Ryujinx/Ryujinx.Graphics/Graphics3d/NvGpuEngine3d.cs

1238 lines
46 KiB
C#
Raw Normal View History

using Ryujinx.Common;
using Ryujinx.Graphics.Gal;
using Ryujinx.Graphics.Memory;
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
using Ryujinx.Graphics.Shader;
using Ryujinx.Graphics.Texture;
using System;
using System.Collections.Generic;
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
using Ryujinx.Profiler;
namespace Ryujinx.Graphics.Graphics3d
{
class NvGpuEngine3d : INvGpuEngine
{
public int[] Registers { get; private set; }
private NvGpu _gpu;
private Dictionary<int, NvGpuMethod> _methods;
private struct ConstBuffer
{
public bool Enabled;
public long Position;
public int Size;
}
private ConstBuffer[][] _constBuffers;
// Viewport dimensions kept for scissor test limits
private int _viewportX0 = 0;
private int _viewportY0 = 0;
private int _viewportX1 = 0;
private int _viewportY1 = 0;
private int _viewportWidth = 0;
private int _viewportHeight = 0;
private int _currentInstance = 0;
2018-08-25 06:16:58 +02:00
public NvGpuEngine3d(NvGpu gpu)
{
_gpu = gpu;
Registers = new int[0xe00];
_methods = new Dictionary<int, NvGpuMethod>();
void AddMethod(int meth, int count, int stride, NvGpuMethod method)
{
while (count-- > 0)
{
_methods.Add(meth, method);
meth += stride;
}
}
AddMethod(0x585, 1, 1, VertexEndGl);
AddMethod(0x674, 1, 1, ClearBuffers);
AddMethod(0x6c3, 1, 1, QueryControl);
AddMethod(0x8e4, 16, 1, CbData);
AddMethod(0x904, 5, 8, CbBind);
_constBuffers = new ConstBuffer[6][];
for (int index = 0; index < _constBuffers.Length; index++)
{
_constBuffers[index] = new ConstBuffer[18];
}
//Ensure that all components are enabled by default.
//FIXME: Is this correct?
WriteRegister(NvGpuEngine3dReg.ColorMaskN, 0x1111);
WriteRegister(NvGpuEngine3dReg.FrameBufferSrgb, 1);
WriteRegister(NvGpuEngine3dReg.FrontFace, (int)GalFrontFace.Cw);
for (int index = 0; index < GalPipelineState.RenderTargetsCount; index++)
{
WriteRegister(NvGpuEngine3dReg.IBlendNEquationRgb + index * 8, (int)GalBlendEquation.FuncAdd);
WriteRegister(NvGpuEngine3dReg.IBlendNFuncSrcRgb + index * 8, (int)GalBlendFactor.One);
WriteRegister(NvGpuEngine3dReg.IBlendNFuncDstRgb + index * 8, (int)GalBlendFactor.Zero);
WriteRegister(NvGpuEngine3dReg.IBlendNEquationAlpha + index * 8, (int)GalBlendEquation.FuncAdd);
WriteRegister(NvGpuEngine3dReg.IBlendNFuncSrcAlpha + index * 8, (int)GalBlendFactor.One);
WriteRegister(NvGpuEngine3dReg.IBlendNFuncDstAlpha + index * 8, (int)GalBlendFactor.Zero);
}
}
public void CallMethod(NvGpuVmm vmm, GpuMethodCall methCall)
{
if (_methods.TryGetValue(methCall.Method, out NvGpuMethod method))
{
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
ProfileConfig profile = Profiles.GPU.Engine3d.CallMethod;
profile.SessionItem = method.Method.Name;
Profile.Begin(profile);
method(vmm, methCall);
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(profile);
}
else
{
WriteRegister(methCall);
}
}
private void VertexEndGl(NvGpuVmm vmm, GpuMethodCall methCall)
{
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.Begin(Profiles.GPU.Engine3d.VertexEnd);
LockCaches();
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.Begin(Profiles.GPU.Engine3d.ConfigureState);
GalPipelineState state = new GalPipelineState();
// Framebuffer must be run configured because viewport dimensions may be used in other methods
SetFrameBuffer(state);
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.ConfigureState);
for (int fbIndex = 0; fbIndex < 8; fbIndex++)
{
SetFrameBuffer(vmm, fbIndex);
}
SetFrontFace(state);
SetCullFace(state);
SetDepth(state);
SetStencil(state);
SetScissor(state);
SetBlending(state);
SetColorMask(state);
SetPrimitiveRestart(state);
SetZeta(vmm);
SetRenderTargets();
long[] keys = UploadShaders(vmm);
_gpu.Renderer.Shader.BindProgram();
UploadTextures(vmm, state, keys);
UploadConstBuffers(vmm, state, keys);
UploadVertexArrays(vmm, state);
DispatchRender(vmm, state);
UnlockCaches();
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.VertexEnd);
}
private void LockCaches()
{
_gpu.Renderer.Buffer.LockCache();
_gpu.Renderer.Rasterizer.LockCaches();
_gpu.Renderer.Texture.LockCache();
}
private void UnlockCaches()
{
_gpu.Renderer.Buffer.UnlockCache();
_gpu.Renderer.Rasterizer.UnlockCaches();
_gpu.Renderer.Texture.UnlockCache();
}
private void ClearBuffers(NvGpuVmm vmm, GpuMethodCall methCall)
{
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.Begin(Profiles.GPU.Engine3d.ClearBuffers);
int attachment = (methCall.Argument >> 6) & 0xf;
GalClearBufferFlags flags = (GalClearBufferFlags)(methCall.Argument & 0x3f);
float red = ReadRegisterFloat(NvGpuEngine3dReg.ClearNColor + 0);
float green = ReadRegisterFloat(NvGpuEngine3dReg.ClearNColor + 1);
float blue = ReadRegisterFloat(NvGpuEngine3dReg.ClearNColor + 2);
float alpha = ReadRegisterFloat(NvGpuEngine3dReg.ClearNColor + 3);
float depth = ReadRegisterFloat(NvGpuEngine3dReg.ClearDepth);
int stencil = ReadRegister(NvGpuEngine3dReg.ClearStencil);
SetFrameBuffer(vmm, attachment);
SetZeta(vmm);
SetRenderTargets();
_gpu.Renderer.RenderTarget.Bind();
_gpu.Renderer.Rasterizer.ClearBuffers(flags, attachment, red, green, blue, alpha, depth, stencil);
_gpu.Renderer.Pipeline.ResetDepthMask();
_gpu.Renderer.Pipeline.ResetColorMask(attachment);
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.ClearBuffers);
}
private void SetFrameBuffer(NvGpuVmm vmm, int fbIndex)
{
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
ProfileConfig profile = Profiles.GPU.Engine3d.SetFrameBuffer;
profile.SessionItem = fbIndex.ToString();
Profile.Begin(profile);
long va = MakeInt64From2xInt32(NvGpuEngine3dReg.FrameBufferNAddress + fbIndex * 0x10);
int surfFormat = ReadRegister(NvGpuEngine3dReg.FrameBufferNFormat + fbIndex * 0x10);
if (va == 0 || surfFormat == 0)
{
_gpu.Renderer.RenderTarget.UnbindColor(fbIndex);
