Fixes Mario Tennis Gimmick Courts and adds support for FastDepthCalc

- Calculate ZSlope every flush but only set PixelShader Constant on Reset Buffer when zfreeze
- Fixed another Pixel Shader bug in D3D that was giving me grief
This commit is contained in:
NanoByte011 2015-01-13 02:55:25 -07:00 committed by Scott Mansell
parent 6d5065c58d
commit add59b3bea
7 changed files with 52 additions and 26 deletions

View File

@ -181,12 +181,10 @@ void VertexManager::vFlush(bool useDstAlpha)
PrepareDrawBuffers(stride);
if (!bpmem.genMode.zfreeze && IndexGenerator::GetIndexLen() >= 3)
{
if (!bpmem.genMode.zfreeze)
CalculateZSlope(stride);
}
// if cull mode is CULL_ALL, ignore triangles and quads
// If cull mode is CULL_ALL, do not render these triangles
if (bpmem.genMode.cullmode == GenMode::CULL_ALL && current_primitive_type == PRIMITIVE_TRIANGLES)
return;
@ -202,6 +200,9 @@ void VertexManager::ResetBuffer(u32 stride)
{
s_pCurBufferPointer = s_pBaseBufferPointer;
IndexGenerator::Start(GetIndexBuffer());
if (bpmem.genMode.zfreeze)
PixelShaderManager::SetZSlope(ZSlope.dfdx, ZSlope.dfdy, ZSlope.f0);
}
} // namespace

View File

@ -89,6 +89,9 @@ void VertexManager::ResetBuffer(u32 stride)
buffer = s_indexBuffer->Map(MAXIBUFFERSIZE * sizeof(u16));
IndexGenerator::Start((u16*)buffer.first);
s_index_offset = buffer.second;
if (bpmem.genMode.zfreeze)
PixelShaderManager::SetZSlope(ZSlope.dfdx, ZSlope.dfdy, ZSlope.f0);
}
void VertexManager::Draw(u32 stride)
@ -140,12 +143,10 @@ void VertexManager::vFlush(bool useDstAlpha)
PrepareDrawBuffers(stride);
if (!bpmem.genMode.zfreeze && IndexGenerator::GetIndexLen() >= 3)
{
if (!bpmem.genMode.zfreeze)
CalculateZSlope(stride);
}
// if cull mode is CULL_ALL, ignore triangles and quads
// If cull mode is CULL_ALL, do not render these triangles
if (bpmem.genMode.cullmode == GenMode::CULL_ALL && current_primitive_type == PRIMITIVE_TRIANGLES)
return;

