Feat/vulkan

.clang-format

@@ -71,8 +71,8 @@ IndentWidth:     4
 AccessModifierOffset: -4
 IndentWrappedFunctionNames: false
 KeepEmptyLinesAtTheStartOfBlocks: false
-MacroBlockBegin: '.*_BEGIN$' # only PREC_BEGIN ?
-MacroBlockEnd:   '.*_END$'
+#MacroBlockBegin: '.*_BEGIN$' # only PREC_BEGIN ?
+#MacroBlockEnd:   '.*_END$'
 MaxEmptyLinesToKeep: 2
 #PenaltyBreakBeforeFirstCallParameter: 19
 #PenaltyBreakComment: 300

.github/workflows/build.yml

@@ -67,7 +67,7 @@ jobs:
           build_param:
           artifact_name: xemu-win-release
     env:
-      DOCKER_IMAGE_NAME: ghcr.io/xemu-project/xemu-win64-toolchain:sha-8152913
+      DOCKER_IMAGE_NAME: ghcr.io/xemu-project/xemu-win64-toolchain:sha-c6bad86
 
     steps:
     - name: Download source package

.gitmodules

@@ -82,9 +82,18 @@
 [submodule "tomlplusplus"]
 	path = tomlplusplus
 	url = https://github.com/marzer/tomlplusplus
-[submodule "hw/xbox/nv2a/thirdparty/nv2a_vsh_cpu"]
-	path = hw/xbox/nv2a/thirdparty/nv2a_vsh_cpu
+[submodule "hw/xbox/nv2a/pgraph/thirdparty/nv2a_vsh_cpu"]
+	path = hw/xbox/nv2a/pgraph/thirdparty/nv2a_vsh_cpu
 	url = https://github.com/abaire/nv2a_vsh_cpu.git
 [submodule "ui/thirdparty/httplib"]
 	path = ui/thirdparty/httplib
 	url = https://github.com/yhirose/cpp-httplib
+[submodule "hw/xbox/nv2a/pgraph/vk/thirdparty/VulkanMemoryAllocator"]
+	path = thirdparty/VulkanMemoryAllocator
+	url = https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator
+[submodule "thirdparty/volk"]
+	path = thirdparty/volk
+	url = https://github.com/zeux/volk
+[submodule "thirdparty/SPIRV-Reflect"]
+	path = thirdparty/SPIRV-Reflect
+	url = https://github.com/KhronosGroup/SPIRV-Reflect

config_spec.yml

@@ -130,6 +130,14 @@ input:
       default: 18 # w
 
 display:
+  renderer:
+    type: enum
+    values: ["NULL", OPENGL, VULKAN]
+    default: OPENGL
+  vulkan:
+    validation_layers: bool
+    debug_shaders: bool
+    assert_on_validation_msg: bool
   quality:
     surface_scale:
       type: integer

configure

@@ -237,7 +237,7 @@ else
     git_submodules_action="ignore"
 fi
 
-git_submodules="ui/keycodemapdb ui/thirdparty/imgui ui/thirdparty/implot ui/thirdparty/httplib util/xxHash tomlplusplus genconfig hw/xbox/nv2a/thirdparty/nv2a_vsh_cpu"
+git_submodules="ui/keycodemapdb ui/thirdparty/imgui ui/thirdparty/implot ui/thirdparty/httplib util/xxHash tomlplusplus genconfig hw/xbox/nv2a/pgraph/thirdparty/nv2a_vsh_cpu thirdparty/volk thirdparty/VulkanMemoryAllocator thirdparty/SPIRV-Reflect"
 git="git"
 
 # Don't accept a target_list environment variable.

debian/control

@@ -16,6 +16,7 @@ Build-Depends: debhelper (>= 11),
  libssl-dev,
  libpcap-dev,
  libslirp-dev,
+ libvulkan-dev,
 Standards-Version: 3.9.8
 Homepage: https://xemu.app
 XS-Debian-Vcs-Browser: https://github.com/mborgerson/xemu

debian/rules

@@ -63,7 +63,7 @@ override_dh_auto_configure:
 
 override_dh_auto_build:
 	./build.sh ${XEMU_BUILD_OPTIONS} ${common_configure_opts} || \
-	{ echo ===== BUILD FAILED ===; tail -n 50 config.log; exit 1; }
+	{ echo ===== BUILD FAILED ===; cat build/meson-logs/meson-log.txt; exit 1; }
 	cp debian/copyright debian/qemu.deb.copyright
 	cp dist/LICENSE.txt debian/copyright
 

hw/xbox/nv2a/debug.h

@@ -1,8 +1,9 @@
 /*
- * QEMU Geforce NV2A debug helpers
+ * QEMU Geforce NV2A profiling and debug helpers
  *
- * Copyright (c) 2015 Jannik Vogel
  * Copyright (c) 2012 espes
+ * Copyright (c) 2015 Jannik Vogel
+ * Copyright (c) 2018-2023 Matt Borgerson
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -18,8 +19,8 @@
  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  */
 