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(profile);
return;
}
long key = vmm.GetPhysicalAddress(va);
int width = ReadRegister(NvGpuEngine3dReg.FrameBufferNWidth + fbIndex * 0x10);
int height = ReadRegister(NvGpuEngine3dReg.FrameBufferNHeight + fbIndex * 0x10);
int arrayMode = ReadRegister(NvGpuEngine3dReg.FrameBufferNArrayMode + fbIndex * 0x10);
int layerCount = arrayMode & 0xFFFF;
int layerStride = ReadRegister(NvGpuEngine3dReg.FrameBufferNLayerStride + fbIndex * 0x10);
int baseLayer = ReadRegister(NvGpuEngine3dReg.FrameBufferNBaseLayer + fbIndex * 0x10);
int blockDim = ReadRegister(NvGpuEngine3dReg.FrameBufferNBlockDim + fbIndex * 0x10);
int gobBlockHeight = 1 << ((blockDim >> 4) & 7);
GalMemoryLayout layout = (GalMemoryLayout)((blockDim >> 12) & 1);
float tx = ReadRegisterFloat(NvGpuEngine3dReg.ViewportNTranslateX + fbIndex * 8);
float ty = ReadRegisterFloat(NvGpuEngine3dReg.ViewportNTranslateY + fbIndex * 8);
float sx = ReadRegisterFloat(NvGpuEngine3dReg.ViewportNScaleX + fbIndex * 8);
float sy = ReadRegisterFloat(NvGpuEngine3dReg.ViewportNScaleY + fbIndex * 8);
_viewportX0 = (int)MathF.Max(0, tx - MathF.Abs(sx));
_viewportY0 = (int)MathF.Max(0, ty - MathF.Abs(sy));
_viewportX1 = (int)(tx + MathF.Abs(sx));
_viewportY1 = (int)(ty + MathF.Abs(sy));
GalImageFormat format = ImageUtils.ConvertSurface((GalSurfaceFormat)surfFormat);
GalImage image = new GalImage(width, height, 1, 1, 1, gobBlockHeight, 1, layout, format, GalTextureTarget.TwoD);
_gpu.ResourceManager.SendColorBuffer(vmm, key, fbIndex, image);
_gpu.Renderer.RenderTarget.SetViewport(fbIndex, _viewportX0, _viewportY0, _viewportX1 - _viewportX0, _viewportY1 - _viewportY0);
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(profile);
}
private void SetFrameBuffer(GalPipelineState state)
{
state.FramebufferSrgb = ReadRegisterBool(NvGpuEngine3dReg.FrameBufferSrgb);
state.FlipX = GetFlipSign(NvGpuEngine3dReg.ViewportNScaleX);
state.FlipY = GetFlipSign(NvGpuEngine3dReg.ViewportNScaleY);
int screenYControl = ReadRegister(NvGpuEngine3dReg.ScreenYControl);
bool negateY = (screenYControl & 1) != 0;
if (negateY)
{
state.FlipY = -state.FlipY;
}
}
private void SetZeta(NvGpuVmm vmm)
{
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.Begin(Profiles.GPU.Engine3d.SetZeta);
long va = MakeInt64From2xInt32(NvGpuEngine3dReg.ZetaAddress);
int zetaFormat = ReadRegister(NvGpuEngine3dReg.ZetaFormat);
int blockDim = ReadRegister(NvGpuEngine3dReg.ZetaBlockDimensions);
int gobBlockHeight = 1 << ((blockDim >> 4) & 7);
GalMemoryLayout layout = (GalMemoryLayout)((blockDim >> 12) & 1); //?
bool zetaEnable = ReadRegisterBool(NvGpuEngine3dReg.ZetaEnable);
if (va == 0 || zetaFormat == 0 || !zetaEnable)
{
_gpu.Renderer.RenderTarget.UnbindZeta();
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.SetZeta);
return;
}
long key = vmm.GetPhysicalAddress(va);
int width = ReadRegister(NvGpuEngine3dReg.ZetaHoriz);
int height = ReadRegister(NvGpuEngine3dReg.ZetaVert);
GalImageFormat format = ImageUtils.ConvertZeta((GalZetaFormat)zetaFormat);
Initial non 2D textures support (#525) * Initial non 2D textures support - Shaders still need to be changed - Some types aren't yet implemented * Start implementing texture instructions suffixes Fix wrong texture type with cube and TEXS Also support array textures in TEX and TEX.B Clean up TEX and TEXS coords managment Fix TEXS.LL with non-2d textures Implement TEX.AOFFI Get the right arguments for TEX, TEXS and TLDS Also, store suffix operands in appropriate values to support multiple suffix combinaisons * Support depth in read/writeTexture Also support WrapR and detect mipmap * Proper cube map textures support + fix TEXS.LZ * Implement depth compare * some code clean up * Implement CubeMap textures in OGLTexture.Create * Implement TLD4 and TLD4S * Add Texture 1D support * updates comments * fix some code style issues * Fix some nits + rename some things to be less confusing * Remove GetSuffix local functions * AOFFI => AOffI * TextureType => GalTextureTarget * finish renaming TextureType to TextureTarget * Disable LL, LZ and LB support in the decompiler This needs more work at the GL level (GLSL implementation should be right) * Revert "Disable LL, LZ and LB support in the decompiler" This reverts commit 64536c3d9f673645faff3152838d1413c3203395. * Fix TEXS ARRAY_2D index * ImageFormat depth should be 1 for all image format * Fix shader build issues with sampler1DShadow and texture * Fix DC & AOFFI combinaison with TEX/TEXS * Support AOFFI with TLD4 and TLD4S * Fix shader compilation error for TLD4.AOFFI with no DC * Fix binding isuses on the 2d copy engine TODO: support 2d array copy * Support 2D array copy operation in the 2D engine This make every copy right in the GPU side. Thie CPU copy probably needs to be updated * Implement GetGpuSize + fix somes issues with 2d engine copies TODO: mipmap level in it * Don't throw an exception in the layer handling * Fix because of rebase * Reject 2d layers of non textures in 2d copy engine * Add 3D textures and mipmap support on BlockLinearSwizzle * Fix naming on new BitUtils methods * gpu cache: Make sure to invalidate textures that doesn't have the same target * Add the concept of layer count for array instead of using depth Also cleanup GetGpuSize as Swizzle can compute the size with mipmap * Support multi layer with mip map in ReadTexture * Add more check for cache invalidation & remove cubemap and cubemap array code for now Also fix compressed 2d array * Fix texelFetchOffset shader build error * Start looking into cube map again Also add some way to log write in register in engines * fix write register log levles * Remove debug logs in WriteRegister * Disable AOFFI support on non NVIDIA drivers * Fix code align
2019-02-28 02:12:24 +01:00
// TODO: Support non 2D?
