Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

Fold the "materialUpdate" flag into the light ubershader part.

Browse source 
This reduces the number of vertex shaders and thus pipelines by quite a
bit more in a few games, like Tekken and GoW, continuing the fight
against shader compile stutter.

The perf impact should be minimal if not positive due to less pipeline
changes.

GLES fixes

Make the vertex input declarations match (always declare fog input).  Fixes D3D11 validation

Tess fix
This commit is contained in:
Henrik Rydgård 2022-09-26 11:20:17 +02:00
parent ad1021ea4b
commit 07ca9e4656
6 changed files with 33 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
GPU/Common/FragmentShaderGenerator.cpp
View file

@ -174,9 +174,7 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
WRITE(p, "layout (location = 1) %s in lowp vec4 v_color0;\n", shading);
if (lmode)
WRITE(p, "layout (location = 2) %s in lowp vec3 v_color1;\n", shading);
if (enableFog) {
WRITE(p, "layout (location = 3) in highp float v_fogdepth;\n");
}
WRITE(p, "layout (location = 3) in highp float v_fogdepth;\n");
if (doTexture) {
WRITE(p, "layout (location = 0) in highp vec3 v_texcoord;\n");
}
@ -279,9 +277,7 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
if (lmode) {
WRITE(p, " vec3 v_color1: COLOR1;\n");
}
if (enableFog) {
WRITE(p, " float v_fogdepth: TEXCOORD1;\n");
}
WRITE(p, " float v_fogdepth: TEXCOORD1;\n");
if (needFragCoord) {
if (compat.shaderLanguage == HLSL_D3D11) {
WRITE(p, " vec4 pixelPos : SV_POSITION;\n");
@ -394,8 +390,8 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
if (enableFog) {
*uniformMask |= DIRTY_FOGCOLOR;
WRITE(p, "uniform vec3 u_fogcolor;\n");
WRITE(p, "%s %s float v_fogdepth;\n", compat.varying_fs, highpFog ? "highp" : "mediump");
}
WRITE(p, "%s %s float v_fogdepth;\n", compat.varying_fs, highpFog ? "highp" : "mediump");
if (doTexture) {
WRITE(p, "%s %s vec3 v_texcoord;\n", compat.varying_fs, highpTexcoord ? "highp" : "mediump");
}

2
GPU/Common/ShaderId.cpp
View file

@ -123,11 +123,11 @@ void ComputeVertexShaderID(VShaderID *id_out, u32 vertType, bool useHWTransform,
if (gstate.isLightingEnabled()) {
// doShadeMapping is stored as UVGenMode, and light type doesn't matter for shade mapping.
id.SetBits(VS_BIT_MATERIAL_UPDATE, 3, gstate.getMaterialUpdate());
id.SetBit(VS_BIT_LIGHTING_ENABLE);
if (gstate_c.Supports(GPU_USE_LIGHT_UBERSHADER)) {
id.SetBit(VS_BIT_LIGHT_UBERSHADER);
} else {
id.SetBits(VS_BIT_MATERIAL_UPDATE, 3, gstate.getMaterialUpdate());
// Light bits
for (int i = 0; i < 4; i++) {
bool chanEnabled = gstate.isLightChanEnabled(i) != 0;

5
GPU/Common/ShaderUniforms.cpp
View file

@ -271,6 +271,8 @@ uint32_t PackLightControlBits() {
// Bit organization
// Bottom 4 bits are enable bits for each light.
// Then, for each light, comes 2 bits for "comp" and 2 bits for "type".
// At the end, at bit 20, we put the three material update bits.
uint32_t lightControl = 0;
for (int i = 0; i < 4; i++) {
if (gstate.isLightChanEnabled(i)) {
@ -282,6 +284,9 @@ uint32_t PackLightControlBits() {
lightControl |= computation << (4 + i * 4);
lightControl |= type << (4 + i * 4 + 2);
}
lightControl |= gstate.getMaterialUpdate() << 20;
return lightControl;
}

