From e720e0a54059a0fb4b12a3993b2cbc63475d4006 Mon Sep 17 00:00:00 2001 From: Triang3l Date: Sun, 5 Sep 2021 21:03:05 +0300 Subject: [PATCH] [Code] Remove game names from code comments (most of at least) --- src/xenia/apu/xma_context.cc | 7 +- src/xenia/cpu/backend/x64/x64_seq_vector.cc | 4 +- src/xenia/cpu/ppc/ppc_emit_altivec.cc | 3 +- src/xenia/gpu/command_processor.cc | 4 +- .../gpu/d3d12/d3d12_command_processor.cc | 13 ++-- src/xenia/gpu/d3d12/d3d12_command_processor.h | 4 +- .../gpu/d3d12/d3d12_render_target_cache.cc | 8 +-- src/xenia/gpu/d3d12/pipeline_cache.cc | 7 +- src/xenia/gpu/d3d12/texture_cache.cc | 10 +-- src/xenia/gpu/d3d12/texture_cache.h | 2 +- src/xenia/gpu/draw_util.cc | 13 ++-- src/xenia/gpu/draw_util.h | 12 ++-- src/xenia/gpu/dxbc_shader_translator.cc | 32 ++++----- src/xenia/gpu/dxbc_shader_translator_fetch.cc | 19 +++-- src/xenia/gpu/dxbc_shader_translator_om.cc | 5 +- src/xenia/gpu/gpu_flags.cc | 12 ++-- src/xenia/gpu/primitive_processor.cc | 18 ++--- src/xenia/gpu/registers.h | 26 +++---- src/xenia/gpu/render_target_cache.cc | 24 +++---- src/xenia/gpu/render_target_cache.h | 16 ++--- src/xenia/gpu/shader.h | 12 ++-- src/xenia/gpu/shader_translator.cc | 4 +- .../gpu/shaders/adaptive_triangle.hs.hlsl | 2 +- src/xenia/gpu/shaders/pixel_formats.hlsli | 8 +-- .../gpu/shaders/tessellation_adaptive.vs.hlsl | 4 +- .../gpu/shaders/texture_load_ctx1.cs.hlsl | 2 +- src/xenia/gpu/shared_memory.cc | 7 +- src/xenia/gpu/texture_conversion.cc | 3 +- src/xenia/gpu/texture_util.cc | 10 +-- src/xenia/gpu/texture_util.h | 49 +++++++------ src/xenia/gpu/ucode.h | 16 ++--- src/xenia/gpu/vulkan/buffer_cache.cc | 2 +- src/xenia/gpu/vulkan/pipeline_cache.cc | 2 +- src/xenia/gpu/xenos.h | 72 +++++++++---------- src/xenia/kernel/xam/apps/xgi_app.cc | 2 +- src/xenia/kernel/xam/apps/xlivebase_app.cc | 2 +- src/xenia/kernel/xam/content_manager.h | 4 +- src/xenia/kernel/xam/user_profile.cc | 5 +- src/xenia/kernel/xam/xam_net.cc | 8 ++- src/xenia/kernel/xam/xam_notify.cc | 4 +- src/xenia/kernel/xam/xam_party.cc | 3 +- src/xenia/kernel/xam/xam_user.cc | 2 +- src/xenia/kernel/xbdm/xbdm_misc.cc | 2 +- .../kernel/xboxkrnl/xboxkrnl_audio_xma.cc | 2 +- src/xenia/kernel/xboxkrnl/xboxkrnl_debug.cc | 4 +- src/xenia/kernel/xboxkrnl/xboxkrnl_io.cc | 8 +-- src/xenia/kernel/xobject.h | 2 +- src/xenia/memory.cc | 11 ++- 48 files changed, 246 insertions(+), 245 deletions(-) diff --git a/src/xenia/apu/xma_context.cc b/src/xenia/apu/xma_context.cc index 9791cba85..a9662a0d8 100644 --- a/src/xenia/apu/xma_context.cc +++ b/src/xenia/apu/xma_context.cc @@ -811,12 +811,13 @@ void XmaContext::ConvertFrame(const uint8_t** samples, bool is_two_channel, // Loop through every sample, convert and drop it into the output array. // If more than one channel, we need to interleave the samples from each // channel next to each other. Always saturate because FFmpeg output is - // not limited to [-1, 1] (for example 1.095 as seen in RDR) + // not limited to [-1, 1] (for example 1.095 as seen in 5454082B). constexpr float scale = (1 << 15) - 1; auto out = reinterpret_cast(output_buffer); - // For testing of vectorized versions, stereo audio is common in Halo 3, since - // the first menu frame; the intro cutscene also has more than 2 channels. + // For testing of vectorized versions, stereo audio is common in 4D5307E6, + // since the first menu frame; the intro cutscene also has more than 2 + // channels. #if XE_ARCH_AMD64 static_assert(kSamplesPerFrame % 8 == 0); const auto in_channel_0 = reinterpret_cast(samples[0]); diff --git a/src/xenia/cpu/backend/x64/x64_seq_vector.cc b/src/xenia/cpu/backend/x64/x64_seq_vector.cc index 4c7fb665a..b5a8e49a2 100644 --- a/src/xenia/cpu/backend/x64/x64_seq_vector.cc +++ b/src/xenia/cpu/backend/x64/x64_seq_vector.cc @@ -1862,8 +1862,8 @@ struct PACK : Sequence> { src = i.src1; } // Saturate to [3,3....] so that only values between 3...[00] and 3...[FF] - // are valid - max before min to pack NaN as zero (Red Dead Redemption is - // heavily affected by the order - packs 0xFFFFFFFF in matrix code to get 0 + // are valid - max before min to pack NaN as zero (5454082B is heavily + // affected by the order - packs 0xFFFFFFFF in matrix code to get a 0 // constant). e.vmaxps(i.dest, src, e.GetXmmConstPtr(XMM3333)); e.vminps(i.dest, i.dest, e.GetXmmConstPtr(XMMPackD3DCOLORSat)); diff --git a/src/xenia/cpu/ppc/ppc_emit_altivec.cc b/src/xenia/cpu/ppc/ppc_emit_altivec.cc index 770def3c1..93f4df2b7 100644 --- a/src/xenia/cpu/ppc/ppc_emit_altivec.cc +++ b/src/xenia/cpu/ppc/ppc_emit_altivec.cc @@ -2069,7 +2069,8 @@ int InstrEmit_vpkd3d128(PPCHIRBuilder& f, const InstrData& i) { v = f.Pack(v, PACK_TYPE_FLOAT16_4); break; case 6: // VPACK_NORMPACKED64 4_20_20_20 w_z_y_x - // Used in 2K games like NBA 2K9, pretty rarely in general. + // Used in 54540829 and other installments in the series, pretty rarely in + // general. v = f.Pack(v, PACK_TYPE_ULONG_4202020); break; default: diff --git a/src/xenia/gpu/command_processor.cc b/src/xenia/gpu/command_processor.cc index 53af72115..2e1a43c36 100644 --- a/src/xenia/gpu/command_processor.cc +++ b/src/xenia/gpu/command_processor.cc @@ -738,7 +738,7 @@ bool CommandProcessor::ExecutePacketType3(RingBuffer* reader, uint32_t packet) { break; } case PM4_WAIT_FOR_IDLE: { - // This opcode is used by "Duke Nukem Forever" while going/being ingame + // This opcode is used by 5454084E while going / being ingame. assert_true(count == 1); uint32_t value = reader->ReadAndSwap(); XELOGGPU("GPU wait for idle = {:08X}", value); @@ -1168,7 +1168,7 @@ bool CommandProcessor::ExecutePacketType3_EVENT_WRITE_ZPD(RingBuffer* reader, // and used to detect a finished query. bool is_end_via_z_pass = pSampleCounts->ZPass_A == kQueryFinished && pSampleCounts->ZPass_B == kQueryFinished; - // Older versions of D3D also checks for ZFail (First Gears of War) + // Older versions of D3D also checks for ZFail (4D5307D5). bool is_end_via_z_fail = pSampleCounts->ZFail_A == kQueryFinished && pSampleCounts->ZFail_B == kQueryFinished; std::memset(pSampleCounts, 0, sizeof(xe_gpu_depth_sample_counts)); diff --git a/src/xenia/gpu/d3d12/d3d12_command_processor.cc b/src/xenia/gpu/d3d12/d3d12_command_processor.cc index fca476673..39bdfb664 100644 --- a/src/xenia/gpu/d3d12/d3d12_command_processor.cc +++ b/src/xenia/gpu/d3d12/d3d12_command_processor.cc @@ -1662,7 +1662,7 @@ void D3D12CommandProcessor::PerformSwap(uint32_t frontbuffer_ptr, gamma_ramp_upload_mapping_ + gamma_ramp_footprint.Offset); for (uint32_t i = 0; i < 256; ++i) { uint32_t value = gamma_ramp_.normal[i].value; - // Swap red and blue (Project Sylpheed has settings allowing separate + // Swap red and blue (535107D4 has settings allowing separate // configuration). mapping[i] = ((value & 1023) << 20) | (value & (1023 << 10)) | ((value >> 20) & 1023); @@ -2076,7 +2076,7 @@ bool D3D12CommandProcessor::IssueDraw(xenos::PrimitiveType primitive_type, memexport_stream.index_count * memexport_format_size; // Try to reduce the number of shared memory operations when writing // different elements into the same buffer through different exports - // (happens in Halo 3). + // (happens in 4D5307E6). bool memexport_range_reused = false; for (uint32_t i = 0; i < memexport_range_count; ++i) { MemExportRange& memexport_range = memexport_ranges[i]; @@ -2878,8 +2878,9 @@ void D3D12CommandProcessor::UpdateSystemConstantValues( // Get the color info register values for each render target. Also, for ROV, // exclude components that don't exist in the format from the write mask. // Don't exclude fully overlapping render targets, however - two render - // targets with the same base address are used in the lighting pass of Halo 3, - // for example, with the needed one picked with dynamic control flow. + // targets with the same base address are used in the lighting pass of + // 4D5307E6, for example, with the needed one picked with dynamic control + // flow. reg::RB_COLOR_INFO color_infos[4]; float rt_clamp[4][4]; uint32_t rt_keep_masks[4][2]; @@ -2898,8 +2899,8 @@ void D3D12CommandProcessor::UpdateSystemConstantValues( } // Disable depth and stencil if it aliases a color render target (for - // instance, during the XBLA logo in Banjo-Kazooie, though depth writing is - // already disabled there). + // instance, during the XBLA logo in 58410954, though depth writing is already + // disabled there). bool depth_stencil_enabled = rb_depthcontrol.stencil_enable || rb_depthcontrol.z_enable; if (edram_rov_used && depth_stencil_enabled) { diff --git a/src/xenia/gpu/d3d12/d3d12_command_processor.h b/src/xenia/gpu/d3d12/d3d12_command_processor.h index 018c568d0..e4d810c41 100644 --- a/src/xenia/gpu/d3d12/d3d12_command_processor.h +++ b/src/xenia/gpu/d3d12/d3d12_command_processor.h @@ -83,9 +83,9 @@ class D3D12CommandProcessor : public CommandProcessor { // Gets the current color write mask, taking the pixel shader's write mask // into account. If a shader doesn't write to a render target, it shouldn't be - // written to and it shouldn't be even bound - otherwise, in Halo 3, one + // written to and it shouldn't be even bound - otherwise, in 4D5307E6, one // render target is being destroyed by a shader not writing anything, and in - // Banjo-Tooie, the result of clearing the top tile is being ignored because + // 58410955, the result of clearing the top tile is being ignored because // there are 4 render targets bound with the same EDRAM base (clearly not // correct usage), but the shader only clears 1, and then EDRAM buffer stores // conflict with each other. diff --git a/src/xenia/gpu/d3d12/d3d12_render_target_cache.cc b/src/xenia/gpu/d3d12/d3d12_render_target_cache.cc index 53df32683..1437e696a 100644 --- a/src/xenia/gpu/d3d12/d3d12_render_target_cache.cc +++ b/src/xenia/gpu/d3d12/d3d12_render_target_cache.cc @@ -3619,7 +3619,7 @@ D3D12RenderTargetCache::GetOrCreateTransferPipelines(TransferShaderKey key) { case xenos::DepthRenderTargetFormat::kD24S8: // Round to the nearest even integer. This seems to be the correct, // adding +0.5 and rounding towards zero results in red instead of - // black in GTA IV and Halo 3 clear