[GPU] Move debug cvars to a new TOML block, misc cleanup

native_2x_msaa is now debug_msaa_4x_as_2x, hopefully clearing up any user ambiguity.

src/xenia/gpu/command_processor.cc

@@ -32,18 +32,18 @@ DEFINE_bool(
     "Only does anything in debug builds, if set will log every write to a gpu "
     "register done by a guest. Does not log writes that are done by the CP on "
     "its own, just ones the guest makes or instructs it to make.",
-    "GPU");
+    "Logging");
 
 DEFINE_bool(disassemble_pm4, false,
             "Only does anything in debug builds, if set will disassemble and "
             "log all PM4 packets sent to the CP.",
-            "GPU");
+            "Logging");
 
 DEFINE_bool(
     log_ringbuffer_kickoff_initiator_bts, false,
     "Only does anything in debug builds, if set will log the pseudo-stacktrace "
     "of the guest thread that wrote the new read position.",
-    "GPU");
+    "Logging");
 
 DEFINE_bool(clear_memory_page_state, false,
             "Refresh state of memory pages to enable gpu written data. (Use "

src/xenia/gpu/d3d12/d3d12_render_target_cache.cc

@@ -36,7 +36,7 @@ DEFINE_bool(
     native_stencil_value_output_d3d12_intel, false,
     "Allow stencil reference output usage on Direct3D 12 on Intel GPUs - not "
     "working on UHD Graphics 630 as of March 2021 (driver 27.20.0100.8336).",
-    "GPU");
+    "GPU.Debug");
 // TODO(Triang3l): Make ROV the default when it's optimized better (for
 // instance, using static shader modifications to pass render target
 // parameters).
@@ -483,7 +483,7 @@ bool D3D12RenderTargetCache::Initialize() {
 
     // Check if 2x MSAA is supported or needs to be emulated with 4x MSAA
     // instead.
-    if (cvars::native_2x_msaa) {
+    if (!cvars::debug_msaa_4x_as_2x) {
       msaa_2x_supported_ = true;
       static constexpr DXGI_FORMAT kRenderTargetDXGIFormats[] = {
           DXGI_FORMAT_R16G16B16A16_FLOAT,

src/xenia/gpu/draw_extent_estimator.cc

@@ -34,7 +34,7 @@ DEFINE_bool(
     "this prevents corruption of other render targets located after the "
     "current ones in the EDRAM by lossy range ownership transfers done for "
     "those draws.",
-    "GPU");
+    "GPU.Debug");
 DEFINE_bool(
     execute_unclipped_draw_vs_on_cpu_with_scissor, false,
     "Don't restrict the usage of execute_unclipped_draw_vs_on_cpu to only "
@@ -45,7 +45,7 @@ DEFINE_bool(
     "Enabling this may cause excessive processing of vertices on the CPU, as "
     "some games draw rectangles (for their UI, for instance) without clipping, "
     "but with a proper scissor rectangle.",
-    "GPU");
+    "GPU.Debug");
 
 namespace xe {
 namespace gpu {

src/xenia/gpu/dxbc_shader_translator.cc

@@ -29,11 +29,11 @@ DEFINE_bool(dxbc_switch, true,
             "(possibly the shader compiler tries to flatten them). On Intel "
             "HD Graphics, this is ignored because of a crash with the switch "
             "instruction.",
-            "GPU");
+            "GPU.Debug");
 DEFINE_bool(dxbc_source_map, false,
             "Disassemble Xenos instructions as comments in the resulting DXBC "
             "for debugging.",
-            "GPU");
+            "GPU.Debug");
 
 namespace xe {
 namespace gpu {

src/xenia/gpu/gpu_flags.cc

@@ -10,13 +10,13 @@
 #include "xenia/gpu/gpu_flags.h"
 
 DEFINE_path(trace_gpu_prefix, "scratch/gpu/",
-            "Prefix path for GPU trace files.", "GPU");
-DEFINE_bool(trace_gpu_stream, false, "Trace all GPU packets.", "GPU");
+            "Prefix path for GPU trace files.", "GPU.Debug");
+DEFINE_bool(trace_gpu_stream, false, "Trace all GPU packets.", "GPU.Debug");
 
 DEFINE_path(
     dump_shaders, "",
     "For shader debugging, path to dump GPU shaders to as they are compiled.",
-    "GPU");
+    "GPU.Debug");
 
 DEFINE_bool(vsync, true, "Enable VSYNC.", "GPU");
 
@@ -43,7 +43,7 @@ DEFINE_bool(
     "Disable filtering between cube map faces near edges where possible "
     "(Vulkan with VK_EXT_non_seamless_cube_map) to reproduce the Direct3D 9 "
     "behavior.",
-    "GPU");
+    "GPU.Debug");
 