GalImage image = new GalImage(width, height, 1, 1, 1, gobBlockHeight, 1, layout, format, GalTextureTarget.TwoD);
_gpu.ResourceManager.SendZetaBuffer(vmm, key, image);
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.SetZeta);
}
private long[] UploadShaders(NvGpuVmm vmm)
{
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.Begin(Profiles.GPU.Engine3d.UploadShaders);
long[] keys = new long[5];
long basePosition = MakeInt64From2xInt32(NvGpuEngine3dReg.ShaderAddress);
int index = 1;
int vpAControl = ReadRegister(NvGpuEngine3dReg.ShaderNControl);
bool vpAEnable = (vpAControl & 1) != 0;
if (vpAEnable)
{
//Note: The maxwell supports 2 vertex programs, usually
//only VP B is used, but in some cases VP A is also used.
//In this case, it seems to function as an extra vertex
//shader stage.
//The graphics abstraction layer has a special overload for this
//case, which should merge the two shaders into one vertex shader.
int vpAOffset = ReadRegister(NvGpuEngine3dReg.ShaderNOffset);
int vpBOffset = ReadRegister(NvGpuEngine3dReg.ShaderNOffset + 0x10);
long vpAPos = basePosition + (uint)vpAOffset;
long vpBPos = basePosition + (uint)vpBOffset;
keys[(int)GalShaderType.Vertex] = vpBPos;
_gpu.Renderer.Shader.Create(vmm, vpAPos, vpBPos, GalShaderType.Vertex);
_gpu.Renderer.Shader.Bind(vpBPos);
index = 2;
}
for (; index < 6; index++)
{
GalShaderType type = GetTypeFromProgram(index);
int control = ReadRegister(NvGpuEngine3dReg.ShaderNControl + index * 0x10);
int offset = ReadRegister(NvGpuEngine3dReg.ShaderNOffset + index * 0x10);
//Note: Vertex Program (B) is always enabled.
bool enable = (control & 1) != 0 || index == 1;
if (!enable)
{
_gpu.Renderer.Shader.Unbind(type);
continue;
}
long key = basePosition + (uint)offset;
keys[(int)type] = key;
_gpu.Renderer.Shader.Create(vmm, key, type);
_gpu.Renderer.Shader.Bind(key);
}
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.UploadShaders);
return keys;
}
private static GalShaderType GetTypeFromProgram(int program)
{
switch (program)
{
case 0:
case 1: return GalShaderType.Vertex;
case 2: return GalShaderType.TessControl;
case 3: return GalShaderType.TessEvaluation;
case 4: return GalShaderType.Geometry;
case 5: return GalShaderType.Fragment;
}
throw new ArgumentOutOfRangeException(nameof(program));
}
private void SetFrontFace(GalPipelineState state)
{
float signX = GetFlipSign(NvGpuEngine3dReg.ViewportNScaleX);
float signY = GetFlipSign(NvGpuEngine3dReg.ViewportNScaleY);
GalFrontFace frontFace = (GalFrontFace)ReadRegister(NvGpuEngine3dReg.FrontFace);
//Flipping breaks facing. Flipping front facing too fixes it
if (signX != signY)
{
switch (frontFace)
{
case GalFrontFace.Cw: frontFace = GalFrontFace.Ccw; break;
case GalFrontFace.Ccw: frontFace = GalFrontFace.Cw; break;
}
}
state.FrontFace = frontFace;
}
private void SetCullFace(GalPipelineState state)
{
state.CullFaceEnabled = ReadRegisterBool(NvGpuEngine3dReg.CullFaceEnable);
if (state.CullFaceEnabled)
{
state.CullFace = (GalCullFace)ReadRegister(NvGpuEngine3dReg.CullFace);
}
}
private void SetDepth(GalPipelineState state)
{
state.DepthTestEnabled = ReadRegisterBool(NvGpuEngine3dReg.DepthTestEnable);
state.DepthWriteEnabled = ReadRegisterBool(NvGpuEngine3dReg.DepthWriteEnable);
if (state.DepthTestEnabled)
{
state.DepthFunc = (GalComparisonOp)ReadRegister(NvGpuEngine3dReg.DepthTestFunction);
}
state.DepthRangeNear = ReadRegisterFloat(NvGpuEngine3dReg.DepthRangeNNear);
state.DepthRangeFar = ReadRegisterFloat(NvGpuEngine3dReg.DepthRangeNFar);
}
private void SetStencil(GalPipelineState state)
{
state.StencilTestEnabled = ReadRegisterBool(NvGpuEngine3dReg.StencilEnable);
if (state.StencilTestEnabled)
{
state.StencilBackFuncFunc = (GalComparisonOp)ReadRegister(NvGpuEngine3dReg.StencilBackFuncFunc);
state.StencilBackFuncRef = ReadRegister(NvGpuEngine3dReg.StencilBackFuncRef);
state.StencilBackFuncMask = (uint)ReadRegister(NvGpuEngine3dReg.StencilBackFuncMask);
state.StencilBackOpFail = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilBackOpFail);
state.StencilBackOpZFail = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilBackOpZFail);
state.StencilBackOpZPass = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilBackOpZPass);