shaders. + // black in the 4D5307E6 clear shader. a.OpMul(dxbc::Dest::R(i, 0b1000), dxbc::Src::R(i, dxbc::Src::kWWWW), dxbc::Src::LF(float(0xFFFFFF))); a.OpRoundNE(dxbc::Dest::R(i, 0b1000), @@ -3804,7 +3804,7 @@ D3D12RenderTargetCache::GetOrCreateTransferPipelines(TransferShaderKey key) { case xenos::DepthRenderTargetFormat::kD24S8: // Round to the nearest even integer. This seems to be the correct, // adding +0.5 and rounding towards zero results in red instead of - // black in GTA IV and Halo 3 clear shaders. + // black in the 4D5307E6 clear shader. a.OpMul(dxbc::Dest::R(1, 0b1000), dxbc::Src::R(1, dxbc::Src::kWWWW), dxbc::Src::LF(float(0xFFFFFF))); a.OpRoundNE(dxbc::Dest::R(1, 0b1000), @@ -4181,7 +4181,7 @@ D3D12RenderTargetCache::GetOrCreateTransferPipelines(TransferShaderKey key) { case xenos::DepthRenderTargetFormat::kD24S8: // Round to the nearest even integer. This seems to be the // correct, adding +0.5 and rounding towards zero results in red - // instead of black in GTA IV and Halo 3 clear shaders. + // instead of black in the 4D5307E6 clear shader. a.OpMul(dxbc::Dest::R(0, 0b0010), dxbc::Src::R(0, dxbc::Src::kXXXX), dxbc::Src::LF(float(0xFFFFFF))); @@ -6228,7 +6228,7 @@ ID3D12PipelineState* D3D12RenderTargetCache::GetOrCreateDumpPipeline( case xenos::DepthRenderTargetFormat::kD24S8: // Round to the nearest even integer. This seems to be the correct, // adding +0.5 and rounding towards zero results in red instead of - // black in GTA IV and Halo 3 clear shaders. + // black in the 4D5307E6 clear shader. a.OpMul(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX), dxbc::Src::LF(float(0xFFFFFF))); a.OpRoundNE(dxbc::Dest::R(1, 0b0001), diff --git a/src/xenia/gpu/d3d12/pipeline_cache.cc b/src/xenia/gpu/d3d12/pipeline_cache.cc index 556c7cacf..879bc8eee 100644 --- a/src/xenia/gpu/d3d12/pipeline_cache.cc +++ b/src/xenia/gpu/d3d12/pipeline_cache.cc @@ -1567,7 +1567,8 @@ bool PipelineCache::GetCurrentStateDescription( /* 16 */ PipelineBlendFactor::kSrcAlphaSat, }; // Like kBlendFactorMap, but with color modes changed to alpha. Some - // pipelines aren't created in Prey because a color mode is used for alpha. + // pipelines aren't created in 545407E0 because a color mode is used for + // alpha. static const PipelineBlendFactor kBlendFactorAlphaMap[32] = { /* 0 */ PipelineBlendFactor::kZero, /* 1 */ PipelineBlendFactor::kOne, @@ -1599,7 +1600,7 @@ bool PipelineCache::GetCurrentStateDescription( // have their sample count matching the one set in the pipeline - however if // we set NumRenderTargets to 0 and also disable depth / stencil, the sample // count must be set to 1 - while the command list may still have - // multisampled render targets bound (happens in Halo 3 main menu). + // multisampled render targets bound (happens in 4D5307E6 main menu). // TODO(Triang3l): Investigate interaction of OMSetRenderTargets with // non-null depth and DSVFormat DXGI_FORMAT_UNKNOWN in the same case. for (uint32_t i = 0; i < 4; ++i) { @@ -2005,7 +2006,7 @@ ID3D12PipelineState* PipelineCache::CreateD3D12Pipeline( state_desc.BlendState.RenderTarget[i]; // Treat 1 * src + 0 * dest as disabled blending (there are opaque // surfaces drawn with blending enabled, but it's 1 * src + 0 * dest, in - // Call of Duty 4 - GPU performance is better when not blending. + // 415607E6 - GPU performance is better when not blending. if (rt.src_blend != PipelineBlendFactor::kOne || rt.dest_blend != PipelineBlendFactor::kZero || rt.blend_op != xenos::BlendOp::kAdd || diff --git a/src/xenia/gpu/d3d12/texture_cache.cc b/src/xenia/gpu/d3d12/texture_cache.cc index 17244c573..8f7fc0fb9 100644 --- a/src/xenia/gpu/d3d12/texture_cache.cc +++ b/src/xenia/gpu/d3d12/texture_cache.cc @@ -121,8 +121,8 @@ namespace shaders { // components of operands in shaders. // For DXT3A and DXT5A, RRRR swizzle is specified in: // http://fileadmin.cs.lth.se/cs/Personal/Michael_Doggett/talks/unc-xenos-doggett.pdf -// Halo 3 also expects replicated components in k_8 sprites. -// DXN is read as RG in Halo 3, but as RA in Call of Duty. +// 4D5307E6 also expects replicated components in k_8 sprites. +// DXN is read as RG in 4D5307E6, but as RA in 415607E6. // TODO(Triang3l): Find out the correct contents of unused texture components. const TextureCache::HostFormat TextureCache::host_formats_[64] = { // k_1_REVERSE @@ -250,9 +250,9 @@ const TextureCache::HostFormat TextureCache::host_formats_[64] = { LoadMode::kUnknown, {2, 1, 0, 3}}, // k_Y1_Cr_Y0_Cb_REP - // Used for videos in NBA 2K9. Red and blue must be swapped. + // Used for videos in 54540829. Red and blue must be swapped. // TODO(Triang3l): D3DFMT_G8R8_G8B8 is DXGI_FORMAT_R8G8_B8G8_UNORM * 255.0f, - // watch out for num_format int, division in shaders, etc., in NBA 2K9 it + // watch out for num_format int, division in shaders, etc., in 54540829 it // works as is. Also need to decompress if the size is uneven, but should be // a very rare case. {DXGI_FORMAT_R8G8_B8G8_UNORM, @@ -1309,7 +1309,7 @@ void TextureCache::RequestTextures(uint32_t used_texture_mask) { // Clear the bindings not only for this draw call, but entirely, because // loading may be needed in some draw call later, which may have the same // key for some binding as before the invalidation, but texture_invalidated_ - // being false (menu background in Halo 3). + // being false (menu background in 4D5307E6). for (size_t i = 0; i < xe::countof(texture_bindings_); ++i) { texture_bindings_[i].Clear(); } diff --git a/src/xenia/gpu/d3d12/texture_cache.h b/src/xenia/gpu/d3d12/texture_cache.h index 5c9c049c4..21a100ea2 100644 --- a/src/xenia/gpu/d3d12/texture_cache.h +++ b/src/xenia/gpu/d3d12/texture_cache.h @@ -418,7 +418,7 @@ class TextureCache { // Uncompression info for when the regular host format for this texture is // block-compressed, but the size is not block-aligned, and thus such // texture cannot be created in Direct3D on PC and needs decompression, - // however, such textures are common, for instance, in Halo 3. This only + // however, such textures are common, for instance, in 4D5307E6. This only // supports unsigned normalized formats - let's hope GPUSIGN_SIGNED was not // used for DXN and DXT5A. DXGI_FORMAT dxgi_format_uncompressed; diff --git a/src/xenia/gpu/draw_util.cc b/src/xenia/gpu/draw_util.cc index 49149b7f9..5157c1a88 100644 --- a/src/xenia/gpu/draw_util.cc +++ b/src/xenia/gpu/draw_util.cc @@ -24,12 +24,13 @@ #include "xenia/gpu/texture_util.h" #include "xenia/gpu/xenos.h" +// Very prominent in 545407F2. DEFINE_bool( resolve_resolution_scale_duplicate_second_pixel, true, "When using resolution scale, apply the hack that duplicates the " "right/lower host pixel in the left and top sides of render target resolve " "areas to eliminate the gap caused by half-pixel offset (this is necessary " - "for certain games like GTA IV to work).", + "for certain games to display the scene graphics).", "GPU"); DEFINE_bool( @@ -952,11 +953,11 @@ bool GetResolveInfo(const RegisterFile& regs, const Memory& memory, dest_dimension = xenos::DataDimension::k2DOrStacked; // RB_COPY_DEST_PITCH::copy_dest_height is the real texture height used // for 3D texture pitch, it's not relative to 0,0 of the coordinate space - // (in Halo 3, the sniper rifle scope has copy_dest_height of 192, but the - // rectangle's Y is 64...256) - provide the real height of the rectangle - // since 32x32 tiles are stored linearly anyway. In addition, the height - // in RB_COPY_DEST_PITCH may be larger than needed - in Red Dead - // Redemption, a UI texture for the letterbox bars alpha is located within + // (in 4D5307E6, the sniper rifle scope has copy_dest_height of 192, but + // the rectangle's Y is 64...256) - provide the real height of the + // rectangle since 32x32 tiles are stored linearly anyway. In addition, + // the height in RB_COPY_DEST_PITCH may be larger than needed - in + // 5454082B, a UI texture for the letterbox bars alpha is located within // the range of a 1280x720 resolve target, so with resolution scaling it's // also wrongly detected as scaled, while only 1280x208 is being resolved. dest_height = uint32_t(y1 - y0); diff --git a/src/xenia/gpu/draw_util.h b/src/xenia/gpu/draw_util.h index d23d0d25a..8aa7287f6 100644 --- a/src/xenia/gpu/draw_util.h +++ b/src/xenia/gpu/draw_util.h @@ -67,7 +67,7 @@ constexpr bool IsPrimitivePolygonal(bool vgt_output_path_is_tessellation_enable, // TODO(Triang3l): Investigate how kRectangleList should be treated - possibly // actually drawn as two polygons on the console, however, the current // geometry shader doesn't care about the winding order - allowing backface - // culling for rectangles currently breaks Gears of War 2. + // culling for rectangles currently breaks 4D53082D. return false; } @@ -112,10 +112,10 @@ constexpr float GetD3D10PolygonOffsetFactor( return float(1 << 24); } // 20 explicit + 1 implicit (1.) mantissa bits. - // 2^20 is not enough for Call of Duty 4 retail version's first mission F.N.G. - // shooting range floor (with the number 1) on Direct3D 12. Tested on Nvidia - // GeForce GTX 1070, the exact formula (taking into account the 0...1 to - // 0...0.5 remapping described below) used for testing is + // 2^20 is not enough for 415607E6 retail version's training mission shooting + // range floor (with the number 1) on Direct3D 12. Tested on Nvidia GeForce + // GTX 1070, the exact formula (taking into account the 0...1 to 0...0.5 + // remapping described below) used for testing is // `int(ceil(offset * 2^20 * 0.5)) * sign(offset)`. With 2^20 * 0.5, there // are various kinds of stripes dependending on the view angle in that // location. With 2^21 * 0.5, the issue is not present. @@ -141,7 +141,7 @@ inline bool DoesCoverageDependOnAlpha(reg::RB_COLORCONTROL rb_colorcontrol) { // pre-passes and shadowmaps. The shader must have its ucode analyzed. If // IsRasterizationPotentiallyDone, this shouldn't be called, and assumed false // instead. Helps reject the pixel shader in some cases - memexport draws in -// Halo 3, and also most of some 1-point draws not covering anything done for +// 4D5307E6, and also most of some 1-point draws