 // Extremely bright screen borders in 4D5307E6.
 // Reading between texels with half-pixel offset in 58410954.
@@ -57,7 +57,7 @@ DEFINE_bool(
     "textures, for instance, when they are read between texels rather than "
     "at texel centers, or the leftmost/topmost pixels may not be fully covered "
     "when MSAA is used with fullscreen passes.",
-    "GPU");
+    "GPU.Debug");
 
 DEFINE_int32(occlusion_query_fake_lower_threshold, 80,
              "Lower end of the fake sample count value written on "
@@ -104,7 +104,7 @@ DEFINE_int32(anisotropic_override, -1,
 DEFINE_bool(no_discard_stencil_in_transfer_pipelines, false,
             "Skip stencil bit discard in render target transfer pipelines. "
             "May improve performance on some GPUs.",
-            "GPU");
+            "GPU.Debug");
 
 DEFINE_bool(gpu_3d_to_2d_texture, true,
             "Handle shaders that sample 3D textures as 2D by creating a 2D "

src/xenia/gpu/graphics_system.cc

@@ -34,7 +34,7 @@ DEFINE_bool(
     store_shaders, true,
     "Store shaders persistently and load them when loading games to avoid "
     "runtime spikes and freezes when playing the game not for the first time.",
-    "GPU");
+    "GPU.Debug");
 
 namespace xe {
 namespace gpu {

src/xenia/gpu/primitive_processor.cc

@@ -30,13 +30,13 @@ DEFINE_bool(
     "For host graphics API downlevel support testing only, force CPU "
     "conversion of triangle fans to triangle lists even if the host supports "
     "triangle fan primitives natively.",
-    "GPU");
+    "GPU.Debug");
 DEFINE_bool(
     force_convert_line_loops_to_strips, false,
     "For host graphics API downlevel support testing only, force CPU "
     "conversion of line loops to line strips even if the host supports line "
     "loop primitives natively.",
-    "GPU");
+    "GPU.Debug");
 DEFINE_bool(
     force_convert_quad_lists_to_triangle_lists, false,
     "For host graphics API downlevel support testing only, force CPU "
@@ -44,14 +44,14 @@ DEFINE_bool(
     "list primitives natively or via geometry shader emulation.\n"
     "May also be useful for graphics debugging when the debugger doesn't "
     "display the geometry generated by geometry shaders properly.",
-    "GPU");
+    "GPU.Debug");
 DEFINE_bool(
     ignore_32bit_vertex_index_support, false,
     "For host graphics API downlevel testing only (useful only for Qualcomm "
     "Adreno 4xx-level host GPU testing), force indirection or pre-masking and "
     "pre-swapping of 32-bit vertex indices as if the host only supports 24-bit "
     "indices.",
-    "GPU");
+    "GPU.Debug");
 // TODO(Triang3l): More investigation of the cache threshold as cache lookups
 // and insertions require global critical region locking, and insertions also
 // require protecting pages. At 1024, the cache only made the performance worse
@@ -64,7 +64,7 @@ DEFINE_int32(
     "Setting this to a very high value may result in excessive CPU processing, "
     "while a very low value may result in excessive locking and lookups.\n"
     "Negative values disable caching.",
-    "GPU");
+    "GPU.Debug");
 
 namespace xe {
 namespace gpu {

src/xenia/gpu/render_target_cache.cc

@@ -21,6 +21,12 @@
 #include "xenia/gpu/registers.h"
 #include "xenia/gpu/xenos.h"
 
+DEFINE_bool(
+    debug_msaa_4x_as_2x, false,
+    "Use 4x MSAA with 2 samples instead of native 2x MSAA when available. "
+    "For scalability testing on host GPU APIs where 2x is not mandatory. MSAA "
+    "will be of a similar or worse quality and use more memory.",
+    "GPU.Debug");
 DEFINE_bool(
     depth_transfer_not_equal_test, true,
     "When transferring data between depth render targets, use the \"not "
@@ -33,7 +39,7 @@ DEFINE_bool(
     "beneficial to subsequent rendering, while setting this to false may "
     "reduce bandwidth usage during transfers as the previous depth won't need "
     "to be read.",
-    "GPU");
+    "GPU.Debug");
 // Lossless round trip: 545407F2.
 // Lossy round trip with the "greater or equal" test afterwards: 4D530919.
 // Lossy round trip with the "equal" test afterwards: 535107F5, 565507EF.
@@ -163,7 +169,7 @@ DEFINE_bool(
     "Greatly increases accuracy for this format, but may result in render "
     "target copying costs if the game switches between 8_8_8_8_GAMMA and "
     "8_8_8_8 views for the same EDRAM render target.",
-    "GPU");
+    "GPU.Debug");
 DEFINE_bool(
     mrt_edram_used_range_clamp_to_min, true,
     "With host render targets, if multiple render targets are bound, estimate "
@@ -173,14 +179,7 @@ DEFINE_bool(
     "Has effect primarily on draws without viewport clipping.\n"
     "Setting this to false results in higher accuracy in rare cases, but may "
     "increase the amount of copying that needs to be done sometimes.",
-    "GPU");
-DEFINE_bool(
-    native_2x_msaa, true,
-    "Use host 2x MSAA when available. Can be disabled for scalability testing "
-    "on host GPU APIs where 2x is not mandatory, in this case, 2 samples of 4x "
-    "MSAA will be used instead (with similar or worse quality and higher "
-    "memory usage).",
-    "GPU");
+    "GPU.Debug");
 DEFINE_bool(
     native_stencil_value_output, true,
     "Use pixel shader stencil reference output where available for purposes "
@@ -193,7 +192,7 @@ DEFINE_bool(
     "When the host can only support 16_16 and 16_16_16_16 render targets as "
     "-1...1, remap -32...32 to -1...1 to use the full possible range of "
     "values, at the expense of multiplicative blending correctness.",
-    "GPU");
+    "GPU.Debug");
 // Enabled by default as the GPU is overall usually the bottleneck when the
 // pixel shader interlock render backend implementation is used, anything that
 // may improve GPU performance is favorable.
@@ -206,7 +205,7 @@ DEFINE_bool(
     "needed when the ownership of a EDRAM range is changed.\n"
     "If this is enabled, excessive barriers may be eliminated when switching "
     "between different render targets in separate EDRAM locations.",
-    "GPU");
+    "GPU.Debug");
 