-#ifndef HW_NV2A_DEBUG_H
-#define HW_NV2A_DEBUG_H
+#ifndef HW_XBOX_NV2A_DEBUG_H
+#define HW_XBOX_NV2A_DEBUG_H
 
 #include <stdint.h>
 
@@ -36,54 +37,6 @@
 # define NV2A_DPRINTF(format, ...)       do { } while (0)
 #endif
 
-// #define DEBUG_NV2A_GL
-#ifdef DEBUG_NV2A_GL
-
-#include <stdbool.h>
-#include "gl/gloffscreen.h"
-#include "config-host.h"
-
-void gl_debug_initialize(void);
-void gl_debug_message(bool cc, const char *fmt, ...);
-void gl_debug_group_begin(const char *fmt, ...);
-void gl_debug_group_end(void);
-void gl_debug_label(GLenum target, GLuint name, const char *fmt, ...);
-void gl_debug_frame_terminator(void);
-
-# define NV2A_GL_DPRINTF(cc, format, ...) \
-    gl_debug_message(cc, "nv2a: " format, ## __VA_ARGS__)
-# define NV2A_GL_DGROUP_BEGIN(format, ...) \
-    gl_debug_group_begin("nv2a: " format, ## __VA_ARGS__)
-# define NV2A_GL_DGROUP_END() \
-    gl_debug_group_end()
-# define NV2A_GL_DLABEL(target, name, format, ...)  \
-    gl_debug_label(target, name, "nv2a: { " format " }", ## __VA_ARGS__)
-#define NV2A_GL_DFRAME_TERMINATOR() \
-    gl_debug_frame_terminator()
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-#ifdef CONFIG_RENDERDOC
-bool nv2a_dbg_renderdoc_available(void);
-void nv2a_dbg_renderdoc_capture_frames(uint32_t num_frames);
-#endif
-
-#ifdef __cplusplus
-}
-#endif
-
-#else
-# define NV2A_GL_DPRINTF(cc, format, ...)          do { \
-        if (cc) NV2A_DPRINTF(format "\n", ##__VA_ARGS__ ); \
-    } while (0)
-# define NV2A_GL_DGROUP_BEGIN(format, ...)         do { } while (0)
-# define NV2A_GL_DGROUP_END()                      do { } while (0)
-# define NV2A_GL_DLABEL(target, name, format, ...) do { } while (0)
-# define NV2A_GL_DFRAME_TERMINATOR()               do { } while (0)
-#endif
-
 /* Debug prints to identify when unimplemented or unconfirmed features
  * are being exercised. These cases likely result in graphical problems of
  * varying degree, but should otherwise not crash the system. Enable this
@@ -111,6 +64,22 @@ void nv2a_dbg_renderdoc_capture_frames(uint32_t num_frames);
 #endif
 