not covering anything done for // some reason in different games with a leftover pixel shader from the previous // draw, but with SQ_PROGRAM_CNTL destroyed, reducing the number of // unpredictable unneeded translations of random shaders with different host diff --git a/src/xenia/gpu/dxbc_shader_translator.cc b/src/xenia/gpu/dxbc_shader_translator.cc index 30582b022..c66ad535f 100644 --- a/src/xenia/gpu/dxbc_shader_translator.cc +++ b/src/xenia/gpu/dxbc_shader_translator.cc @@ -23,11 +23,12 @@ #include "xenia/gpu/xenos.h" #include "xenia/ui/graphics_provider.h" +// The test case for AMD is 4D5307E6 (checked in 2018). DEFINE_bool(dxbc_switch, true, "Use switch rather than if for flow control. Turning this off or " "on may improve stability, though this heavily depends on the " "driver - on AMD, it's recommended to have this set to true, as " - "Halo 3 appears to crash when if is used for flow control " + "some titles appear to crash when if is used for flow control " "(possibly the shader compiler tries to flatten them). On Intel " "HD Graphics, this is ignored because of a crash with the switch " "instruction.", @@ -398,7 +399,7 @@ void DxbcShaderTranslator::StartVertexOrDomainShader() { assert_true(register_count() >= 2); if (register_count() >= 1) { // Copy the domain location to r0.xyz. - // ZYX swizzle according to Call of Duty 3 and Viva Pinata. + // ZYX swizzle according to 415607E1 and 4D5307F2. in_domain_location_used_ |= 0b0111; a_.OpMov(uses_register_dynamic_addressing ? dxbc::Dest::X(0, 0, 0b0111) : dxbc::Dest::R(0, 0b0111), @@ -425,7 +426,7 @@ void DxbcShaderTranslator::StartVertexOrDomainShader() { if (register_count() >= 1) { // Copy the domain location to r0.xyz. // ZYX swizzle with r1.y == 0, according to the water shader in - // Banjo-Kazooie: Nuts & Bolts. + // 4D5307ED. in_domain_location_used_ |= 0b0111; a_.OpMov(uses_register_dynamic_addressing ? dxbc::Dest::X(0, 0, 0b0111) : dxbc::Dest::R(0, 0b0111), @@ -447,10 +448,10 @@ void DxbcShaderTranslator::StartVertexOrDomainShader() { // appears that the tessellator offloads the reordering of coordinates // for edges to game shaders. // - // In Banjo-Kazooie: Nuts & Bolts, the water shader multiplies the - // first control point's position by r0.z, the second CP's by r0.y, - // and the third CP's by r0.x. But before doing that it swizzles - // r0.xyz the following way depending on the value in r1.y: + // In 4D5307ED, the water shader multiplies the first control point's + // position by r0.z, the second CP's by r0.y, and the third CP's by + // r0.x. But before doing that it swizzles r0.xyz the following way + // depending on the value in r1.y: // - ZXY for 1.0. // - YZX for 2.0. // - XZY for 4.0. @@ -478,9 +479,9 @@ void DxbcShaderTranslator::StartVertexOrDomainShader() { a_.OpMov(uses_register_dynamic_addressing ? dxbc::Dest::X(0, 0, 0b0011) : dxbc::Dest::R(0, 0b0011), dxbc::Src::VDomain()); - // Control point indices according to the shader from the main menu of - // Defender, which starts from `cndeq r2, c255.xxxy, r1.xyzz, r0.zzzz`, - // where c255.x is 0, and c255.y is 1. + // Control point indices according the main menu of 58410823, with + // `cndeq r2, c255.xxxy, r1.xyzz, r0.zzzz` in the prologue of the + // shader, where c255.x is 0, and c255.y is 1. // r0.z for (1 - r0.x) * (1 - r0.y) // r1.x for r0.x * (1 - r0.y) // r1.y for r0.x * r0.y @@ -509,7 +510,7 @@ void DxbcShaderTranslator::StartVertexOrDomainShader() { assert_true(register_count() >= 2); if (register_count() >= 1) { // Copy the domain location to r0.yz. - // XY swizzle according to the ground shader in Viva Pinata. + // XY swizzle according to the ground shader in 4D5307F2. in_domain_location_used_ |= 0b0011; a_.OpMov(uses_register_dynamic_addressing ? dxbc::Dest::X(0, 0, 0b0110) : dxbc::Dest::R(0, 0b0110), @@ -530,9 +531,8 @@ void DxbcShaderTranslator::StartVertexOrDomainShader() { // the tessellator offloads the reordering of coordinates for edges to // game shaders. // - // In Viva Pinata, if we assume that r0.y is U and r0.z is V, the - // factors each control point value is multiplied by are the - // following: + // In 4D5307F2, if we assume that r0.y is U and r0.z is V, the factors + // each control point value is multiplied by are the following: // - (1-u)*(1-v), u*(1-v), (1-u)*v, u*v for 0.0 (identity swizzle). // - u*(1-v), (1-u)*(1-v), u*v, (1-u)*v for 1.0 (YXWZ). // - u*v, (1-u)*v, u*(1-v), (1-u)*(1-v) for 2.0 (WZYX). @@ -1452,7 +1452,7 @@ void DxbcShaderTranslator::StoreResult(const InstructionResult& result, dest = dxbc::Dest::R(system_temp_point_size_edge_flag_kill_vertex_); break; case InstructionStorageTarget::kExportAddress: - // Validate memexport writes (Halo 3 has some weird invalid ones). + // Validate memexport writes (4D5307E6 has some completely invalid ones). if (!can_store_memexport_address || memexport_alloc_current_count_ == 0 || memexport_alloc_current_count_ > Shader::kMaxMemExports || system_temps_memexport_address_[memexport_alloc_current_count_ - 1] == @@ -1463,7 +1463,7 @@ void DxbcShaderTranslator::StoreResult(const InstructionResult& result, system_temps_memexport_address_[memexport_alloc_current_count_ - 1]); break; case InstructionStorageTarget::kExportData: { - // Validate memexport writes (Halo 3 has some weird invalid ones). + // Validate memexport writes (4D5307E6 has some completely invalid ones). if (memexport_alloc_current_count_ == 0 || memexport_alloc_current_count_ > Shader::kMaxMemExports || system_temps_memexport_data_[memexport_alloc_current_count_ - 1] diff --git a/src/xenia/gpu/dxbc_shader_translator_fetch.cc b/src/xenia/gpu/dxbc_shader_translator_fetch.cc index cb2d529da..4033239e0 100644 --- a/src/xenia/gpu/dxbc_shader_translator_fetch.cc +++ b/src/xenia/gpu/dxbc_shader_translator_fetch.cc @@ -705,10 +705,10 @@ void DxbcShaderTranslator::ProcessTextureFetchInstruction( // Add a small epsilon to the offset (1.5/4 the fixed-point texture // coordinate ULP - shouldn't significantly effect the fixed-point // conversion; 1/4 is also not enough with 3x resolution scaling very - // noticeably on the weapon in Halo 3) to resolve ambiguity when fetching + // noticeably on the weapon in 4D5307E6) to resolve ambiguity when fetching // point-sampled textures between texels. This applies to both normalized - // (Banjo-Kazooie Xbox Live Arcade logo, coordinates interpolated between - // vertices with half-pixel offset) and unnormalized (Halo 3 lighting + // (58410954 Xbox Live Arcade logo, coordinates interpolated between + // vertices with half-pixel offset) and unnormalized (4D5307E6 lighting // G-buffer reading, ps_param_gen pixels) coordinates. On Nvidia Pascal, // without this adjustment, blockiness is visible in both cases. Possibly // there is a better way, however, an attempt was made to error-correct @@ -1595,13 +1595,12 @@ void DxbcShaderTranslator::ProcessTextureFetchInstruction( // - Data. - // Viva Pinata uses vertex displacement map textures for tessellated - // models like the beehive tree with explicit LOD with point sampling - // (they store values packed in two components), however, the fetch - // constant has anisotropic filtering enabled. However, Direct3D 12 - // doesn't allow mixing anisotropic and point filtering. Possibly - // anistropic filtering should be disabled when explicit LOD is used - do - // this here. + // 4D5307F2 uses vertex displacement map textures for tessellated models + // like the beehive tree with explicit LOD with point sampling (they store + // values packed in two components), however, the fetch constant has + // anisotropic filtering enabled. However, Direct3D 12 doesn't allow + // mixing anisotropic and point filtering. Possibly anistropic filtering + // should be disabled when explicit LOD is used - do this here. uint32_t sampler_binding_index = FindOrAddSamplerBinding( tfetch_index, instr.attributes.mag_filter, instr.attributes.min_filter, instr.attributes.mip_filter, diff --git a/src/xenia/gpu/dxbc_shader_translator_om.cc b/src/xenia/gpu/dxbc_shader_translator_om.cc index 1002bf14d..a49500ec2 100644 --- a/src/xenia/gpu/dxbc_shader_translator_om.cc +++ b/src/xenia/gpu/dxbc_shader_translator_om.cc @@ -287,8 +287,7 @@ void DxbcShaderTranslator::StartPixelShader_LoadROVParameters() { dxbc::Src::R(system_temp_rov_params_, dxbc::Src::kYYYY)); // Choose in which 40-sample half of the tile the pixel is, for swapping // 40-sample columns when accessing the depth buffer - games expect this - // behavior when writing depth back to the EDRAM via color writing (GTA IV, - // Halo 3). + // behavior when writing depth back to the EDRAM via color writing (4D5307E6). // system_temp_rov_params_.x = tile-local sample 0 X >= 40 // system_temp_rov_params_.y = row offset // system_temp_rov_params_.z = X sample 0 position within the tile @@ -3282,7 +3281,7 @@ void DxbcShaderTranslator::ROV_DepthTo24Bit(uint32_t d24_temp, dxbc::Src::LF(float(0xFFFFFF))); // Round to the nearest even integer. This seems to be the correct way: // rounding towards zero gives 0xFF instead of 0x100 in clear shaders in, - // for instance, Halo 3, but other clear shaders in it are also broken if + // for instance, 4D5307E6, but other clear shaders in it are also broken if // 0.5 is added before ftou instead of round_ne. a_.OpRoundNE(d24_dest, d24_src); // Convert to fixed-point. diff --git a/src/xenia/gpu/gpu_flags.cc b/src/xenia/gpu/gpu_flags.cc index 5f73fd3c2..1510eeec0 100644 --- a/src/xenia/gpu/gpu_flags.cc +++ b/src/xenia/gpu/gpu_flags.cc @@ -28,16 +28,18 @@ DEFINE_bool( "the real reason why they're invalid is found.", "GPU"); +// Extremely bright screen borders in 4D5307E6. +// Reading between texels with half-pixel offset in 58410954. DEFINE_bool( half_pixel_offset, true, "Enable support of vertex half-pixel offset (D3D9 PA_SU_VTX_CNTL " "PIX_CENTER). Generally games are aware of the half-pixel offset, and " "having this enabled is the correct behavior (disabling this may " - "significantly break post-processing in some games, like Halo 3), but in " - "some games it might have been ignored, resulting in slight blurriness of " - "UI