state.StencilBackMask = (uint)ReadRegister(NvGpuEngine3dReg.StencilBackMask);
state.StencilFrontFuncFunc = (GalComparisonOp)ReadRegister(NvGpuEngine3dReg.StencilFrontFuncFunc);
state.StencilFrontFuncRef = ReadRegister(NvGpuEngine3dReg.StencilFrontFuncRef);
state.StencilFrontFuncMask = (uint)ReadRegister(NvGpuEngine3dReg.StencilFrontFuncMask);
state.StencilFrontOpFail = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilFrontOpFail);
state.StencilFrontOpZFail = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilFrontOpZFail);
state.StencilFrontOpZPass = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilFrontOpZPass);
state.StencilFrontMask = (uint)ReadRegister(NvGpuEngine3dReg.StencilFrontMask);
}
}
private void SetScissor(GalPipelineState state)
{
int count = 0;
for (int index = 0; index < GalPipelineState.RenderTargetsCount; index++)
{
state.ScissorTestEnabled[index] = ReadRegisterBool(NvGpuEngine3dReg.ScissorEnable + index * 4);
if (state.ScissorTestEnabled[index])
{
uint scissorHorizontal = (uint)ReadRegister(NvGpuEngine3dReg.ScissorHorizontal + index * 4);
uint scissorVertical = (uint)ReadRegister(NvGpuEngine3dReg.ScissorVertical + index * 4);
int left = (int)(scissorHorizontal & 0xFFFF); // Left, lower 16 bits
int right = (int)(scissorHorizontal >> 16); // Right, upper 16 bits
int bottom = (int)(scissorVertical & 0xFFFF); // Bottom, lower 16 bits
int top = (int)(scissorVertical >> 16); // Top, upper 16 bits
int width = Math.Abs(right - left);
int height = Math.Abs(top - bottom);
// If the scissor test covers the whole possible viewport, i.e. uninitialized, disable scissor test
if ((width > NvGpu.MaxViewportSize && height > NvGpu.MaxViewportSize) || width <= 0 || height <= 0)
{
state.ScissorTestEnabled[index] = false;
continue;
}
// Keep track of how many scissor tests are active.
// If only 1, and it's the first user should apply to all viewports
count++;
// Flip X
if (state.FlipX == -1)
{
left = _viewportX1 - (left - _viewportX0);
right = _viewportX1 - (right - _viewportX0);
}
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
// Ensure X is in the right order
if (left > right)
{
int temp = left;
left = right;
right = temp;
}
// Flip Y
if (state.FlipY == -1)
{
bottom = _viewportY1 - (bottom - _viewportY0);
top = _viewportY1 - (top - _viewportY0);
}
// Ensure Y is in the right order
if (bottom > top)
{
int temp = top;
top = bottom;
bottom = temp;
}
// Handle out of active viewport dimensions
left = Math.Clamp(left, _viewportX0, _viewportX1);
right = Math.Clamp(right, _viewportX0, _viewportX1);
top = Math.Clamp(top, _viewportY0, _viewportY1);
bottom = Math.Clamp(bottom, _viewportY0, _viewportY1);
// Save values to state
state.ScissorTestX[index] = left;
state.ScissorTestY[index] = bottom;
state.ScissorTestWidth[index] = right - left;
state.ScissorTestHeight[index] = top - bottom;
}
}
state.ScissorTestCount = count;
}
private void SetBlending(GalPipelineState state)
{
bool blendIndependent = ReadRegisterBool(NvGpuEngine3dReg.BlendIndependent);
state.BlendIndependent = blendIndependent;
for (int index = 0; index < GalPipelineState.RenderTargetsCount; index++)
{
if (blendIndependent)
{
state.Blends[index].Enabled = ReadRegisterBool(NvGpuEngine3dReg.IBlendNEnable + index);
if (state.Blends[index].Enabled)
{
state.Blends[index].SeparateAlpha = ReadRegisterBool(NvGpuEngine3dReg.IBlendNSeparateAlpha + index * 8);
state.Blends[index].EquationRgb = ReadBlendEquation(NvGpuEngine3dReg.IBlendNEquationRgb + index * 8);
state.Blends[index].FuncSrcRgb = ReadBlendFactor (NvGpuEngine3dReg.IBlendNFuncSrcRgb + index * 8);
state.Blends[index].FuncDstRgb = ReadBlendFactor (NvGpuEngine3dReg.IBlendNFuncDstRgb + index * 8);
state.Blends[index].EquationAlpha = ReadBlendEquation(NvGpuEngine3dReg.IBlendNEquationAlpha + index * 8);
state.Blends[index].FuncSrcAlpha = ReadBlendFactor (NvGpuEngine3dReg.IBlendNFuncSrcAlpha + index * 8);
state.Blends[index].FuncDstAlpha = ReadBlendFactor (NvGpuEngine3dReg.IBlendNFuncDstAlpha + index * 8);
}
}
else
{
//It seems that even when independent blend is disabled, the first IBlend enable
//register is still set to indicate whenever blend is enabled or not (?).