 #define NV2A_PROF_COUNTERS_XMAC \
+    _X(NV2A_PROF_FINISH_VERTEX_BUFFER_DIRTY) \
+    _X(NV2A_PROF_FINISH_SURFACE_CREATE) \
+    _X(NV2A_PROF_FINISH_SURFACE_DOWN) \
+    _X(NV2A_PROF_FINISH_NEED_BUFFER_SPACE) \
+    _X(NV2A_PROF_FINISH_FRAMEBUFFER_DIRTY) \
+    _X(NV2A_PROF_FINISH_PRESENTING) \
+    _X(NV2A_PROF_FINISH_FLIP_STALL) \
+    _X(NV2A_PROF_FINISH_FLUSH) \
+    _X(NV2A_PROF_FINISH_STALLED) \
+    _X(NV2A_PROF_CLEAR) \
+    _X(NV2A_PROF_QUEUE_SUBMIT) \
+    _X(NV2A_PROF_QUEUE_SUBMIT_AUX) \
+    _X(NV2A_PROF_PIPELINE_NOTDIRTY) \
+    _X(NV2A_PROF_PIPELINE_GEN) \
+    _X(NV2A_PROF_PIPELINE_BIND) \
+    _X(NV2A_PROF_PIPELINE_RENDERPASSES) \
     _X(NV2A_PROF_BEGIN_ENDS) \
     _X(NV2A_PROF_DRAW_ARRAYS) \
     _X(NV2A_PROF_INLINE_BUFFERS) \
@@ -120,18 +89,26 @@ void nv2a_dbg_renderdoc_capture_frames(uint32_t num_frames);
     _X(NV2A_PROF_SHADER_GEN) \
     _X(NV2A_PROF_SHADER_BIND) \
     _X(NV2A_PROF_SHADER_BIND_NOTDIRTY) \
+    _X(NV2A_PROF_SHADER_UBO_DIRTY) \
+    _X(NV2A_PROF_SHADER_UBO_NOTDIRTY) \
     _X(NV2A_PROF_ATTR_BIND) \
     _X(NV2A_PROF_TEX_UPLOAD) \
-    _X(NV2A_PROF_TEX_BIND) \
     _X(NV2A_PROF_GEOM_BUFFER_UPDATE_1) \
     _X(NV2A_PROF_GEOM_BUFFER_UPDATE_2) \
     _X(NV2A_PROF_GEOM_BUFFER_UPDATE_3) \
     _X(NV2A_PROF_GEOM_BUFFER_UPDATE_4) \
     _X(NV2A_PROF_GEOM_BUFFER_UPDATE_4_NOTDIRTY) \
+    _X(NV2A_PROF_SURF_SWIZZLE) \
+    _X(NV2A_PROF_SURF_CREATE) \
     _X(NV2A_PROF_SURF_DOWNLOAD) \
     _X(NV2A_PROF_SURF_UPLOAD) \
     _X(NV2A_PROF_SURF_TO_TEX) \
     _X(NV2A_PROF_SURF_TO_TEX_FALLBACK) \
+    _X(NV2A_PROF_QUEUE_SUBMIT_1) \
+    _X(NV2A_PROF_QUEUE_SUBMIT_2) \
+    _X(NV2A_PROF_QUEUE_SUBMIT_3) \
+    _X(NV2A_PROF_QUEUE_SUBMIT_4) \
+    _X(NV2A_PROF_QUEUE_SUBMIT_5) \
 
 enum NV2A_PROF_COUNTERS_ENUM {
     #define _X(x) x,
@@ -161,6 +138,21 @@ extern NV2AStats g_nv2a_stats;
 
 const char *nv2a_profile_get_counter_name(unsigned int cnt);
 int nv2a_profile_get_counter_value(unsigned int cnt);
+void nv2a_profile_increment(void);
+void nv2a_profile_flip_stall(void);
+
+static inline void nv2a_profile_inc_counter(enum NV2A_PROF_COUNTERS_ENUM cnt)
+{
+    g_nv2a_stats.frame_working.counters[cnt] += 1;
+}
+
+#ifdef CONFIG_RENDERDOC
+void nv2a_dbg_renderdoc_init(void);
+void *nv2a_dbg_renderdoc_get_api(void);
+bool nv2a_dbg_renderdoc_available(void);
+void nv2a_dbg_renderdoc_capture_frames(int num_frames);
+extern int renderdoc_capture_frames;
+#endif
 
 #ifdef __cplusplus
 }

hw/xbox/nv2a/meson.build

@@ -1,27 +1,17 @@
 specific_ss.add(files(
 	'nv2a.c',
-	'debug.c',
 	'pbus.c',
 	'pcrtc.c',
 	'pfb.c',
 	'pfifo.c',
-	'pgraph.c',
 	'pmc.c',
 	'pramdac.c',
 	'prmcio.c',
 	'prmdio.c',
 	'prmvio.c',
-	'psh.c',
 	'ptimer.c',
 	'pvideo.c',
-	'shaders.c',
 	'stubs.c',
 	'user.c',
-	'vsh.c',
-	'swizzle.c',
-	's3tc.c',
 	))
-subdir('gl')
-
-subdir('thirdparty')
-specific_ss.add(nv2a_vsh_cpu)
+subdir('pgraph')

hw/xbox/nv2a/nv2a.c

@@ -172,6 +172,16 @@ static void nv2a_get_offsets(VGACommonState *s,
     *pline_compare = line_compare;
 }
 