textures, for instance, when they are read between texels rather than " - "at texel centers (Banjo-Kazooie), or the leftmost/topmost pixels may not " - "be fully covered when MSAA is used with fullscreen passes.", + "significantly break post-processing in some games), but in certain games " + "it might have been ignored, resulting in slight blurriness of UI " + "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"); DEFINE_int32(query_occlusion_fake_sample_count, 1000, diff --git a/src/xenia/gpu/primitive_processor.cc b/src/xenia/gpu/primitive_processor.cc index b00e4ce50..12ba9cbe5 100644 --- a/src/xenia/gpu/primitive_processor.cc +++ b/src/xenia/gpu/primitive_processor.cc @@ -57,7 +57,7 @@ DEFINE_bool( // 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 -// (Tony Hawk's American Wasteland, 16-bit primitive reset index replacement). +// (415607D4, 16-bit primitive reset index replacement). DEFINE_int32( primitive_processor_cache_min_indices, 4096, "Smallest number of guest indices to store in the cache to try reusing " @@ -247,14 +247,14 @@ bool PrimitiveProcessor::Process(ProcessingResult& result_out) { // games using tessellated strips / fans so far. switch (tessellation_mode) { case xenos::TessellationMode::kDiscrete: - // - Call of Duty 3 - nets above barrels in the beginning of the - // first mission (turn right after the end of the intro) - + // - 415607E1 - nets above barrels in the beginning of the first + // mission (turn right after the end of the intro) - // kTriangleList. host_vertex_shader_type = Shader::HostVertexShaderType::kTriangleDomainCPIndexed; break; case xenos::TessellationMode::kContinuous: - // - Viva Pinata - tree building with a beehive in the beginning + // - 4D5307F2 - tree building with a beehive in the beginning // (visible on the start screen behind the logo), waterfall in the // beginning - kTriangleList. host_vertex_shader_type = @@ -276,7 +276,7 @@ bool PrimitiveProcessor::Process(ProcessingResult& result_out) { Shader::HostVertexShaderType::kQuadDomainCPIndexed; break; case xenos::TessellationMode::kContinuous: - // - Defender - retro screen and beams in the main menu - kQuadList. + // - 58410823 - retro screen and beams in the main menu - kQuadList. host_vertex_shader_type = Shader::HostVertexShaderType::kQuadDomainCPIndexed; break; @@ -285,14 +285,14 @@ bool PrimitiveProcessor::Process(ProcessingResult& result_out) { } break; case xenos::PrimitiveType::kTrianglePatch: - // - Banjo-Kazooie: Nuts & Bolts - water - adaptive. - // - Halo 3 - water - adaptive. + // - 4D5307E6 - water - adaptive. + // - 4D5307ED - water - adaptive. host_vertex_shader_type = Shader::HostVertexShaderType::kTriangleDomainPatchIndexed; break; case xenos::PrimitiveType::kQuadPatch: - // - Fable II - continuous. - // - Viva Pinata - garden ground - adaptive. + // - 4D5307F1 - continuous. + // - 4D5307F2 - garden ground - adaptive. host_vertex_shader_type = Shader::HostVertexShaderType::kQuadDomainPatchIndexed; break; diff --git a/src/xenia/gpu/registers.h b/src/xenia/gpu/registers.h index 2fc23b2c8..69d922d8b 100644 --- a/src/xenia/gpu/registers.h +++ b/src/xenia/gpu/registers.h @@ -335,10 +335,10 @@ union alignas(uint32_t) PA_SU_SC_MODE_CNTL { uint32_t cull_back : 1; // +1 // 0 - front is CCW, 1 - front is CW. uint32_t face : 1; // +2 - // The game Fuse uses poly_mode 2 for triangles, which is "reserved" on R6xx - // and not defined on Adreno 2xx, but polymode_front/back_ptype are 0 - // (points) in this case in Fuse, which should not be respected for - // non-kDualMode as the game wants to draw filled triangles. + // 4541096E uses poly_mode 2 for triangles, which is "reserved" on R6xx and + // not defined on Adreno 2xx, but polymode_front/back_ptype are 0 (points) + // in this case in 4541096E, which should not be respected for non-kDualMode + // as the title wants to draw filled triangles. xenos::PolygonModeEnable poly_mode : 2; // +3 xenos::PolygonType polymode_front_ptype : 3; // +5 xenos::PolygonType polymode_back_ptype : 3; // +8 @@ -559,16 +559,16 @@ union alignas(uint32_t) RB_COLORCONTROL { // (gl_FragCoord.y near 0 in the top, near 1 in the bottom here - D3D-like.) // For 2 samples, the top sample (closer to gl_FragCoord.y 0) is covered // when alpha is in [0.5, 1), the bottom sample is covered when the alpha is - // [1. With these thresholds, however, in Red Dead Redemption, almost all - // distant trees are transparent, this is asymmetric - fully transparent for - // a quarter of the range (or even half of the range for 2x and almost the - // entire range for 1x), but fully opaque only in one value. + // [1. With these thresholds, however, in 5454082B, almost all distant trees + // are transparent, this is asymmetric - fully transparent for a quarter of + // the range (or even half of the range for 2x and almost the entire range + // for 1x), but fully opaque only in one value. // Though, 2, 2, 2, 2 offset values are commonly used for undithered alpha - // to coverage (in games such as Red Dead Redemption, and overall in AMD - // driver implementations) - it appears that 2, 2, 2, 2 offsets are supposed - // to make this symmetric. - // Both Red Dead Redemption and RADV (which used AMDVLK as a reference) use - // 3, 1, 0, 2 offsets for dithered alpha to mask. + // to coverage (in games such as 5454082B, and overall in AMD driver + // implementations) - it appears that 2, 2, 2, 2 offsets are supposed to + // make this symmetric. + // Both 5454082B and RADV (which used AMDVLK as a reference) use 3, 1, 0, 2 + // offsets for dithered alpha to mask. // https://gitlab.freedesktop.org/nchery/mesa/commit/8a52e4cc4fad4f1c75acc0badd624778f9dfe202 // It appears that the offsets lower the thresholds by (offset / 4 / // sample count). That's consistent with both 2, 2, 2, 2 making the test diff --git a/src/xenia/gpu/render_target_cache.cc b/src/xenia/gpu/render_target_cache.cc index 76a3a2ef8..2dcf93789 100644 --- a/src/xenia/gpu/render_target_cache.cc +++ b/src/xenia/gpu/render_target_cache.cc @@ -40,6 +40,7 @@ DEFINE_bool( "reduce bandwidth usage during transfers as the previous depth won't need " "to be read.", "GPU"); +// The round trip is done, in particular, in 545407F2. DEFINE_string( depth_float24_conversion, "", "Method for converting 32-bit Z values to 20e4 floating point when using " @@ -56,8 +57,8 @@ DEFINE_string( " + Highest performance, allows early depth test and writing.\n" " + Host MSAA is possible with pixel-rate shading where supported.\n" " - EDRAM > RAM > EDRAM depth buffer round trip done in certain games " - "(such as GTA IV) destroys precision irreparably, causing artifacts if " - "another rendering pass is done after the EDRAM reupload.\n" + "destroys precision irreparably, causing artifacts if another rendering " + "pass is done after the EDRAM reupload.\n" " truncate:\n" " Convert to 20e4 directly in pixel shaders, always rounding down.\n" " + Average performance, conservative early depth test is possible.\n" @@ -96,18 +97,15 @@ DEFINE_bool( "bloom, etc., in some cases.", "GPU"); // Disabled by default because of full-screen effects that occur when game -// shaders assume piecewise linear, much more severe than blending-related -// issues. +// shaders assume piecewise linear (4541080F), much more severe than +// blending-related issues. DEFINE_bool( gamma_render_target_as_srgb, false, "When the host can't write piecewise linear gamma directly with correct " "blending, use sRGB output on the host for conceptually correct blending " - "in linear color space (to prevent issues such as bright squares around " - "bullet holes and overly dark lighting in Halo 3) while having slightly " - "different precision distribution in the render target and severely " - "incorrect values if the game accesses the resulting colors directly as " - "raw data (the whole screen in The Orange Box, for instance, since when " - "the first loading bar appears).", + "in linear color space while having slightly different precision " + "distribution in the render target and severely incorrect values if the " + "game accesses the resulting colors directly as raw data.", "GPU"); DEFINE_bool( mrt_edram_used_range_clamp_to_min, true, @@ -493,9 +491,9 @@ bool RenderTargetCache::Update(bool is_rasterization_done, // (issues caused by color and depth render target collisions haven't been // found yet), but render targets with smaller index are considered more // important - specifically, because of the usage in the lighting pass of - // Halo 3, which can be checked in the vertical look calibration sequence in + // 4D5307E6, which can be checked in the vertical look calibration sequence in // the beginning of the game: if render target 0 is removed in favor of 1, the - // UNSC servicemen and the world will be too dark, like fully in shadow - + // characters and the world will be too dark, like fully in shadow - // especially prominent on the helmet. This happens because the shader picks // between two render targets to write dynamically (though with a static, bool // constant condition), but all other state is set up in a way that implies @@ -624,7 +622,7 @@ bool RenderTargetCache::Update(bool is_rasterization_done, // "As if it was 64bpp" (contribution of 32bpp render targets multiplied by 2, // and clamping for 32bpp render targets divides this by 2) because 32bpp // render targets can be combined with twice as long 64bpp render targets. An - // example is the Dead Space 3 menu background (1-sample 1152x720, or 1200x720 + // example is the 4541099D menu background (1-sample 1152x720, or 1200x720 // after rounding to tiles, with a 32bpp depth buffer at 0 requiring 675 // tiles, and a 64bpp color buffer at 675 requiring 1350 tiles, but the // smallest distance between two render target bases is 675 tiles). diff --git a/src/xenia/gpu/render_target_cache.h b/src/xenia/gpu/render_target_cache.h index a3d580356..bf7c9a83e 100644 --- a/src/xenia/gpu/render_target_cache.h +++ b/src/xenia/gpu/render_target_cache.h @@ -70,10 +70,10 @@ class RenderTargetCache { // Significant differences: // - 8_8_8_8_GAMMA - the piecewise linear gamma curve is very different than // sRGB, one possible path is conversion in shaders (resulting in - // incorrect blending, especially visible on decals in Halo 3), another is - // using sRGB render targets and either conversion on resolve or reading - // the resolved data as a true sRGB texture (incorrect when the game - // accesses the data directly, like The Orange Box). + // incorrect blending, especially visible on decals in 4D5307E6), another + // is using sRGB render targets and either conversion on resolve or + // reading the resolved data as a true sRGB texture (incorrect when the + // game accesses the data directly, like 4541080F). // - 2_10_10_10_FLOAT - ranges significantly different than in float16, much // smaller RGB range, and alpha is fixed-point and has only 2 bits. // - 16_16, 16_16_16_16 - has -32 to 32 range, not -1 to 1 - need either to @@ -445,9 +445,9 @@ class RenderTargetCache { // aliasing naively, precision may be lost - host depth must only be // overwritten if the new guest value is different than the current host depth // when converted to the guest format (this catches the usual case of - // overwriting the depth buffer for clearing it mostly). Sonic the Hedgehog's - // intro cutscene, for example, has a good example of corruption that happens - // if this is not handled - the upper 1280x384 pixels are rendered in a very + // overwriting the depth buffer for clearing it mostly). 