != prev_vram) {
+            __prof_plot_i(plot_name, vram);
+        }
+        prev_vram = vram;
+# endif
+    }
+#endif
+
     i32 flags = 0;
     if (vsync == 0) {
         flags = DXGI_PRESENT_ALLOW_TEARING;

src/sim_step.c

@@ -403,6 +403,9 @@ INTERNAL void test_generate_walls(struct sim_snapshot *world)
     }
 
     /* Sort tile chunks  */
+    /* NOTE: We sort tile chunks from top-left to bottom-right so that cross
+     * chunk tile checks only have to happen in one direction (because we
+     * know the left & top chunks have already been processed) */
     struct sim_ent **sorted_tile_chunks = arena_dry_push(scratch.arena, struct sim_ent *);
     u64 sorted_tile_chunks_count = 0;
     for (u64 ent_index = 0; ent_index < world->num_ents_reserved; ++ent_index) {
@@ -482,6 +485,7 @@ INTERNAL void test_generate_walls(struct sim_snapshot *world)
                             u64 start_hash = rand_u64_from_seed(*(u64 *)&start);
                             struct dict_entry *entry = dict_get_entry(&horizontal_ends_dict, start_hash);
                             if (entry) {
+                                /* Existing wall exists accross chunk boundary */
                                 node = (struct wall_node *)entry->value;
                                 dict_remove_entry(&horizontal_ends_dict, entry);
                             }
@@ -554,6 +558,7 @@ INTERNAL void test_generate_walls(struct sim_snapshot *world)
                             u64 start_hash = rand_u64_from_seed(*(u64 *)&start);
                             struct dict_entry *entry = dict_get_entry(&vertical_ends_dict, start_hash);
                             if (entry) {
+                                /* Existing wall exists accross chunk boundary */
                                 node = (struct wall_node *)entry->value;
                                 dict_remove_entry(&vertical_ends_dict, entry);
                             }

src/user.c

@@ -1849,16 +1849,23 @@ INTERNAL void user_update(void)
             pos.y += spacing;
             pos.y += spacing;
 
-            draw_text(G.ui_cmd_buffer, font, pos, string_format(temp.arena, LIT("Memory usage:                  %F MiB"), FMT_FLOAT((f64)gstat_get(GSTAT_MEMORY_COMMITTED) / 1024 / 1024)));
+            draw_text(G.ui_cmd_buffer, font, pos, string_format(temp.arena, LIT("Memory committed:              %F MiB"), FMT_FLOAT((f64)gstat_get(GSTAT_MEMORY_COMMITTED) / 1024 / 1024)));
             pos.y += spacing;
 
-            draw_text(G.ui_cmd_buffer, font, pos, string_format(temp.arena, LIT("Virtual memory usage:          %F TiB"), FMT_FLOAT((f64)gstat_get(GSTAT_MEMORY_RESERVED) / 1024 / 1024 / 1024 / 1024)));
+            draw_text(G.ui_cmd_buffer, font, pos, string_format(temp.arena, LIT("Virtual memory reserved:       %F TiB"), FMT_FLOAT((f64)gstat_get(GSTAT_MEMORY_RESERVED) / 1024 / 1024 / 1024 / 1024)));
             pos.y += spacing;
 
             draw_text(G.ui_cmd_buffer, font, pos, string_format(temp.arena, LIT("Arenas allocated:              %F"), FMT_UINT(gstat_get(GSTAT_NUM_ARENAS))));
             pos.y += spacing;
             pos.y += spacing;
 
+            draw_text(G.ui_cmd_buffer, font, pos, string_format(temp.arena, LIT("Video memory usage:            %F MiB"), FMT_FLOAT((f64)gstat_get(GSTAT_VRAM_USAGE) / 1024 / 1024)));
+            pos.y += spacing;
+
+            draw_text(G.ui_cmd_buffer, font, pos, string_format(temp.arena, LIT("Video memory budget:           %F GiB"), FMT_FLOAT((f64)gstat_get(GSTAT_VRAM_BUDGET) / 1024 / 1024 / 1024)));
+            pos.y += spacing;
+            pos.y += spacing;
+
 #if 0
             draw_text(G.ui_cmd_buffer, font, pos, string_format(temp.arena, LIT("screen_size: (%F, %F)"), FMT_FLOAT((f64)G.screen_size.x), FMT_FLOAT((f64)G.screen_size.y)));
             pos.y += spacing;
@@ -2029,7 +2036,6 @@ INTERNAL void user_update(void)
 INTERNAL SYS_THREAD_ENTRY_POINT_FUNC_DEF(user_thread_entry_point, arg)
 {
     (UNUSED)arg;
-
     i64 last_frame_ns = 0;
     i64 target_dt_ns = NS_FROM_SECONDS(USER_FPS_LIMIT > (0) ? (1.0 / USER_FPS_LIMIT) : 0);