 namespace xe {
 namespace gpu {

src/xenia/gpu/render_target_cache.h

@@ -28,12 +28,12 @@
 #include "xenia/gpu/shader.h"
 #include "xenia/gpu/xenos.h"
 
+DECLARE_bool(debug_msaa_4x_as_2x);
 DECLARE_bool(depth_transfer_not_equal_test);
 DECLARE_bool(depth_float24_round);
 DECLARE_bool(depth_float24_convert_in_pixel_shader);
 DECLARE_bool(draw_resolution_scaled_texture_offsets);
 DECLARE_bool(gamma_render_target_as_unorm16);
-DECLARE_bool(native_2x_msaa);
 DECLARE_bool(native_stencil_value_output);
 DECLARE_bool(snorm16_render_target_full_range);
 

src/xenia/gpu/texture_cache.cc

@@ -40,17 +40,17 @@ DEFINE_uint32(
     texture_cache_memory_limit_soft, 384,
     "Maximum host texture memory usage (in megabytes) above which old textures "
     "will be destroyed.",
-    "GPU");
+    "GPU.Debug");
 DEFINE_uint32(
     texture_cache_memory_limit_soft_lifetime, 30,
     "Seconds a texture should be unused to be considered old enough to be "
     "deleted if texture memory usage exceeds texture_cache_memory_limit_soft.",
-    "GPU");
+    "GPU.Debug");
 DEFINE_uint32(
     texture_cache_memory_limit_hard, 768,
     "Maximum host texture memory usage (in megabytes) above which textures "
     "will be destroyed as soon as possible.",
-    "GPU");
+    "GPU.Debug");
 DEFINE_uint32(
     texture_cache_memory_limit_render_to_texture, 24,
     "Part of the host texture memory budget (in megabytes) that will be scaled "
@@ -60,11 +60,11 @@ DEFINE_uint32(
     "render-to-texture (resolve) targets and 384 - 24 = 360 MB of regular "
     "textures - so with 2x2 resolution scaling, the soft limit will be 360 + "
     "96 MB, and with 3x3, it will be 360 + 216 MB.",
-    "GPU");
+    "GPU.Debug");
 DEFINE_bool(tiled_shared_memory, true,
             "Enable tiled/sparse resources for efficient large address space "
             "support. Disable for graphics debugger compatibility.",
-            "GPU");
+            "GPU.Debug");
 
 namespace xe {
 namespace gpu {

src/xenia/gpu/trace_dump.cc

@@ -35,8 +35,9 @@
 #pragma clang diagnostic pop
 #endif
 
-DEFINE_path(target_trace_file, "", "Specifies the trace file to load.", "GPU");
-DEFINE_path(trace_dump_path, "", "Output path for dumped files.", "GPU");
+DEFINE_path(target_trace_file, "", "Specifies the trace file to load.",
+            "GPU.Debug");
+DEFINE_path(trace_dump_path, "", "Output path for dumped files.", "GPU.Debug");
 
 namespace xe {
 namespace gpu {

src/xenia/gpu/vulkan/vulkan_render_target_cache.cc

@@ -265,7 +265,7 @@ bool VulkanRenderTargetCache::Initialize(uint32_t shared_memory_binding_count) {
   // 2x MSAA support.
   // TODO(Triang3l): Handle sampledImageIntegerSampleCounts 4 not supported in
   // transfers.
-  if (cvars::native_2x_msaa) {
+  if (!cvars::debug_msaa_4x_as_2x) {
     // Multisampled integer sampled images are optional in Vulkan and in Xenia.
     msaa_2x_attachments_supported_ =
         (device_properties.framebufferColorSampleCounts &