534507D6 intro + // cutscene, for example, has a good example of corruption that happens if + // this is not handled - the upper 1280x384 pixels are rendered in a very // "striped" way if the depth precision is lost (if this is made always return // false). virtual bool IsHostDepthEncodingDifferent( @@ -627,7 +627,7 @@ class RenderTargetCache { // surface info was changed), to avoid unneeded render target switching (which // is especially undesirable on tile-based GPUs) in the implementation if // simply disabling depth / stencil test or color writes and then re-enabling - // (Banjo-Kazooie does this often with color). Must also be used to determine + // (58410954 does this often with color). Must also be used to determine // whether it's safe to enable depth / stencil or writing to a specific color // render target in the pipeline for this draw call. // Only valid for non-pixel-shader-interlock paths. diff --git a/src/xenia/gpu/shader.h b/src/xenia/gpu/shader.h index f2f944155..f7c52cab0 100644 --- a/src/xenia/gpu/shader.h +++ b/src/xenia/gpu/shader.h @@ -551,7 +551,7 @@ struct ParsedAluInstruction { InstructionResult scalar_result; // Both operations must be executed before any result is stored if vector and // scalar operations are paired. There are cases of vector result being used - // as scalar operand or vice versa (the halo on Avalanche in Halo 3, for + // as scalar operand or vice versa (the ring on Avalanche in 4D5307E6, for // example), in this case there must be no dependency between the two // operations. @@ -851,11 +851,11 @@ class Shader { // highest static register address + 1, or 0 if no registers referenced this // way. SQ_PROGRAM_CNTL is not always reliable - some draws (like single point // draws with oPos = 0001 that are done by Xbox 360's Direct3D 9 sometimes; - // can be reproduced by launching Arrival in Halo 3 from the campaign lobby) - // that aren't supposed to cover any pixels use an invalid (zero) - // SQ_PROGRAM_CNTL, but with an outdated pixel shader loaded, in this case - // SQ_PROGRAM_CNTL may contain a number smaller than actually needed by the - // pixel shader - SQ_PROGRAM_CNTL should be used to go above this count if + // can be reproduced by launching the intro mission in 4D5307E6 from the + // campaign lobby) that aren't supposed to cover any pixels use an invalid + // (zero) SQ_PROGRAM_CNTL, but with an outdated pixel shader loaded, in this + // case SQ_PROGRAM_CNTL may contain a number smaller than actually needed by + // the pixel shader - SQ_PROGRAM_CNTL should be used to go above this count if // uses_register_dynamic_addressing is true. uint32_t register_static_address_bound() const { return register_static_address_bound_; diff --git a/src/xenia/gpu/shader_translator.cc b/src/xenia/gpu/shader_translator.cc index 8f6c61399..b1d1a060e 100644 --- a/src/xenia/gpu/shader_translator.cc +++ b/src/xenia/gpu/shader_translator.cc @@ -388,8 +388,8 @@ void Shader::GatherAluInstructionInformation( // allocation in shader translator implementations. // eA is (hopefully) always written to using: // mad eA, r#, const0100, c# - // (though there are some exceptions, shaders in Halo 3 for some reason set eA - // to zeros, but the swizzle of the constant is not .xyzw in this case, and + // (though there are some exceptions, shaders in 4D5307E6 for some reason set + // eA to zeros, but the swizzle of the constant is not .xyzw in this case, and // they don't write to eM#). // Export is done to vector_dest of the ucode instruction for both vector and // scalar operations - no need to check separately. diff --git a/src/xenia/gpu/shaders/adaptive_triangle.hs.hlsl b/src/xenia/gpu/shaders/adaptive_triangle.hs.hlsl index 31f7c36dd..c88563d79 100644 --- a/src/xenia/gpu/shaders/adaptive_triangle.hs.hlsl +++ b/src/xenia/gpu/shaders/adaptive_triangle.hs.hlsl @@ -36,7 +36,7 @@ XeHSConstantDataOutput XePatchConstant( // 2) r0.zyx -> r0.zyx by the guest (because r1.y is set to 0 by Xenia, which // apparently means identity swizzle to games). // 3) r0.z * v0 + r0.y * v1 + r0.x * v2 by the guest. - // With this order, there are no cracks in Halo 3 water. + // With this order, there are no cracks in 4D5307E6 water. [unroll] for (i = 0u; i < 3u; ++i) { output.edges[i] = xe_input_patch[(i + 1u) % 3u].edge_factor; } diff --git a/src/xenia/gpu/shaders/pixel_formats.hlsli b/src/xenia/gpu/shaders/pixel_formats.hlsli index 91bde0378..473a0620d 100644 --- a/src/xenia/gpu/shaders/pixel_formats.hlsli +++ b/src/xenia/gpu/shaders/pixel_formats.hlsli @@ -986,11 +986,11 @@ uint4 XeDXT3AAs1111TwoBlocksRowToBGRA4(uint2 halfblocks) { // DXT1/DXT3/DXT5 color components and CTX1 X/Y are ordered in: // http://fileadmin.cs.lth.se/cs/Personal/Michael_Doggett/talks/unc-xenos-doggett.pdf // (LSB on the right, MSB on the left.) - // TODO(Triang3l): Investigate this better, Halo: Reach is the only known game + // TODO(Triang3l): Investigate this better, 4D53085B is the only known game // that uses it (for lighting in certain places - one of easy to notice usages - // is the T-shaped (or somewhat H-shaped) metal beams in the beginning of - // Winter Contingency), however the contents don't say anything about the - // channel order. + // is the T-shaped (or somewhat H-shaped) metal beams in the beginning of the + // first mission), however the contents don't say anything about the channel + // order. uint4 row = (((halfblocks.xxyy >> uint2(3u, 11u).xyxy) & 1u) << 8u) | (((halfblocks.xxyy >> uint2(7u, 15u).xyxy) & 1u) << 24u) | (((halfblocks.xxyy >> uint2(2u, 10u).xyxy) & 1u) << 4u) | diff --git a/src/xenia/gpu/shaders/tessellation_adaptive.vs.hlsl b/src/xenia/gpu/shaders/tessellation_adaptive.vs.hlsl index fff6c501c..87ed1320a 100644 --- a/src/xenia/gpu/shaders/tessellation_adaptive.vs.hlsl +++ b/src/xenia/gpu/shaders/tessellation_adaptive.vs.hlsl @@ -5,8 +5,8 @@ XeHSControlPointInputAdaptive main(uint xe_edge_factor : SV_VertexID) { XeHSControlPointInputAdaptive output; // The Xbox 360's GPU accepts the float32 tessellation factors for edges // through a special kind of an index buffer. - // While Viva Pinata sets the factors to 0 for frustum-culled (quad) patches, - // in Halo 3 only allowing patches with factors above 0 makes distant + // While 4D5307F2 sets the factors to 0 for frustum-culled (quad) patches, in + // 4D5307E6 only allowing patches with factors above 0 makes distant // (triangle) patches disappear - it appears that there are no special values // for culled patches on the Xbox 360 (unlike zero, negative and NaN on // Direct3D 11). diff --git a/src/xenia/gpu/shaders/texture_load_ctx1.cs.hlsl b/src/xenia/gpu/shaders/texture_load_ctx1.cs.hlsl index eb908cc3a..6523cc92d 100644 --- a/src/xenia/gpu/shaders/texture_load_ctx1.cs.hlsl +++ b/src/xenia/gpu/shaders/texture_load_ctx1.cs.hlsl @@ -11,7 +11,7 @@ RWBuffer xe_texture_load_dest : register(u0); // Dword 1: // rrrrrrrrgggggggg // RRRRRRRRGGGGGGGG -// (R is in the higher bits, according to how this format is used in Halo 3). +// (R is in the higher bits, according to how this format is used in 4D5307E6). // Dword 2: // AA BB CC DD // EE FF GG HH diff --git a/src/xenia/gpu/shared_memory.cc b/src/xenia/gpu/shared_memory.cc index 0e47bd912..2b05821dc 100644 --- a/src/xenia/gpu/shared_memory.cc +++ b/src/xenia/gpu/shared_memory.cc @@ -465,9 +465,10 @@ std::pair SharedMemory::MemoryInvalidationCallback( // invalidated - if no GPU-written data nearby that was not intended to be // invalidated since it's not in sync with CPU memory and can't be // reuploaded. It's a lot cheaper to upload some excess data than to catch - // access violations - with 4 KB callbacks, the original Doom runs at 4 FPS - // on Intel Core i7-3770, with 64 KB the CPU game code takes 3 ms to run per - // frame, but with 256 KB it's 0.7 ms. + // access violations - with 4 KB callbacks, 58410824 (being a + // software-rendered game) runs at 4 FPS on Intel Core i7-3770, with 64 KB, + // the CPU game code takes 3 ms to run per frame, but with 256 KB, it's + // 0.7 ms. if (page_first & 63) { uint64_t gpu_written_start = system_page_flags_[block_first].valid_and_gpu_written; diff --git a/src/xenia/gpu/texture_conversion.cc b/src/xenia/gpu/texture_conversion.cc index bd028f47e..225e7feed 100644 --- a/src/xenia/gpu/texture_conversion.cc +++ b/src/xenia/gpu/texture_conversion.cc @@ -49,7 +49,8 @@ void CopySwapBlock(xenos::Endian endian, void* output, const void* input, void ConvertTexelCTX1ToR8G8(xenos::Endian endian, void* output, const void* input, size_t length) { // https://fileadmin.cs.lth.se/cs/Personal/Michael_Doggett/talks/unc-xenos-doggett.pdf - // (R is in the higher bits, according to how this format is