state.Blends[index].Enabled = ReadRegisterBool(NvGpuEngine3dReg.IBlendNEnable);
if (state.Blends[index].Enabled)
{
state.Blends[index].SeparateAlpha = ReadRegisterBool(NvGpuEngine3dReg.BlendSeparateAlpha);
state.Blends[index].EquationRgb = ReadBlendEquation(NvGpuEngine3dReg.BlendEquationRgb);
state.Blends[index].FuncSrcRgb = ReadBlendFactor (NvGpuEngine3dReg.BlendFuncSrcRgb);
state.Blends[index].FuncDstRgb = ReadBlendFactor (NvGpuEngine3dReg.BlendFuncDstRgb);
state.Blends[index].EquationAlpha = ReadBlendEquation(NvGpuEngine3dReg.BlendEquationAlpha);
state.Blends[index].FuncSrcAlpha = ReadBlendFactor (NvGpuEngine3dReg.BlendFuncSrcAlpha);
state.Blends[index].FuncDstAlpha = ReadBlendFactor (NvGpuEngine3dReg.BlendFuncDstAlpha);
}
}
}
}
private GalBlendEquation ReadBlendEquation(NvGpuEngine3dReg register)
{
return (GalBlendEquation)ReadRegister(register);
}
private GalBlendFactor ReadBlendFactor(NvGpuEngine3dReg register)
{
return (GalBlendFactor)ReadRegister(register);
}
private void SetColorMask(GalPipelineState state)
{
bool colorMaskCommon = ReadRegisterBool(NvGpuEngine3dReg.ColorMaskCommon);
state.ColorMaskCommon = colorMaskCommon;
for (int index = 0; index < GalPipelineState.RenderTargetsCount; index++)
{
int colorMask = ReadRegister(NvGpuEngine3dReg.ColorMaskN + (colorMaskCommon ? 0 : index));
state.ColorMasks[index].Red = ((colorMask >> 0) & 0xf) != 0;
state.ColorMasks[index].Green = ((colorMask >> 4) & 0xf) != 0;
state.ColorMasks[index].Blue = ((colorMask >> 8) & 0xf) != 0;
state.ColorMasks[index].Alpha = ((colorMask >> 12) & 0xf) != 0;
}
}
private void SetPrimitiveRestart(GalPipelineState state)
2018-07-08 18:14:35 +02:00
{
state.PrimitiveRestartEnabled = ReadRegisterBool(NvGpuEngine3dReg.PrimRestartEnable);
2018-07-08 18:14:35 +02:00
if (state.PrimitiveRestartEnabled)
2018-07-08 18:14:35 +02:00
{
state.PrimitiveRestartIndex = (uint)ReadRegister(NvGpuEngine3dReg.PrimRestartIndex);
2018-07-08 18:14:35 +02:00
}
}
private void SetRenderTargets()
{
//Commercial games do not seem to
//bool SeparateFragData = ReadRegisterBool(NvGpuEngine3dReg.RTSeparateFragData);
uint control = (uint)(ReadRegister(NvGpuEngine3dReg.RtControl));
uint count = control & 0xf;
if (count > 0)
{
int[] map = new int[count];
for (int index = 0; index < count; index++)
{
int shift = 4 + index * 3;
map[index] = (int)((control >> shift) & 7);
}
_gpu.Renderer.RenderTarget.SetMap(map);
}
else
{
_gpu.Renderer.RenderTarget.SetMap(null);
}
}
private void UploadTextures(NvGpuVmm vmm, GalPipelineState state, long[] keys)
{
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.Begin(Profiles.GPU.Engine3d.UploadTextures);
long baseShPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.ShaderAddress);
int textureCbIndex = ReadRegister(NvGpuEngine3dReg.TextureCbIndex);
List<(long, GalImage, GalTextureSampler)> unboundTextures = new List<(long, GalImage, GalTextureSampler)>();
for (int index = 0; index < keys.Length; index++)
{
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
foreach (TextureDescriptor desc in _gpu.Renderer.Shader.GetTextureUsage(keys[index]))
{
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
int textureHandle;
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
if (desc.IsBindless)
{
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
long position = _constBuffers[index][desc.CbufSlot].Position;
textureHandle = vmm.ReadInt32(position + desc.CbufOffset * 4);
}
else
{
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
long position = _constBuffers[index][textureCbIndex].Position;
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
textureHandle = vmm.ReadInt32(position + desc.HandleIndex * 4);
}
unboundTextures.Add(UploadTexture(vmm, textureHandle));
}
}
for (int index = 0; index < unboundTextures.Count; index++)
{
(long key, GalImage image, GalTextureSampler sampler) = unboundTextures[index];
if (key == 0)
{
continue;
}
_gpu.Renderer.Texture.Bind(key, index, image);
_gpu.Renderer.Texture.SetSampler(image, sampler);
}
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.UploadTextures);
}
private (long, GalImage, GalTextureSampler) UploadTexture(NvGpuVmm vmm, int textureHandle)
{
if (textureHandle == 0)
{
//FIXME: Some games like puyo puyo will use handles with the value 0.
//This is a bug, most likely caused by sync issues.
return (0, default(GalImage), default(GalTextureSampler));
}
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.Begin(Profiles.GPU.Engine3d.UploadTexture);
bool linkedTsc = ReadRegisterBool(NvGpuEngine3dReg.LinkedTsc);
int ticIndex = (textureHandle >> 0) & 0xfffff;
int tscIndex = linkedTsc ? ticIndex : (textureHandle >> 20) & 0xfff;
long ticPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.TexHeaderPoolOffset);
long tscPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.TexSamplerPoolOffset);
ticPosition += ticIndex * 0x20;
tscPosition += tscIndex * 0x20;
GalImage image = TextureFactory.MakeTexture(vmm, ticPosition);
GalTextureSampler sampler = TextureFactory.MakeSampler(_gpu, vmm, tscPosition);
long key = vmm.ReadInt64(ticPosition + 4) & 0xffffffffffff;
if (image.Layout == GalMemoryLayout.BlockLinear)
{
key &= ~0x1ffL;
}
else if (image.Layout == GalMemoryLayout.Pitch)
{
key &= ~0x1fL;
}
key = vmm.GetPhysicalAddress(key);
if (key == -1)
{
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.UploadTexture);
//FIXME: Shouldn't ignore invalid addresses.
