// Syncs the shader constant buffers with xfmem
// TODO: A cleaner way to control the matrices without making a mess in the parameters field
void VertexShaderManager::SetConstants()
{
if (nTransformMatricesChanged[0] >= 0)
{
int startn = nTransformMatricesChanged[0] / 4;
int endn = (nTransformMatricesChanged[1] + 3) / 4;
memcpy(constants.transformmatrices[startn].data(), &xfmem.posMatrices[startn * 4],
(endn - startn) * sizeof(float4));
dirty = true;
nTransformMatricesChanged[0] = nTransformMatricesChanged[1] = -1;
}
if (nNormalMatricesChanged[0] >= 0)
{
int startn = nNormalMatricesChanged[0] / 3;
int endn = (nNormalMatricesChanged[1] + 2) / 3;
for (int i = startn; i < endn; i++)
{
memcpy(constants.normalmatrices[i].data(), &xfmem.normalMatrices[3 * i], 12);
}
dirty = true;
nNormalMatricesChanged[0] = nNormalMatricesChanged[1] = -1;
}
if (nPostTransformMatricesChanged[0] >= 0)
{
int startn = nPostTransformMatricesChanged[0] / 4;
int endn = (nPostTransformMatricesChanged[1] + 3) / 4;
memcpy(constants.posttransformmatrices[startn].data(), &xfmem.postMatrices[startn * 4],
(endn - startn) * sizeof(float4));
dirty = true;
nPostTransformMatricesChanged[0] = nPostTransformMatricesChanged[1] = -1;
}
if (nLightsChanged[0] >= 0)
{
// TODO: Outdated comment
// lights don't have a 1 to 1 mapping, the color component needs to be converted to 4 floats
int istart = nLightsChanged[0] / 0x10;
int iend = (nLightsChanged[1] + 15) / 0x10;
for (int i = istart; i < iend; ++i)
{
const Light& light = xfmem.lights[i];
VertexShaderConstants::Light& dstlight = constants.lights[i];
// xfmem.light.color is packed as abgr in u8[4], so we have to swap the order
dstlight.color[0] = light.color[3];
dstlight.color[1] = light.color[2];
dstlight.color[2] = light.color[1];
dstlight.color[3] = light.color[0];
dstlight.cosatt[0] = light.cosatt[0];
dstlight.cosatt[1] = light.cosatt[1];
dstlight.cosatt[2] = light.cosatt[2];
if (fabs(light.distatt[0]) < 0.00001f && fabs(light.distatt[1]) < 0.00001f &&
fabs(light.distatt[2]) < 0.00001f)
{
// dist attenuation, make sure not equal to 0!!!
dstlight.distatt[0] = .00001f;
}
else
{
dstlight.distatt[0] = light.distatt[0];
}
dstlight.distatt[1] = light.distatt[1];
dstlight.distatt[2] = light.distatt[2];
dstlight.pos[0] = light.dpos[0];
dstlight.pos[1] = light.dpos[1];
dstlight.pos[2] = light.dpos[2];
double norm = double(light.ddir[0]) * double(light.ddir[0]) +
double(light.ddir[1]) * double(light.ddir[1]) +
double(light.ddir[2]) * double(light.ddir[2]);
norm = 1.0 / sqrt(norm);
float norm_float = static_cast<float>(norm);
dstlight.dir[0] = light.ddir[0] * norm_float;
dstlight.dir[1] = light.ddir[1] * norm_float;
dstlight.dir[2] = light.ddir[2] * norm_float;
}
dirty = true;
nLightsChanged[0] = nLightsChanged[1] = -1;
}
for (int i : nMaterialsChanged)
{
u32 data = i >= 2 ? xfmem.matColor[i - 2] : xfmem.ambColor[i];
constants.materials[i][0] = (data >> 24) & 0xFF;
constants.materials[i][1] = (data >> 16) & 0xFF;
constants.materials[i][2] = (data >> 8) & 0xFF;
constants.materials[i][3] = data & 0xFF;
dirty = true;
}
nMaterialsChanged = BitSet32(0);
if (bPosNormalMatrixChanged)
{
bPosNormalMatrixChanged = false;
const float* pos = &xfmem.posMatrices[g_main_cp_state.matrix_index_a.PosNormalMtxIdx * 4];
const float* norm =
&xfmem.normalMatrices[3 * (g_main_cp_state.matrix_index_a.PosNormalMtxIdx & 31)];
memcpy(constants.posnormalmatrix.data(), pos, 3 * sizeof(float4));
memcpy(constants.posnormalmatrix[3].data(), norm, 3 * sizeof(float));
memcpy(constants.posnormalmatrix[4].data(), norm + 3, 3 * sizeof(float));
memcpy(constants.posnormalmatrix[5].data(), norm + 6, 3 * sizeof(float));
dirty = true;
}
if (bTexMatricesChanged[0])
{
bTexMatricesChanged[0] = false;
const float* pos_matrix_ptrs[] = {