used in Halo 3). + // (R is in the higher bits, according to how this format is used in + // 4D5307E6). union { uint8_t data[8]; struct { diff --git a/src/xenia/gpu/texture_util.cc b/src/xenia/gpu/texture_util.cc index bd7a78c6c..e64e4c205 100644 --- a/src/xenia/gpu/texture_util.cc +++ b/src/xenia/gpu/texture_util.cc @@ -352,11 +352,11 @@ TextureGuestLayout GetGuestTextureLayout( xenos::kTextureSubresourceAlignmentBytes); // Estimate the memory amount actually referenced by the texture, which may - // be smaller (especially in the 1280x720 linear k_8_8_8_8 case in Ridge - // Racer Unbounded, for which memory exactly for 1280x720 is allocated, and - // aligning the height to 32 would cause access of an unallocated page) or - // bigger than the stride. For tiled textures, this is the dimensions - // aligned to 32x32x4 blocks (or x1 for the missing dimensions). + // be smaller (especially in the 1280x720 linear k_8_8_8_8 case in 4E4D083E, + // for which memory exactly for 1280x720 is allocated, and aligning the + // height to 32 would cause access of an unallocated page) or bigger than + // the stride. For tiled textures, this is the dimensions aligned to 32x32x4 + // blocks (or x1 for the missing dimensions). uint32_t level_width_blocks = xe::align(std::max(width_texels >> level, uint32_t(1)), format_info->block_width) / diff --git a/src/xenia/gpu/texture_util.h b/src/xenia/gpu/texture_util.h index 3d9bc0e99..733b3e88f 100644 --- a/src/xenia/gpu/texture_util.h +++ b/src/xenia/gpu/texture_util.h @@ -64,14 +64,14 @@ bool GetPackedMipOffset(uint32_t width, uint32_t height, uint32_t depth, // implies 32-block alignment for both uncompressed and compressed textures) // stored in the fetch constant, and height aligned to 32 blocks for Z slice // and array layer stride calculation purposes. The pitch can be different -// from the actual width - an example is Plants vs. Zombies, using 1408 pitch -// for a 1280x menu background). +// from the actual width - an example is 584109FF, using 1408 pitch for a +// 1280x menu background). // - The mip levels use `max(next_pow2(width or height in texels) >> level, 1)` // aligned to 32 blocks for the same purpose, likely disregarding the pitch // from the fetch constant. // // There is also mip tail packing if the fetch constant specifies that packed -// mips are enabled, for both tiled and linear textures (Prey uses linear +// mips are enabled, for both tiled and linear textures (545407E0 uses linear // DXT-compressed textures with packed mips very extensively for the game world // materials). In this case, mips with width or height of 16 or smaller are // stored not individually, but instead, in 32-texel (note: not 32-block - mip @@ -99,7 +99,7 @@ bool GetPackedMipOffset(uint32_t width, uint32_t height, uint32_t depth, // tail, and the offset calculation function doesn't have level == 0 checks in // it, only early-out if level < packed tail level (which can be 0). There are // examples of textures with packed base, for example, in the intro level of -// Prey (8x8 linear DXT1 - pairs of orange lights in the bottom of gambling +// 545407E0 (8x8 linear DXT1 - pairs of orange lights in the bottom of gambling // machines). // // Linear texture rows are aligned to 256 bytes, for both the base and the mips @@ -107,22 +107,21 @@ bool GetPackedMipOffset(uint32_t width, uint32_t height, uint32_t depth, // fetch constant). // // However, all the 32x32x4 padding, being just padding, is not necessarily -// being actually accessed, especially for linear textures. Ridge Racer -// Unbounded has a 1280x720 k_8_8_8_8 linear texture, and allocates memory for -// exactly 1280x720, so aligning the height to 32 to 1280x736 results in access -// violations. So, while for stride calculations all the padding must be -// respected, for actual memory loads it's better to avoid trying to access it -// when possible: +// being actually accessed, especially for linear textures. 4E4D083E has a +// 1280x720 k_8_8_8_8 linear texture, and allocates memory for exactly 1280x720, +// so aligning the height to 32 to 1280x736 results in access violations. So, +// while for stride calculations all the padding must be respected, for actual +// memory loads it's better to avoid trying to access it when possible: // - If the pitch is bigger than the width, it's better to calculate the last // row's length from the width rather than the pitch (this also possibly works // in the other direction though - pitch < width is a weird situation, but // probably legal, and may lead to reading data from beyond the calculated // subresource stride). -// - For linear textures (like that 1280x720 example from Ridge Racer -// Unbounded), it's easy to calculate the exact memory extent that may be -// accessed knowing the dimensions (unlike for tiled textures with complex -// addressing within 32x32x4-block tiles), so there's no need to align them to -// 32x32x4 for memory extent calculation. +// - For linear textures (like that 1280x720 example from 4E4D083E), it's easy +// to calculate the exact memory extent that may be accessed knowing the +// dimensions (unlike for tiled textures with complex addressing within +// 32x32x4-block tiles), so there's no need to align them to 32x32x4 for +// memory extent calculation. // - For the linear packed mip tail, the extent can be calculated as max of // (block offsets + block extents) of all levels stored in it. // @@ -152,16 +151,16 @@ struct TextureGuestLayout { // tiled textures, this will be rounded to 32x32x4 blocks (or 32x32x1 // depending on the dimension), but for the linear subresources, this may be // significantly (including less 4 KB pages) smaller than the aligned size - // (like for Ridge Racer Unbounded where aligning the height of a 1280x720 - // linear texture results in access violations). For the linear mip tail, - // this includes all the mip levels stored in it. If the width is bigger - // than the pitch, this will also be taken into account for the last row so - // all memory actually used by the texture will be loaded, and may be bigger - // than the distance between array slices or levels. The purpose of this - // parameter is to make the memory amount that needs to be resident as close - // to the real amount as possible, to make sure all the needed data will be - // read, but also, if possible, unneeded memory pages won't be accessed - // (since that may trigger an access violation on the CPU). + // (like for 4E4D083E where aligning the height of a 1280x720 linear texture + // results in access violations). For the linear mip tail, this includes all + // the mip levels stored in it. If the width is bigger than the pitch, this + // will also be taken into account for the last row so all memory actually + // used by the texture will be loaded, and may be bigger than the distance + // between array slices or levels. The purpose of this parameter is to make + // the memory amount that needs to be resident as close to the real amount + // as possible, to make sure all the needed data will be read, but also, if + // possible, unneeded memory pages won't be accessed (since that may trigger + // an access violation on the CPU). uint32_t x_extent_blocks; uint32_t y_extent_blocks; uint32_t z_extent; diff --git a/src/xenia/gpu/ucode.h b/src/xenia/gpu/ucode.h index a4a169eb0..9d5ed7dca 100644 --- a/src/xenia/gpu/ucode.h +++ b/src/xenia/gpu/ucode.h @@ -483,7 +483,7 @@ enum class FetchOpcode : uint32_t { // - 3D (used for both 3D and stacked 2D texture): U, V, W (normalized or // unnormalized - same for both 3D W and stack layer; also VolMagFilter / // VolMinFilter between stack layers is supported, used for color correction - // in Burnout Revenge). + // in 454107DC). // - Cube: SC, TC (between 1 and 2 for normalized), face ID (0.0 to 5.0), the // cube vector ALU instruction is used to calculate them. // https://gpuopen.com/learn/fetching-from-cubes-and-octahedrons/ @@ -495,9 +495,9 @@ enum class FetchOpcode : uint32_t { // The total LOD for a sample is additive and is based on what is enabled. // // For cube maps, according to what texCUBEgrad compiles to in a modified - // HLSL shader of Brave: A Warrior's Tale and to XNA assembler output for PC - // SM3 texldd, register gradients are in cube space (not in SC/TC space, - // unlike the coordinates themselves). This isn't true for the GCN, however. + // HLSL shader of 455607D1 and to XNA assembler output for PC SM3 texldd, + // register gradients are in cube space (not in SC/TC space, unlike the + // coordinates themselves). This isn't true for the GCN, however. // // TODO(Triang3l): Find if gradients are unnormalized for cube maps if // coordinates are unnormalized. Since texldd doesn't perform any @@ -814,8 +814,8 @@ static_assert_size(TextureFetchInstruction, sizeof(uint32_t) * 3); // (mul, mad, dp, etc.) and for NaN in min/max. It's very important to respect // this rule for multiplication, as games often rely on it in vector // normalization (rcp and mul), Infinity * 0 resulting in NaN breaks a lot of -// things in games - causes white screen in Halo 3, white specular on -// characters in GTA IV. The result is always positive zero in this case, no +// things in games - causes white screen in 4D5307E6, white specular on +// characters in 545407F2. The result is always positive zero in this case, no // matter what the signs of the other operands are, according to R5xx // Acceleration section 8.7.5 "Legacy multiply behavior" and testing on // Adreno 200. This means that the following need to be taken into account @@ -1628,8 +1628,8 @@ enum class ExportRegister : uint32_t { // X - PSIZE (gl_PointSize). // Y - EDGEFLAG (glEdgeFlag) for PrimitiveType::kPolygon wireframe/point // drawing. - // Z - KILLVERTEX flag (used in Banjo-Kazooie: Nuts & Bolts for grass), set - // for killing primitives based on PA_CL_CLIP_CNTL::VTX_KILL_OR condition. + // Z - KILLVERTEX flag (used in 4D5307ED for grass), set for killing + // primitives based on PA_CL_CLIP_CNTL::VTX_KILL_OR condition. kVSPointSizeEdgeFlagKillVertex = 63, kPSColor0 = 0, diff --git