+const uint8_t *nv2a_get_dac_palette(void)
+{
+    return g_nv2a->puserdac.palette;
+}
+
+int nv2a_get_screen_off(void)
+{
+    return g_nv2a->vga.sr[VGA_SEQ_CLOCK_MODE] & VGA_SR01_SCREEN_OFF;
+}
+
 static void nv2a_vga_gfx_update(void *opaque)
 {
     VGACommonState *vga = opaque;
@@ -277,7 +287,7 @@ static void nv2a_reset(NV2AState *d)
     }
 
     memset(d->pfifo.regs, 0, sizeof(d->pfifo.regs));
-    memset(d->pgraph.regs, 0, sizeof(d->pgraph.regs));
+    memset(d->pgraph.regs_, 0, sizeof(d->pgraph.regs_));
     memset(d->pvideo.regs, 0, sizeof(d->pvideo.regs));
 
     d->pcrtc.start = 0;
@@ -365,11 +375,10 @@ static void nv2a_vm_state_change(void *opaque, bool running, RunState state)
     if (state == RUN_STATE_SAVE_VM) {
         nv2a_lock_fifo(d);
         qatomic_set(&d->pfifo.halt, true);
-        qatomic_set(&d->pgraph.download_dirty_surfaces_pending, true);
-        qemu_event_reset(&d->pgraph.dirty_surfaces_download_complete);
+        pgraph_pre_savevm_trigger(d);
         nv2a_unlock_fifo(d);
         qemu_mutex_unlock_iothread();
-        qemu_event_wait(&d->pgraph.dirty_surfaces_download_complete);
+        pgraph_pre_savevm_wait(d);
         qemu_mutex_lock_iothread();
         nv2a_lock_fifo(d);
     } else if (state == RUN_STATE_RESTORE_VM) {
@@ -382,11 +391,10 @@ static void nv2a_vm_state_change(void *opaque, bool running, RunState state)
         nv2a_unlock_fifo(d);
     } else if (state == RUN_STATE_SHUTDOWN) {
         nv2a_lock_fifo(d);
-        qatomic_set(&d->pgraph.shader_cache_writeback_pending, true);
-        qemu_event_reset(&d->pgraph.shader_cache_writeback_complete);
+        pgraph_pre_shutdown_trigger(d);
         nv2a_unlock_fifo(d);
         qemu_mutex_unlock_iothread();
-        qemu_event_wait(&d->pgraph.shader_cache_writeback_complete);
+        pgraph_pre_shutdown_wait(d);
         qemu_mutex_lock_iothread();
     }
 }
@@ -515,9 +523,9 @@ static const VMStateDescription vmstate_nv2a = {
         VMSTATE_UINT32(pgraph.inline_buffer_length, NV2AState), // fixme
         VMSTATE_UINT32(pgraph.draw_arrays_length, NV2AState),
         VMSTATE_UINT32(pgraph.draw_arrays_max_count, NV2AState),
-        VMSTATE_INT32_ARRAY(pgraph.gl_draw_arrays_start, NV2AState, 1250),
-        VMSTATE_INT32_ARRAY(pgraph.gl_draw_arrays_count, NV2AState, 1250),
-        VMSTATE_UINT32_ARRAY(pgraph.regs, NV2AState, 0x2000),
+        VMSTATE_INT32_ARRAY(pgraph.draw_arrays_start, NV2AState, 1250),
+        VMSTATE_INT32_ARRAY(pgraph.draw_arrays_count, NV2AState, 1250),
+        VMSTATE_UINT32_ARRAY(pgraph.regs_, NV2AState, 0x2000),
         VMSTATE_UINT32(pmc.pending_interrupts, NV2AState),
         VMSTATE_UINT32(pmc.enabled_interrupts, NV2AState),
         VMSTATE_UINT32(pfifo.pending_interrupts, NV2AState),

hw/xbox/nv2a/nv2a.h

@@ -22,8 +22,9 @@
 #define HW_NV2A_H
 
 void nv2a_init(PCIBus *bus, int devfn, MemoryRegion *ram);
-void nv2a_gl_context_init(void);
+void nv2a_context_init(void);
 int nv2a_get_framebuffer_surface(void);
+void nv2a_release_framebuffer_surface(void);
 void nv2a_set_surface_scale_factor(unsigned int scale);
 unsigned int nv2a_get_surface_scale_factor(void);
 const uint8_t *nv2a_get_dac_palette(void);