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

Shuffle constants around, squeezing them into gaps. Saves another 16 bytes.

Browse source
This commit is contained in:
Henrik Rydgård 2022-09-25 10:43:17 +02:00
parent f4b71e2dc7
commit cfa427c37a
2 changed files with 13 additions and 12 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

23
GPU/Common/ShaderUniforms.h
View file

@ -20,7 +20,8 @@ enum : uint64_t {
// TODO: Split into two structs, one for software transform and one for hardware transform, to save space.
// Currently 512 bytes. Probably can't get to 256 (nVidia's UBO alignment).
// Every line here is a 4-float.
struct UB_VS_FS_Base {
struct alignas(16) UB_VS_FS_Base {
float proj[16];
float proj_through[16];
float view[12];
@ -29,7 +30,6 @@ struct UB_VS_FS_Base {
float uvScaleOffset[4];
float depthRange[4];
// Rotation is used only for software transform.
float fogCoef[2]; float stencil; float rotation;
float matAmbient[4];
float cullRangeMin[4];
float cullRangeMax[4];
@ -40,10 +40,11 @@ struct UB_VS_FS_Base {
float texEnvColor[3];
uint32_t colorTestMask;
int alphaColorRef[4];
float blendFixA[4]; // .w is unused
float blendFixB[4]; // .w is unused
float blendFixA[3]; float stencil;
float blendFixB[3]; float rotation;
float texClamp[4];
float texClampOffset[4]; // .zw are unused
float texClampOffset[2]; // .zw are unused
float fogCoef[2];
};
static const char * const ub_baseStr =
@ -54,9 +55,6 @@ R"( mat4 u_proj;
mat3x4 u_texmtx;
vec4 u_uvscaleoffset;
vec4 u_depthRange;
vec2 u_fogcoef;
float u_stencilReplaceValue;
float u_rotation;
vec4 u_matambientalpha;
vec4 u_cullRangeMin;
vec4 u_cullRangeMax;
@ -68,15 +66,16 @@ R"( mat4 u_proj;
vec3 u_texenv;
uint u_alphacolormask;
ivec4 u_alphacolorref;
vec3 u_blendFixA;
vec3 u_blendFixB;
vec3 u_blendFixA; float u_stencilReplaceValue;
vec3 u_blendFixB; float u_rotation;
vec4 u_texclamp;
vec2 u_texclampoff;
vec2 u_fogcoef;
)";
// 512 bytes. Would like to shrink more. Some colors only have 8-bit precision and we expand
// them to float unnecessarily, could just as well expand in the shader.
struct UB_VS_Lights {
struct alignas(16) UB_VS_Lights {
float ambientColor[4];
float materialDiffuse[4];
float materialSpecular[4];
@ -129,7 +128,7 @@ R"( vec4 u_ambient;
// With some cleverness, we could get away with uploading just half this when only the four or five first
// bones are being used. This is 384b.
struct UB_VS_Bones {
struct alignas(16) UB_VS_Bones {
float bones[8][12];
};

2
GPU/Vulkan/DrawEngineVulkan.cpp
View file

@ -91,6 +91,8 @@ DrawEngineVulkan::DrawEngineVulkan(Draw::DrawContext *draw)
indexGen.Setup(decIndex);
InitDeviceObjects();
INFO_LOG(G3D, "sizeof(UB_VS_FS_Base) = %d", (int)sizeof(UB_VS_FS_Base));
}
void DrawEngineVulkan::InitDeviceObjects() {