return (0, default(GalImage), default(GalTextureSampler));
}
_gpu.ResourceManager.SendTexture(vmm, key, image);
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.UploadTexture);
return (key, image, sampler);
}
private void UploadConstBuffers(NvGpuVmm vmm, GalPipelineState state, long[] keys)
{
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.Begin(Profiles.GPU.Engine3d.UploadConstBuffers);
for (int stage = 0; stage < keys.Length; stage++)
{
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
foreach (CBufferDescriptor desc in _gpu.Renderer.Shader.GetConstBufferUsage(keys[stage]))
{
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
ConstBuffer cb = _constBuffers[stage][desc.Slot];
if (!cb.Enabled)
{
continue;
}
long key = vmm.GetPhysicalAddress(cb.Position);
if (_gpu.ResourceManager.MemoryRegionModified(vmm, key, cb.Size, NvGpuBufferType.ConstBuffer))
{
if (vmm.TryGetHostAddress(cb.Position, cb.Size, out IntPtr cbPtr))
{
_gpu.Renderer.Buffer.SetData(key, cb.Size, cbPtr);
}
else
{
_gpu.Renderer.Buffer.SetData(key, vmm.ReadBytes(cb.Position, cb.Size));
}
}
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
state.ConstBufferKeys[stage][desc.Slot] = key;
}
}
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.UploadConstBuffers);
}
private void UploadVertexArrays(NvGpuVmm vmm, GalPipelineState state)
{
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.Begin(Profiles.GPU.Engine3d.UploadVertexArrays);
long ibPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.IndexArrayAddress);
long iboKey = vmm.GetPhysicalAddress(ibPosition);
int indexEntryFmt = ReadRegister(NvGpuEngine3dReg.IndexArrayFormat);
int indexCount = ReadRegister(NvGpuEngine3dReg.IndexBatchCount);
int primCtrl = ReadRegister(NvGpuEngine3dReg.VertexBeginGl);
GalPrimitiveType primType = (GalPrimitiveType)(primCtrl & 0xffff);
GalIndexFormat indexFormat = (GalIndexFormat)indexEntryFmt;
int indexEntrySize = 1 << indexEntryFmt;
if (indexEntrySize > 4)
{
throw new InvalidOperationException("Invalid index entry size \"" + indexEntrySize + "\"!");
}
if (indexCount != 0)
{
int ibSize = indexCount * indexEntrySize;
bool iboCached = _gpu.Renderer.Rasterizer.IsIboCached(iboKey, (uint)ibSize);
bool usesLegacyQuads =
primType == GalPrimitiveType.Quads ||
primType == GalPrimitiveType.QuadStrip;
if (!iboCached || _gpu.ResourceManager.MemoryRegionModified(vmm, iboKey, (uint)ibSize, NvGpuBufferType.Index))
{
if (!usesLegacyQuads)
{
if (vmm.TryGetHostAddress(ibPosition, ibSize, out IntPtr ibPtr))
{
_gpu.Renderer.Rasterizer.CreateIbo(iboKey, ibSize, ibPtr);
}
else
{
_gpu.Renderer.Rasterizer.CreateIbo(iboKey, ibSize, vmm.ReadBytes(ibPosition, ibSize));
}
}
else
{
byte[] buffer = vmm.ReadBytes(ibPosition, ibSize);
if (primType == GalPrimitiveType.Quads)
{
buffer = QuadHelper.ConvertQuadsToTris(buffer, indexEntrySize, indexCount);
}
else /* if (PrimType == GalPrimitiveType.QuadStrip) */
{
buffer = QuadHelper.ConvertQuadStripToTris(buffer, indexEntrySize, indexCount);
}
_gpu.Renderer.Rasterizer.CreateIbo(iboKey, ibSize, buffer);
}
}
if (!usesLegacyQuads)
{
_gpu.Renderer.Rasterizer.SetIndexArray(ibSize, indexFormat);
}
else
{
if (primType == GalPrimitiveType.Quads)
{
_gpu.Renderer.Rasterizer.SetIndexArray(QuadHelper.ConvertSizeQuadsToTris(ibSize), indexFormat);
}
else /* if (PrimType == GalPrimitiveType.QuadStrip) */
{
_gpu.Renderer.Rasterizer.SetIndexArray(QuadHelper.ConvertSizeQuadStripToTris(ibSize), indexFormat);
}
}
}
List<GalVertexAttrib>[] attribs = new List<GalVertexAttrib>[32];
for (int attr = 0; attr < 16; attr++)
{
int packed = ReadRegister(NvGpuEngine3dReg.VertexAttribNFormat + attr);
int arrayIndex = packed & 0x1f;
if (attribs[arrayIndex] == null)
{
attribs[arrayIndex] = new List<GalVertexAttrib>();
}
long vbPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.VertexArrayNAddress + arrayIndex * 4);
if (vbPosition == 0)
{
continue;
}
bool isConst = ((packed >> 6) & 1) != 0;
int offset = (packed >> 7) & 0x3fff;
GalVertexAttribSize size = (GalVertexAttribSize)((packed >> 21) & 0x3f);
GalVertexAttribType type = (GalVertexAttribType)((packed >> 27) & 0x7);
bool isRgba = ((packed >> 31) & 1) != 0;
// Check vertex array is enabled to avoid out of bounds exception when reading bytes
bool enable = (ReadRegister(NvGpuEngine3dReg.VertexArrayNControl + arrayIndex * 4) & 0x1000) != 0;
//Note: 16 is the maximum size of an attribute,
//having a component size of 32-bits with 4 elements (a vec4).
if (enable)
{
byte[] data = vmm.ReadBytes(vbPosition + offset, 16);
attribs[arrayIndex].Add(new GalVertexAttrib(attr, isConst, offset, data, size, type, isRgba));
}
}
state.VertexBindings = new GalVertexBinding[32];
for (int index = 0; index < 32; index++)
{
if (attribs[index] == null)
{
continue;
}
int control = ReadRegister(NvGpuEngine3dReg.VertexArrayNControl + index * 4);
bool enable = (control & 0x1000) != 0;
if (!enable)
{
continue;
}
long vbPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.VertexArrayNAddress + index * 4);
long vbEndPos = MakeInt64From2xInt32(NvGpuEngine3dReg.VertexArrayNEndAddr + index * 2);
int vertexDivisor = ReadRegister(NvGpuEngine3dReg.VertexArrayNDivisor + index * 4);
2018-08-25 06:16:58 +02:00
bool instanced = ReadRegisterBool(NvGpuEngine3dReg.VertexArrayNInstance + index);
int stride = control & 0xfff;
if (instanced && vertexDivisor != 0)
2018-08-25 06:16:58 +02:00
{
vbPosition += stride * (_currentInstance / vertexDivisor);
2018-08-25 06:16:58 +02:00
}
if (vbPosition > vbEndPos)
2018-08-25 06:16:58 +02:00
{
//Instance is invalid, ignore the draw call
continue;
}
long vboKey = vmm.GetPhysicalAddress(vbPosition);
2018-08-25 06:16:58 +02:00
long vbSize = (vbEndPos - vbPosition) + 1;
int modifiedVbSize = (int)vbSize;
// If quads convert size to triangle length
if (stride == 0)
{
if (primType == GalPrimitiveType.Quads)
{
modifiedVbSize = QuadHelper.ConvertSizeQuadsToTris(modifiedVbSize);
}
else if (primType == GalPrimitiveType.QuadStrip)
{
modifiedVbSize = QuadHelper.ConvertSizeQuadStripToTris(modifiedVbSize);
}
}
bool vboCached = _gpu.Renderer.Rasterizer.IsVboCached(vboKey, modifiedVbSize);
if (!vboCached || _gpu.ResourceManager.MemoryRegionModified(vmm, vboKey, vbSize, NvGpuBufferType.Vertex))
{
if ((primType == GalPrimitiveType.Quads | primType == GalPrimitiveType.QuadStrip) && stride != 0)
{
// Convert quad buffer to triangles
byte[] data = vmm.ReadBytes(vbPosition, vbSize);
if (primType == GalPrimitiveType.Quads)
{
data = QuadHelper.ConvertQuadsToTris(data, stride, (int)(vbSize / stride));
}
else
{
data = QuadHelper.ConvertQuadStripToTris(data, stride, (int)(vbSize / stride));
}
_gpu.Renderer.Rasterizer.CreateVbo(vboKey, data);
}
else if (vmm.TryGetHostAddress(vbPosition, vbSize, out IntPtr vbPtr))
{
_gpu.Renderer.Rasterizer.CreateVbo(vboKey, (int)vbSize, vbPtr);
}
else
{
_gpu.Renderer.Rasterizer.CreateVbo(vboKey, vmm.ReadBytes(vbPosition, vbSize));
}
}
state.VertexBindings[index].Enabled = true;
state.VertexBindings[index].Stride = stride;
state.VertexBindings[index].VboKey = vboKey;
state.VertexBindings[index].Instanced = instanced;