21
GPU/Common/VertexShaderGenerator.cpp
View file

@ -552,7 +552,7 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag
if (enableLighting) {
WRITE(p, "uniform lowp vec4 u_ambient;\n");
*uniformMask |= DIRTY_AMBIENT;
if ((matUpdate & 2) == 0 || !hasColor) {
if (lightUberShader || (matUpdate & 2) == 0 || !hasColor) {
WRITE(p, "uniform lowp vec3 u_matdiffuse;\n");
*uniformMask |= DIRTY_MATDIFFUSE;
}
@ -952,13 +952,22 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag
const char *srcCol = "color0";
if (doBezier || doSpline) {
// TODO: Probably, should use hasColorTess but FF4 has a problem with drawing the background.
srcCol = "tess";
srcCol = "tess.col";
}
p.F(" vec4 ambientColor = %s;\n", (matUpdate & 1) && hasColor ? srcCol : "u_matambientalpha");
if (enableLighting) {
p.F(" vec3 diffuseColor = %s.rgb;\n", (matUpdate & 2) && hasColor ? srcCol : "u_matdiffuse");
p.F(" vec3 specularColor = %s.rgb;\n", (matUpdate & 4) && hasColor ? srcCol : "u_matspecular");
if (lightUberShader && hasColor) {
p.F(" vec4 ambientColor = ((u_lightControl & (1u << 20u)) != 0u) ? %s : u_matambientalpha;\n", srcCol);
if (enableLighting) {
p.F(" vec3 diffuseColor = ((u_lightControl & (1u << 21u)) != 0u) ? %s.rgb : u_matdiffuse;\n", srcCol);
p.F(" vec3 specularColor = ((u_lightControl & (1u << 22u)) != 0u) ? %s.rgb : u_matspecular.rgb;\n", srcCol);
}
} else {
// This path also takes care of the lightUberShader && !hasColor path, because all comparisons fail.
p.F(" vec4 ambientColor = %s;\n", (matUpdate & 1) && hasColor ? srcCol : "u_matambientalpha");
if (enableLighting) {
p.F(" vec3 diffuseColor = %s.rgb;\n", (matUpdate & 2) && hasColor ? srcCol : "u_matdiffuse");
p.F(" vec3 specularColor = %s.rgb;\n", (matUpdate & 4) && hasColor ? srcCol : "u_matspecular");
}
}
bool diffuseIsZero = true;

8
GPU/GPUCommon.cpp
View file

@ -465,6 +465,14 @@ void GPUCommon::UpdateCmdInfo() {
cmdInfo_[GE_CMD_LIGHTTYPE0 + i].AddDirty(DIRTY_VERTEXSHADER_STATE);
}
}
if (gstate_c.Supports(GPU_USE_LIGHT_UBERSHADER)) {
cmdInfo_[GE_CMD_MATERIALUPDATE].RemoveDirty(DIRTY_VERTEXSHADER_STATE);
cmdInfo_[GE_CMD_MATERIALUPDATE].AddDirty(DIRTY_LIGHT_CONTROL);
} else {
cmdInfo_[GE_CMD_MATERIALUPDATE].RemoveDirty(DIRTY_LIGHT_CONTROL);
cmdInfo_[GE_CMD_MATERIALUPDATE].AddDirty(DIRTY_VERTEXSHADER_STATE);
}
}
void GPUCommon::BeginHostFrame() {

2
GPU/Vulkan/ShaderManagerVulkan.cpp
View file

@ -376,7 +376,7 @@ VulkanFragmentShader *ShaderManagerVulkan::GetFragmentShaderFromModule(VkShaderM
// instantaneous.
#define CACHE_HEADER_MAGIC 0xff51f420
#define CACHE_VERSION 27
#define CACHE_VERSION 28
struct VulkanCacheHeader {
uint32_t magic;
uint32_t version;