View File

@ -271,7 +271,11 @@ static inline void GeneratePixelShader(T& out, DSTALPHA_MODE dstAlphaMode, API_T
GenerateVSOutputMembers<T>(out, ApiType);
out.Write("};\n");
const bool forced_early_z = g_ActiveConfig.backend_info.bSupportsEarlyZ && bpmem.UseEarlyDepthTest() && (g_ActiveConfig.bFastDepthCalc || bpmem.alpha_test.TestResult() == AlphaTest::UNDETERMINED);
const bool forced_early_z = g_ActiveConfig.backend_info.bSupportsEarlyZ && bpmem.UseEarlyDepthTest()
&& (g_ActiveConfig.bFastDepthCalc || bpmem.alpha_test.TestResult() == AlphaTest::UNDETERMINED)
// We can't allow early_ztest for zfreeze because a reference poly is used
// to control the depth and we need a depth test after the alpha test.
&& !bpmem.genMode.zfreeze;
const bool per_pixel_depth = (bpmem.ztex2.op != ZTEXTURE_DISABLE && bpmem.UseLateDepthTest()) || (!g_ActiveConfig.bFastDepthCalc && bpmem.zmode.testenable && !forced_early_z) || bpmem.genMode.zfreeze;
if (forced_early_z)
@ -365,7 +369,7 @@ static inline void GeneratePixelShader(T& out, DSTALPHA_MODE dstAlphaMode, API_T
out.Write("void main(\n");
out.Write(" out float4 ocol0 : SV_Target0,%s%s\n in float4 rawpos : SV_Position,\n",
dstAlphaMode == DSTALPHA_DUAL_SOURCE_BLEND ? "\n out float4 ocol1 : SV_Target1," : "",
per_pixel_depth ? "\n out float depth : SV_Depth," : "");
(per_pixel_depth && bpmem.zmode.testenable) ? "\n out float depth : SV_Depth," : "");
out.Write(" in centroid float4 colors_0 : COLOR0,\n");
out.Write(" in centroid float4 colors_1 : COLOR1\n");
@ -1023,7 +1027,11 @@ static inline void WriteAlphaTest(T& out, pixel_shader_uid_data* uid_data, API_T
// Tests seem to have proven that writing depth even when the alpha test fails is more
// important that a reliable alpha test, so we just force the alpha test to always succeed.
// At least this seems to be less buggy.
uid_data->alpha_test_use_zcomploc_hack = bpmem.UseEarlyDepthTest() && bpmem.zmode.updateenable && !g_ActiveConfig.backend_info.bSupportsEarlyZ;
uid_data->alpha_test_use_zcomploc_hack = bpmem.UseEarlyDepthTest()
&& bpmem.zmode.updateenable
&& !g_ActiveConfig.backend_info.bSupportsEarlyZ
&& !bpmem.genMode.zfreeze; // Might not be neccessary
if (!uid_data->alpha_test_use_zcomploc_hack)
{
out.Write("\t\tdiscard;\n");
@ -1114,10 +1122,10 @@ static inline void WritePerPixelDepth(T& out, pixel_shader_uid_data* uid_data, A
out.Write("\tfloat2 screenpos = rawpos.xy * " I_EFBSCALE".xy;\n");
// Opengl has reversed vertical screenspace coordiantes
if(ApiType == API_OPENGL)
if (ApiType == API_OPENGL)
out.Write("\tscreenpos.y = %i - screenpos.y - 1;\n", EFB_HEIGHT);
out.Write("\tdepth = float(" I_ZSLOPE".z + " I_ZSLOPE".x * screenpos.x + " I_ZSLOPE".y * screenpos.y) / float(0xffffff);\n");
out.Write("\tdepth = float(" I_ZSLOPE".z + " I_ZSLOPE".x * screenpos.x + " I_ZSLOPE".y * screenpos.y) / float(0xFFFFFF);\n");
}
else
{

View File

@ -50,7 +50,7 @@ void PixelShaderManager::Dirty()
SetZTextureBias();
SetViewportChanged();
SetEfbScaleChanged();
SetZSlope(0, 0, 1);
SetZSlope(0, 0, (float)0xFFFFFF);
SetIndTexScaleChanged(false);
SetIndTexScaleChanged(true);
SetIndMatrixChanged(0);
@ -116,7 +116,8 @@ void PixelShaderManager::SetConstants()
s_bViewPortChanged = false;
}
if (s_bEFBScaleChanged) {
if (s_bEFBScaleChanged)
{
constants.efbscale[0] = 1.0f / float(Renderer::EFBToScaledXf(1));
constants.efbscale[1] = 1.0f / float(Renderer::EFBToScaledYf(1));
dirty = true;