state.VertexBindings[index].Divisor = vertexDivisor;
state.VertexBindings[index].Attribs = attribs[index].ToArray();
}
GPU Profiling (#570) * Profiler initial setup * Capture actual timing data * Profiling data dumped to file on close * Support for multiple sessions under the same name * Service profiling * Sort output for easier read * csv output * Split session into 2 seperate values * Refactor name to category * Basic profiling window dummy. Toggle with F1 or set key with config No actual data displayed yet, just a pretty triangle * Simple font rendering * Display some actual timing data * Fix font bearing being ignored * x bearing and advance. Fixed y bearing calc * Different coloured lines to make reading easier * Scrolling * Multiple columns for name * Column titles * display in ms rather than ticks * Bars to display times * Sortable columns * Regex filtering * Better instant timing calculation Fixed minor regex bug * Better filtering Better max value calculation Skip some rendering to reduce profiler weight * Variable update rate * Show/hide inactive button Some other touchups * Add missing project reference * Hide inactive and pause * Fix viewport errors * Update initial window position * Variable name cleanup * Disable timing dump by default * Internal Profile refactor and cleanup * Timing info cleanup * Profile config cleanup * Settings cleanup * Button refactor * Profile refactor * Profile window cleanup * Window manager refactor * Font service cleanup * Fixed bug in profiling method where method was called twice without profiling enabled * Allow update rates of less than 1hz * Stop using window.run because it's apparently not great for performance. Some other performance things, should only draw a new frame when something has changed * Improved time tracking to keep history * Profile window was getting too long so I added regions and split bar rendering out into partial class * Dummy graph view with button to toggle * Realtime graphing initial commit * Display totals on new bar * Simple zooming support with arrow keys * Limit graph zoom and label start and stop * Added support for timing flags * Stop data running away when paused and frame updated * Manual step button * Update at when flag issued (ie every frame) * Removed useless finish profiling call * Enable and disable profiling at compile time. * Better plage for frame swap flag, also kept enough flags to cover larger time spans * No more stopwatches created, uses PerformanceCounter now * public and internal fields to props * Move visible update to update rather than draw as it causes a lockup if called from draw Also added profile window disposal so closing main window closes profiler too * Fixed optimization settings for profiled builds * Appveyer script guess to add profiling builds * Quotes * 1 less quote * Maybe escape space? * Specify config * Different approach * Fix file paths * Fix another path * Better artifact naming * Missing - * test string * Removed for, to test * readd for * moved dashes around so artifacts can begin with letters * quote env vars * martix * Removed configs * Much more efficient capture, ConcurrentDictionary was causing too much overhead * Skip repeating pixels during draw * Stop ram usage getting too high. Compensating for cleanup doing more now * Profile CPU, execute skipped because it's just too much work * Fixed bug with skipping draws. Furthest needed to be reset every loop * Less distracting colour for timing flags * Removed profile method function. It just doesn't play nice with conditional compilation so best to remove it now before it's used a lot * Profile vertex end * Null check for category, group and item * Forgot to reset instant count/time * Increment line when blank * Fix threading conflict Fixed instant count and time. Now accuratly represents the total time and count in the buffer * Fixed bug in time rendering where times were being trimmed to an int. Also added microsecond/millisecond formatting to reduce the number of decimal places needed * Profiled methods used by VertexEndGl * Texture 2d copy profiled * Support for multiple profiling levels * Sometimes it would have to wait a long time for lock to clear so moved it to a tryenter and skip if already locked * Dumb bug regarding clearing of timestamps. Start is already removed so no need to add it to the start * Optimisations in drawing routine: Only calculate bar top and bottom once per bar rather than once per timestamp Pre-calculate the right side of the graph as it was being calculated multiple times per bar Skip rendering timestamps that occupy the same pixel space now uses the raw timestamp to decide. While technically not as accurate it's much easier as the right side of the bar doesn't have to be calculated for a skipped timestamp * Couple alignment changes * Engine 3d call method profiled * Custom equals overload for profile config. The default implpmentation was just too slow * Bump cleanup thread priority. It clears the timer queue so it need to be run frequently * Fixed bug with scrolling caused by recent rendering optimisations. Simply forgot to increment the line index on a skipped line * Stopped blocking memory disposal so much. Also parralised(?) cleanup call * Profiled M2mf, P2mf and shader compilation * Missing file * Removed deprecated config * Removed legacy npaddebug * Using statement no longer needed * Apply suggestions from AcK Co-Authored-By: Ac_K <Acoustik666@gmail.com> * Apply suggestions from code review Co-Authored-By: Ac_K <Acoustik666@gmail.com>
2019-06-01 02:13:57 +02:00
Profile.End(Profiles.GPU.Engine3d.UploadVertexArrays);
}
private void DispatchRender(NvGpuVmm vmm, GalPipelineState state)
{
int indexCount = ReadRegister(NvGpuEngine3dReg.IndexBatchCount);
int primCtrl = ReadRegister(NvGpuEngine3dReg.VertexBeginGl);
GalPrimitiveType primType = (GalPrimitiveType)(primCtrl & 0xffff);
bool instanceNext = ((primCtrl >> 26) & 1) != 0;
bool instanceCont = ((primCtrl >> 27) & 1) != 0;
2018-08-25 06:16:58 +02:00
if (instanceNext && instanceCont)
2018-08-25 06:16:58 +02:00
{
throw new InvalidOperationException("GPU tried to increase and reset instance count at the same time");
}
if (instanceNext)
2018-08-25 06:16:58 +02:00
{
_currentInstance++;
2018-08-25 06:16:58 +02:00
}
else if (!instanceCont)
2018-08-25 06:16:58 +02:00
{
_currentInstance = 0;
2018-08-25 06:16:58 +02:00
}
state.Instance = _currentInstance;
2018-08-25 06:16:58 +02:00
_gpu.Renderer.Pipeline.Bind(state);
_gpu.Renderer.RenderTarget.Bind();
if (indexCount != 0)
{
int indexEntryFmt = ReadRegister(NvGpuEngine3dReg.IndexArrayFormat);
int indexFirst = ReadRegister(NvGpuEngine3dReg.IndexBatchFirst);
int vertexBase = ReadRegister(NvGpuEngine3dReg.VertexArrayElemBase);
long indexPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.IndexArrayAddress);
long iboKey = vmm.GetPhysicalAddress(indexPosition);
//Quad primitive types were deprecated on OpenGL 3.x,
//they are converted to a triangles index buffer on IB creation,
//so we should use the triangles type here too.