a/src/xenia/gpu/vulkan/buffer_cache.cc b/src/xenia/gpu/vulkan/buffer_cache.cc index a02d9b54e..426fad033 100644 --- a/src/xenia/gpu/vulkan/buffer_cache.cc +++ b/src/xenia/gpu/vulkan/buffer_cache.cc @@ -507,7 +507,7 @@ std::pair BufferCache::UploadVertexBuffer( uint32_t upload_size = source_length; // Ping the memory subsystem for allocation size. - // TODO(DrChat): Artifacting occurring in GripShift with this enabled. + // TODO(DrChat): Artifacting occurring in 5841089E with this enabled. // physical_heap->QueryBaseAndSize(&upload_base, &upload_size); assert(upload_base <= source_addr); uint32_t source_offset = source_addr - upload_base; diff --git a/src/xenia/gpu/vulkan/pipeline_cache.cc b/src/xenia/gpu/vulkan/pipeline_cache.cc index 4defbcf04..128c5b133 100644 --- a/src/xenia/gpu/vulkan/pipeline_cache.cc +++ b/src/xenia/gpu/vulkan/pipeline_cache.cc @@ -758,7 +758,7 @@ bool PipelineCache::SetDynamicState(VkCommandBuffer command_buffer, depth_bias_scale = depth_bias_scales[1]; depth_bias_offset = depth_bias_offsets[1]; } - // Convert to Vulkan units based on the values in Call of Duty 4: + // Convert to Vulkan units based on the values in 415607E6: // r_polygonOffsetScale is -1 there, but 32 in the register. // r_polygonOffsetBias is -1 also, but passing 2/65536. // 1/65536 and 2 scales are applied separately, however, and for shadow maps diff --git a/src/xenia/gpu/xenos.h b/src/xenia/gpu/xenos.h index d3a202dd6..2a0b6c938 100644 --- a/src/xenia/gpu/xenos.h +++ b/src/xenia/gpu/xenos.h @@ -235,10 +235,10 @@ enum class SurfaceNumFormat : uint32_t { // // Depth surfaces are also stored as 32bpp tiles, however, as opposed to color // surfaces, 40x16-sample halves of each tile are swapped - game shaders (for -// example, in GTA IV, Halo 3) perform this swapping when writing specific -// depth/stencil values by drawing to a depth buffer's memory through a color -// render target (to reupload a depth/stencil surface previously evicted from -// the EDRAM to the main memory, for instance). +// example, in 4D5307E6 main menu, 545407F2) perform this swapping when writing +// specific depth/stencil values by drawing to a depth buffer's memory through a +// color render target (to reupload a depth/stencil surface previously evicted +// from the EDRAM to the main memory, for instance). enum class MsaaSamples : uint32_t { k1X = 0, @@ -728,12 +728,12 @@ enum class SampleControl : uint32_t { // - sample_control is SQ_CONTEXT_MISC::sc_sample_cntl. // - interpolator_control_sampling_pattern is // SQ_INTERPOLATOR_CNTL::sampling_pattern. -// Centroid interpolation can be tested in Red Dead Redemption. If the GPU host -// backend implements guest MSAA properly, using host MSAA, with everything -// interpolated at centers, the Diez Coronas start screen background may have -// a few distinctly bright pixels on the mesas/buttes, where extrapolation -// happens. Interpolating certain values (ones that aren't used for gradient -// calculation, not texture coordinates) at centroids fixes this issue. +// Centroid interpolation can be tested in 5454082B. If the GPU host backend +// implements guest MSAA properly, using host MSAA, with everything interpolated +// at centers, the Monument Valley start screen background may have a few +// distinctly bright pixels on the mesas/buttes, where extrapolation happens. +// Interpolating certain values (ones that aren't used for gradient calculation, +// not texture coordinates) at centroids fixes this issue. inline uint32_t GetInterpolatorSamplingPattern( MsaaSamples msaa_samples, SampleControl sample_control, uint32_t interpolator_control_sampling_pattern) { @@ -763,9 +763,9 @@ enum class TessellationMode : uint32_t { enum class PolygonModeEnable : uint32_t { kDisabled = 0, // Render triangles. kDualMode = 1, // Send 2 sets of 3 polygons with the specified polygon type. - // The game Fuse uses 2 for triangles, which is "reserved" on R6xx and not - // defined on Adreno 2xx, but polymode_front/back_ptype are 0 (points) in this - // case in Fuse, which should not be respected for non-kDualMode as the game + // 4541096E uses 2 for triangles, which is "reserved" on R6xx and not defined + // on Adreno 2xx, but polymode_front/back_ptype are 0 (points) in this case in + // 4541096E, which should not be respected for non-kDualMode as the title // wants to draw filled triangles. }; @@ -785,17 +785,15 @@ enum class ModeControl : uint32_t { // for it especially since the Xbox 360 doesn't have early per-sample depth / // stencil, only early hi-Z / hi-stencil, and other registers possibly // toggling pixel shader execution are yet to be found): - // - Most of depth pre-pass draws in Call of Duty 4 use the kDepth more with - // a `oC0 = tfetch2D(tf0, r0.xy) * r1` shader, some use `oC0 = r0` though. + // - Most of depth pre-pass draws in 415607E6 use the kDepth more with a + // `oC0 = tfetch2D(tf0, r0.xy) * r1` shader, some use `oC0 = r0` though. // However, when alphatested surfaces are drawn, kColorDepth is explicitly // used with the same shader performing the texture fetch. - // - Red Dead Redemption has some kDepth draws with alphatest enabled, but the - // shader is `oC0 = r0`, which makes no sense (alphatest based on an - // interpolant from the vertex shader) as no texture alpha cutout is - // involved. - // - Red Dead Redemption also has kDepth draws with pretty complex shaders - // clearly for use only in the color pass - even fetching and filtering a - // shadowmap. + // - 5454082B has some kDepth draws with alphatest enabled, but the shader is + // `oC0 = r0`, which makes no sense (alphatest based on an interpolant from + // the vertex shader) as no texture alpha cutout is involved. + // - 5454082B also has kDepth draws with pretty complex shaders clearly for + // use only in the color pass - even fetching and filtering a shadowmap. // For now, based on these, let's assume the pixel shader is never used with // kDepth. kDepth = 5, @@ -833,10 +831,10 @@ enum class ModeControl : uint32_t { // coordinates of the corners). // // The rectangle is used for both the source render target and the destination -// texture, according to how it's used in Tales of Vesperia. +// texture, according to how it's used in 4E4D07E9. // // Direct3D 9 gives the rectangle in source render target coordinates (for -// example, in Halo 3, the sniper rifle scope has a (128,64)->(448,256) +// example, in 4D5307E6, the sniper rifle scope has a (128,64)->(448,256) // rectangle). It doesn't adjust the EDRAM base pointer, otherwise (taking into // account that 4x MSAA is used for the scope) it would have been // (8,0)->(328,192), but it's not. However, it adjusts the destination texture @@ -851,7 +849,7 @@ enum class ModeControl : uint32_t { // RB_COPY_DEST_PITCH's purpose appears to be not clamping or something like // that, but just specifying pitch for going between rows, and height for going // between 3D texture slices. copy_dest_pitch is rounded to 32 by Direct3D 9, -// copy_dest_height is not. In the Halo 3 sniper rifle scope example, +// copy_dest_height is not. In the 4D5307E6 sniper rifle scope example, // copy_dest_pitch is 320, and copy_dest_height is 192 - the same as the resolve // rectangle size (resolving from a 320x192 portion of the surface at 128,64 to // the whole texture, at 0,0). Relative to RB_COPY_DEST_BASE, the height should @@ -860,17 +858,17 @@ enum class ModeControl : uint32_t { // of the register) that it exists purely to be able to go between 3D texture // slices. // -// Window scissor must also be applied - in the jigsaw puzzle in Banjo-Tooie, -// there are 1280x720 resolve rectangles, but only the scissored 1280x256 -// needs to be copied, otherwise it overflows even beyond the EDRAM, and the -// depth buffer is visible on the screen. It also ensures the coordinates are -// not negative (in F.E.A.R., for example, the right tile is resolved with -// vertices (-640,0)->(640,720), however, the destination texture pointer is -// adjusted properly to the right half of the texture, and the source render -// target has a pitch of 800). +// Window scissor must also be applied - in the jigsaw puzzle in 58410955, there +// are 1280x720 resolve rectangles, but only the scissored 1280x256 needs to be +// copied, otherwise it overflows even beyond the EDRAM, and the depth buffer is +// visible on the screen. It also ensures the coordinates are not negative (in +// 565507D9, for example, the right tile is resolved with vertices +// (-640,0)->(640,720), however, the destination texture pointer is adjusted +// properly to the right half of the texture, and the source render target has a +// pitch of 800). // Granularity of offset and size in resolve operations is 8x8 pixels -// (GPU_RESOLVE_ALIGNMENT - for example, Halo 3 resolves a 24x16 region for a +// (GPU_RESOLVE_ALIGNMENT - for example, 4D5307E6 resolves a 24x16 region for a // 18x10 texture, 8x8 region for a 1x1 texture). // https://github.com/jmfauvel/CSGO-SDK/blob/master/game/client/view.cpp#L944 // https://github.com/stanriders/hl2-asw-port/blob/master/src/game/client/vgui_int.cpp#L901 @@ -1072,9 +1070,9 @@ union alignas(uint32_t) xe_gpu_texture_fetch_t { // pitch is irrelevant to them (but the 256-byte alignment requirement still // applies to linear textures). // Examples of pitch > aligned width: - // - Plants vs. Zombies (loading screen and menu backgrounds, 1408 for a - // 1280x linear k_DXT4_5 texture, which corresponds to 22 * 256 bytes - // rather than 20 * 256 for just 1280x). + // - 584109FF (loading screen and menu backgrounds, 1408 for a 1280x linear + // k_DXT4_5 texture, which corresponds to 22 * 256 bytes rather than + // 20 * 256 for just 1280x). uint32_t pitch : 9; // +22 uint32_t tiled : 1; // +31 diff --git a/src/xenia/kernel/xam/apps/xgi_app.cc b/src/xenia/kernel/xam/apps/xgi_app.cc index 771341cdb..8fdf699f4 100644 --- a/src/xenia/kernel/xam/apps/xgi_app.cc +++ b/src/xenia/kernel/xam/apps/xgi_app.cc @@ -98,7 +98,7 @@ X_HRESULT XgiApp::DispatchMessageSync(uint32_t message, uint32_t buffer_ptr, return X_E_SUCCESS; } case 0x000B0014: { - // Gets Jetpac XBLA in game + // Gets 584107FB in game. // get high score table? XELOGD("XGI_unknown"); return X_STATUS_SUCCESS; diff --git a/src/xenia/kernel/xam/apps/xlivebase_app.cc b/src/xenia/kernel/xam/apps/xlivebase_app.cc index 45440fa55..57bfc4fd4 100644 --- a/src/xenia/kernel/xam/apps/xlivebase_app.cc +++ b/src/xenia/kernel/xam/apps/xlivebase_app.cc @@ -66,7 +66,7 @@ X_HRESULT XLiveBaseApp::DispatchMessageSync(uint32_t message, return X_E_FAIL; } case 0x00058046: { - // Required to be successful for Forza 4 to detect signed-in profile + // Required to be successful for 4D530910 to detect signed-in profile // Doesn't seem to set anything in the given buffer, probably only takes // input XELOGD("XLiveBaseUnk58046({:08X}, {:08X}) unimplemented", buffer_ptr, diff --git a/src/xenia/kernel/xam/content_manager.h b/src/xenia/kernel/xam/content_manager.h index 4876bfd1c..0db9f0cb5 100644 --- a/src/xenia/kernel/xam/content_manager.h +++ b/src/xenia/kernel/xam/content_manager.h @@ -73,8 +73,8 @@ struct XCONTENT_DATA { } void set_display_name(const std::u16string_view value) { - // Some games (eg Goldeneye XBLA) require multiple null-terminators for it - // to read the string properly, blanking the array should take care of that + // Some games (e.g. 584108A9) require multiple null-terminators for it to + // read the string properly, blanking the array should take care of that std::fill_n(display_name_raw.chars, countof(display_name_raw.chars), 0); string_util::copy_and_swap_truncating(display_name_raw.chars, value, diff --git a/src/xenia/kernel/xam/user_profile.cc b/src/xenia/kernel/xam/user_profile.cc index 4982a02c6..3bfa40f48 100644 --- a/src/xenia/kernel/xam/user_profile.cc +++ b/src/xenia/kernel/xam/user_profile.cc @@ -20,9 +20,8 @@ namespace kernel { namespace xam { UserProfile::UserProfile() { - // NeoGeo Battle Coliseum checks the user XUID against a mask of - // 0x00C0000000000000 (3<<54), if non-zero, it prevents the user from playing - // the game. + // 58410A1F checks the user XUID against a mask of 0x00C0000000000000 (3<<54), + // if non-zero, it prevents the user from playing the game. // "You do not have permissions to perform this operation." xuid_ = 0xB13EBABEBABEBABE; name_ = "User"; diff --git a/src/xenia/kernel/xam/xam_net.cc b/src/xenia/kernel/xam/xam_net.cc index dbdff9dc1..413639156 100644 --- a/src/xenia/kernel/xam/xam_net.cc +++ b/src/xenia/kernel/xam/xam_net.cc @@ -249,8 +249,10 @@ dword_result_t NetDll_WSAStartup(dword_t caller, word_t version, data_ptr->max_sockets = wsaData.iMaxSockets; data_ptr->max_udpdg = wsaData.iMaxUdpDg; - // Some games (PoG) want this value round-tripped - they'll compare if it - // changes and bugcheck if it does. + // Some games (5841099F) want this value round-tripped - they'll compare if + // it changes and bugcheck if it does. + // TODO(Triang3l): Verify if the title ID in the comment is correct - added + // by benvanik as an acronym initially. uint32_t vendor_ptr = xe::load_and_swap(data_out + 0x190); xe::store_and_swap(data_out + 0x190, vendor_ptr); } @@ -459,7 +461,7 @@ dword_result_t NetDll_XNetGetTitleXnAddr(dword_t caller, // TODO(gibbed): A proper mac address. // RakNet's 360 version appears to depend on abEnet to create "random" 64-bit // numbers. A zero value will cause RakPeer::Startup to fail. This causes - // Peggle 2 to crash on startup. + // 58411436 to crash on startup. // The 360-specific code is scrubbed from the RakNet repo, but there's still // traces of what it's doing which match the game code. // https://github.com/facebookarchive/RakNet/blob/master/Source/RakPeer.cpp#L382 diff --git a/src/xenia/kernel/xam/xam_notify.cc b/src/xenia/kernel/xam/xam_notify.cc index 32fb2635d..6d09df7f3 100644 --- a/src/xenia/kernel/xam/xam_notify.cc +++ b/src/xenia/kernel/xam/xam_notify.cc @@ -79,8 +79,8 @@ dword_result_t XNotifyGetNext(dword_t handle, dword_t match_id, } *id_ptr = dequeued ? id : 0; - // param_ptr may be null - Ghost Recon Advanced Warfighter 2 Demo explicitly - // passes nullptr in the code. + // param_ptr may be null - 555307F0 Demo explicitly passes nullptr in the + // code. // https://github.com/xenia-project/xenia/pull/1577 if (param_ptr) { *param_ptr = dequeued ? param : 0; diff --git a/src/xenia/kernel/xam/xam_party.cc b/src/xenia/kernel/xam/xam_party.cc index fd0556baa..5476f0b33 100644 --- a/src/xenia/kernel/xam/xam_party.cc +++ b/src/xenia/kernel/xam/xam_party.cc @@ -17,8 +17,7 @@ namespace kernel { namespace xam { dword_result_t XamPartyGetUserList(dword_t player_count, lpdword_t party_list) { - // Sonic & All-Stars Racing Transformed want specificly this code - // to skip loading party data. + // 5345085D wants specifically this code to skip loading party data. // This code is not documented in NT_STATUS code list return 0x807D0003; } diff --git a/src/xenia/kernel/xam/xam_user.cc b/src/xenia/kernel/xam/xam_user.cc index aed0c765b..612cc326c 100644 --- a/src/xenia/kernel/xam/xam_user.cc +++ b/src/xenia/kernel/xam/xam_user.cc @@ -716,7 +716,7 @@ dword_result_t XamUserCreateAchievementEnumerator(dword_t title_id, i, // dummy image id 0, {0, 0}, - 8}; // flags=8 makes dummy achievements show up in Crackdown's + 8}; // flags=8 makes dummy achievements show up in 4D5307DC // achievements list. e->AppendItem(item); } diff --git a/src/xenia/kernel/xbdm/xbdm_misc.cc b/src/xenia/kernel/xbdm/xbdm_misc.cc index c352b2434..e15af9eb6 100644 --- a/src/xenia/kernel/xbdm/xbdm_misc.cc +++ b/src/xenia/kernel/xbdm/xbdm_misc.cc @@ -34,7 +34,7 @@ DECLARE_XBDM_EXPORT1(DmCloseLoadedModules, kDebug, kStub); MAKE_DUMMY_STUB_STATUS(DmFreePool); dword_result_t DmGetXbeInfo() { - // TODO(gibbed): Crackdown appears to expect this as success? + // TODO(gibbed): 4D5307DC appears to expect this as success? // Unknown arguments -- let's hope things don't explode. return 0x02DA0000; } diff --git a/src/xenia/kernel/xboxkrnl/xboxkrnl_audio_xma.cc b/src/xenia/kernel/xboxkrnl/xboxkrnl_audio_xma.cc index 951732494..f4c2c156e 100644 --- a/src/xenia/kernel/xboxkrnl/xboxkrnl_audio_xma.cc +++ b/src/xenia/kernel/xboxkrnl/xboxkrnl_audio_xma.cc @@ -119,7 +119,7 @@ static_assert_size(XMA_CONTEXT_INIT, 56); dword_result_t XMAInitializeContext(lpvoid_t context_ptr, pointer_t context_init) { - // Input buffers may be null (buffer 1 in Tony Hawk's American Wasteland). + // Input buffers may be null (buffer 1 in 415607D4). // Convert to host endianness. uint32_t input_buffer_0_guest_ptr = context_init->input_buffer_0_ptr; uint32_t input_buffer_0_physical_address = 0; diff --git a/src/xenia/kernel/xboxkrnl/xboxkrnl_debug.cc b/src/xenia/kernel/xboxkrnl/xboxkrnl_debug.cc index a3e92c55f..5dd7b3309 100644 --- a/src/xenia/kernel/xboxkrnl/xboxkrnl_debug.cc +++ b/src/xenia/kernel/xboxkrnl/xboxkrnl_debug.cc @@ -47,8 +47,8 @@ void HandleSetThreadName(pointer_t record) { return; } - // Shadowrun (and its demo) has a bug where it ends up passing freed memory - // for the name, so at the point of SetThreadName it's filled with junk. + // 4D5307D6 (and its demo) has a bug where it ends up passing freed memory for + // the name, so at the point of SetThreadName it's filled with junk. // TODO(gibbed): cvar for thread name encoding for conversion, some games use // SJIS and there's no way to automatically know this. diff --git a/src/xenia/kernel/xboxkrnl/xboxkrnl_io.cc b/src/xenia/kernel/xboxkrnl/xboxkrnl_io.cc index f7dee19b9..37dc42aec 100644 --- a/src/xenia/kernel/xboxkrnl/xboxkrnl_io.cc +++ b/src/xenia/kernel/xboxkrnl/xboxkrnl_io.cc @@ -49,10 +49,10 @@ static bool IsValidPath(const std::string_view s, bool is_pattern) { if (got_asterisk) { // * must be followed by a . (*.) // - // Viva Piñata: Party Animals (4D530819) has a bug in its game code where - // it attempts to FindFirstFile() with filters of "Game:\\*_X3.rkv", - // "Game:\\m*_X3.rkv", and "Game:\\w*_X3.rkv" and will infinite loop if - // the path filter is allowed. + // 4D530819 has a bug in its game code where it attempts to + // FindFirstFile() with filters of "Game:\\*_X3.rkv", "Game:\\m*_X3.rkv", + // and "Game:\\w*_X3.rkv" and will infinite loop if the path filter is + // allowed. if (c != '.') { return false; } diff --git a/src/xenia/kernel/xobject.h b/src/xenia/kernel/xobject.h index 11d8ecb3c..67518b35c 100644 --- a/src/xenia/kernel/xobject.h +++ b/src/xenia/kernel/xobject.h @@ -111,7 +111,7 @@ struct X_OBJECT_TYPE { class XObject { public: - // Burnout Paradise needs proper handle value for certain calculations + // 45410806 needs proper handle value for certain calculations // It gets handle value from TLS (without base handle value is 0x88) // and substract 0xF8000088. Without base we're receiving wrong address // Instead of receiving address that starts with 0x82... we're receiving diff --git a/src/xenia/memory.cc b/src/xenia/memory.cc index 5b631e416..c7494e9fa 100644 --- a/src/xenia/memory.cc +++ b/src/xenia/memory.cc @@ -1531,12 +1531,11 @@ bool PhysicalHeap::Release(uint32_t base_address, uint32_t* out_region_size) { // Must invalidate here because the range being released may be reused in // another mapping of physical memory - but callback flags are set in each // heap separately (https://github.com/xenia-project/xenia/issues/1559 - - // dynamic vertices in Viva Pinata start screen and menu allocated in - // 0xA0000000 at addresses that overlap intro video textures in 0xE0000000, - // with the state of the allocator as of February 24th, 2020). If memory is - // invalidated in Alloc instead, Alloc won't be aware of callbacks enabled in - // other heaps, thus callback handlers will keep considering this range valid - // forever. + // dynamic vertices in 4D5307F2 start screen and menu allocated in 0xA0000000 + // at addresses that overlap intro video textures in 0xE0000000, with the + // state of the allocator as of February 24th, 2020). If memory is invalidated + // in Alloc instead, Alloc won't be aware of callbacks enabled in other heaps, + // thus callback handlers will keep considering this range valid forever. uint32_t region_size; if (QuerySize(base_address, ®ion_size)) { TriggerCallbacks(std::move(global_lock), base_address, region_size, true,