View File

@ -25,6 +25,8 @@ u8 *VertexManager::s_pEndBufferPointer;
PrimitiveType VertexManager::current_primitive_type;
Slope VertexManager::ZSlope;
bool VertexManager::IsFlushed;
static const PrimitiveType primitive_from_gx[8] = {
@ -246,6 +248,8 @@ void VertexManager::CalculateZSlope(u32 stride)
{
float vtx[9];
float out[12];
float viewOffset[2] = { xfmem.viewport.xOrig - bpmem.scissorOffset.x * 2,
xfmem.viewport.yOrig - bpmem.scissorOffset.y * 2};
// Lookup vertices of the last rendered triangle and software-transform them
// This allows us to determine the depth slope, which will be used if zfreeze
@ -260,9 +264,11 @@ void VertexManager::CalculateZSlope(u32 stride)
VertexShaderManager::TransformToClipSpace(&vtx[i * 3], &out[i * 4]);
// Transform to Screenspace
out[0 + i * 4] = out[0 + i * 4] / out[3 + i * 4] * xfmem.viewport.wd + (xfmem.viewport.xOrig - 342);
out[1 + i * 4] = out[1 + i * 4] / out[3 + i * 4] * xfmem.viewport.ht + (xfmem.viewport.yOrig - 342);
out[2 + i * 4] = out[2 + i * 4] / out[3 + i * 4] * xfmem.viewport.zRange + xfmem.viewport.farZ;
float w = out[3 + i * 4];
out[0 + i * 4] = out[0 + i * 4] / w * xfmem.viewport.wd + viewOffset[0];
out[1 + i * 4] = out[1 + i * 4] / w * xfmem.viewport.ht + viewOffset[1];
out[2 + i * 4] = out[2 + i * 4] / w * xfmem.viewport.zRange + xfmem.viewport.farZ;
}
float dx31 = out[8] - out[0];
@ -276,9 +282,11 @@ void VertexManager::CalculateZSlope(u32 stride)
float b = dx31 * DF21 + dx12 * DF31;
float c = -dx12 * dy31 - dx31 * -dy12;
float slope_dfdx = -a / c;
float slope_dfdy = -b / c;
float slope_f0 = out[2] - (out[0] * slope_dfdx + out[1] * slope_dfdy);
// Stop divide by zero
if (c == 0)
return;
PixelShaderManager::SetZSlope(slope_dfdx, slope_dfdy, slope_f0);
ZSlope.dfdx = -a / c;
ZSlope.dfdy = -b / c;
ZSlope.f0 = out[2] - (out[0] * ZSlope.dfdx + out[1] * ZSlope.dfdy);
}

View File

@ -14,6 +14,13 @@ enum PrimitiveType {
PRIMITIVE_TRIANGLES,
};
struct Slope
{
float dfdx;
float dfdy;
float f0;
};
class VertexManager
{
private:
@ -41,8 +48,6 @@ public:
static void DoState(PointerWrap& p);
static void CalculateZSlope(u32 stride);
protected:
virtual void vDoState(PointerWrap& p) { }
@ -57,6 +62,9 @@ protected:
static u32 GetRemainingSize();
static u32 GetRemainingIndices(int primitive);
static Slope ZSlope;
static void CalculateZSlope(u32 stride);
private:
static bool IsFlushed;

View File

@ -692,6 +692,7 @@ void VertexShaderManager::ResetView()
void VertexShaderManager::TransformToClipSpace(const float* data, float *out)
{
// Can we use constants.posnormalmatrix here instead?
const float *world_matrix = (const float *)xfmem.posMatrices + g_main_cp_state.matrix_index_a.PosNormalMtxIdx * 4;
const float *proj_matrix = &g_fProjectionMatrix[0];
@ -700,8 +701,6 @@ void VertexShaderManager::TransformToClipSpace(const float* data, float *out)
t[1] = data[0] * world_matrix[4] + data[1] * world_matrix[5] + data[2] * world_matrix[6] + world_matrix[7];
t[2] = data[0] * world_matrix[8] + data[1] * world_matrix[9] + data[2] * world_matrix[10] + world_matrix[11];
// TODO: this requires g_fProjectionMatrix to be up to date, which is not really a good design decision.
out[0] = t[0] * proj_matrix[0] + t[1] * proj_matrix[1] + t[2] * proj_matrix[2] + proj_matrix[3];
out[1] = t[0] * proj_matrix[4] + t[1] * proj_matrix[5] + t[2] * proj_matrix[6] + proj_matrix[7];
out[2] = t[0] * proj_matrix[8] + t[1] * proj_matrix[9] + t[2] * proj_matrix[10] + proj_matrix[11];