if (primType == GalPrimitiveType.Quads || primType == GalPrimitiveType.QuadStrip)
{
//Note: We assume that index first points to the first
//vertex of a quad, if it points to the middle of a
//quad (First % 4 != 0 for Quads) then it will not work properly.
if (primType == GalPrimitiveType.Quads)
{
indexFirst = QuadHelper.ConvertSizeQuadsToTris(indexFirst);
}
else // QuadStrip
{
indexFirst = QuadHelper.ConvertSizeQuadStripToTris(indexFirst);
}
primType = GalPrimitiveType.Triangles;
}
_gpu.Renderer.Rasterizer.DrawElements(iboKey, indexFirst, vertexBase, primType);
}
else
{
int vertexFirst = ReadRegister(NvGpuEngine3dReg.VertexArrayFirst);
int vertexCount = ReadRegister(NvGpuEngine3dReg.VertexArrayCount);
//Quad primitive types were deprecated on OpenGL 3.x,
//they are converted to a triangles index buffer on IB creation,
//so we should use the triangles type here too.
if (primType == GalPrimitiveType.Quads || primType == GalPrimitiveType.QuadStrip)
{
//Note: We assume that index first points to the first
//vertex of a quad, if it points to the middle of a
//quad (First % 4 != 0 for Quads) then it will not work properly.
if (primType == GalPrimitiveType.Quads)
{
vertexFirst = QuadHelper.ConvertSizeQuadsToTris(vertexFirst);
}
else // QuadStrip
{
vertexFirst = QuadHelper.ConvertSizeQuadStripToTris(vertexFirst);
}
primType = GalPrimitiveType.Triangles;
vertexCount = QuadHelper.ConvertSizeQuadsToTris(vertexCount);
}
_gpu.Renderer.Rasterizer.DrawArrays(vertexFirst, vertexCount, primType);
}
// Reset pipeline for host OpenGL calls
_gpu.Renderer.Pipeline.Unbind(state);
//Is the GPU really clearing those registers after draw?
WriteRegister(NvGpuEngine3dReg.IndexBatchFirst, 0);
WriteRegister(NvGpuEngine3dReg.IndexBatchCount, 0);
}
private enum QueryMode
{
WriteSeq,
Sync,
WriteCounterAndTimestamp
}
private void QueryControl(NvGpuVmm vmm, GpuMethodCall methCall)
{
WriteRegister(methCall);
long position = MakeInt64From2xInt32(NvGpuEngine3dReg.QueryAddress);
int seq = Registers[(int)NvGpuEngine3dReg.QuerySequence];
int ctrl = Registers[(int)NvGpuEngine3dReg.QueryControl];
QueryMode mode = (QueryMode)(ctrl & 3);
switch (mode)
{
case QueryMode.WriteSeq: vmm.WriteInt32(position, seq); break;
case QueryMode.WriteCounterAndTimestamp:
{
//TODO: Implement counters.
long counter = 1;
long timestamp = PerformanceCounter.ElapsedMilliseconds;
vmm.WriteInt64(position + 0, counter);
vmm.WriteInt64(position + 8, timestamp);
break;
}
}
}
private void CbData(NvGpuVmm vmm, GpuMethodCall methCall)
{
long position = MakeInt64From2xInt32(NvGpuEngine3dReg.ConstBufferAddress);
int offset = ReadRegister(NvGpuEngine3dReg.ConstBufferOffset);
vmm.WriteInt32(position + offset, methCall.Argument);
WriteRegister(NvGpuEngine3dReg.ConstBufferOffset, offset + 4);
_gpu.ResourceManager.ClearPbCache(NvGpuBufferType.ConstBuffer);
}
private void CbBind(NvGpuVmm vmm, GpuMethodCall methCall)
{
int stage = (methCall.Method - 0x904) >> 3;
int index = methCall.Argument;
bool enabled = (index & 1) != 0;
index = (index >> 4) & 0x1f;
long position = MakeInt64From2xInt32(NvGpuEngine3dReg.ConstBufferAddress);
long cbKey = vmm.GetPhysicalAddress(position);
int size = ReadRegister(NvGpuEngine3dReg.ConstBufferSize);
if (!_gpu.Renderer.Buffer.IsCached(cbKey, size))
{
_gpu.Renderer.Buffer.Create(cbKey, size);
}
ConstBuffer cb = _constBuffers[stage][index];
if (cb.Position != position || cb.Enabled != enabled || cb.Size != size)
{
_constBuffers[stage][index].Position = position;
_constBuffers[stage][index].Enabled = enabled;
_constBuffers[stage][index].Size = size;
}
}
private float GetFlipSign(NvGpuEngine3dReg reg)
{
return MathF.Sign(ReadRegisterFloat(reg));
}
private long MakeInt64From2xInt32(NvGpuEngine3dReg reg)
{
return
(long)Registers[(int)reg + 0] << 32 |
(uint)Registers[(int)reg + 1];
}
private void WriteRegister(GpuMethodCall methCall)
{
Registers[methCall.Method] = methCall.Argument;
}
private int ReadRegister(NvGpuEngine3dReg reg)
{
return Registers[(int)reg];
}
private float ReadRegisterFloat(NvGpuEngine3dReg reg)
{
return BitConverter.Int32BitsToSingle(ReadRegister(reg));
}
private bool ReadRegisterBool(NvGpuEngine3dReg reg)
{
return (ReadRegister(reg) & 1) != 0;
}
private void WriteRegister(NvGpuEngine3dReg reg, int value)
{
Registers[(int)reg] = value;
}
}
}