&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex0MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex1MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex2MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex3MtxIdx * 4]};
for (size_t i = 0; i < ArraySize(pos_matrix_ptrs); ++i)
{
memcpy(constants.texmatrices[3 * i].data(), pos_matrix_ptrs[i], 3 * sizeof(float4));
}
dirty = true;
}
if (bTexMatricesChanged[1])
{
bTexMatricesChanged[1] = false;
const float* pos_matrix_ptrs[] = {
&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex4MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex5MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex6MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex7MtxIdx * 4]};
for (size_t i = 0; i < ArraySize(pos_matrix_ptrs); ++i)
{
memcpy(constants.texmatrices[3 * i + 12].data(), pos_matrix_ptrs[i], 3 * sizeof(float4));
}
dirty = true;
}
if (bViewportChanged)
{
bViewportChanged = false;
// The console GPU places the pixel center at 7/12 unless antialiasing
// is enabled, while D3D and OpenGL place it at 0.5. See the comment
// in VertexShaderGen.cpp for details.
// NOTE: If we ever emulate antialiasing, the sample locations set by
// BP registers 0x01-0x04 need to be considered here.
const float pixel_center_correction = 7.0f / 12.0f - 0.5f;
const bool bUseVertexRounding = g_ActiveConfig.bVertexRounding && g_ActiveConfig.iEFBScale != 1;
const float viewport_width = bUseVertexRounding ?
(2.f * xfmem.viewport.wd) :
g_renderer->EFBToScaledXf(2.f * xfmem.viewport.wd);
const float viewport_height = bUseVertexRounding ?
(2.f * xfmem.viewport.ht) :
g_renderer->EFBToScaledXf(2.f * xfmem.viewport.ht);
const float pixel_size_x = 2.f / viewport_width;
const float pixel_size_y = 2.f / viewport_height;
constants.pixelcentercorrection[0] = pixel_center_correction * pixel_size_x;
constants.pixelcentercorrection[1] = pixel_center_correction * pixel_size_y;
// By default we don't change the depth value at all in the vertex shader.
constants.pixelcentercorrection[2] = 1.0f;
constants.pixelcentercorrection[3] = 0.0f;
constants.viewport[0] = (2.f * xfmem.viewport.wd);
constants.viewport[1] = (2.f * xfmem.viewport.ht);
if (g_renderer->UseVertexDepthRange())
{
// Oversized depth ranges are handled in the vertex shader. We need to reverse
// the far value to use the reversed-Z trick.
if (g_ActiveConfig.backend_info.bSupportsReversedDepthRange)
{
// Sometimes the console also tries to use the reversed-Z trick. We can only do
// that with the expected accuracy if the backend can reverse the depth range.
constants.pixelcentercorrection[2] = fabs(xfmem.viewport.zRange) / 16777215.0f;
if (xfmem.viewport.zRange < 0.0f)
constants.pixelcentercorrection[3] = xfmem.viewport.farZ / 16777215.0f;
else
constants.pixelcentercorrection[3] = 1.0f - xfmem.viewport.farZ / 16777215.0f;
}
else
{
// For backends that don't support reversing the depth range we can still render
// cases where the console uses the reversed-Z trick. But we simply can't provide
// the expected accuracy, which might result in z-fighting.
constants.pixelcentercorrection[2] = xfmem.viewport.zRange / 16777215.0f;
constants.pixelcentercorrection[3] = 1.0f - xfmem.viewport.farZ / 16777215.0f;
}
}
dirty = true;
BPFunctions::SetViewport();
// Update projection if the viewport isn't 1:1 useable
if (!g_ActiveConfig.backend_info.bSupportsOversizedViewports)
{
ViewportCorrectionMatrix(s_viewportCorrection);
bProjectionChanged = true;
}
}
if (bProjectionChanged)
{
bProjectionChanged = false;
float* rawProjection = xfmem.projection.rawProjection;
switch (xfmem.projection.type)
{
case GX_PERSPECTIVE:
g_fProjectionMatrix[0] = rawProjection[0] * g_ActiveConfig.fAspectRatioHackW;
g_fProjectionMatrix[1] = 0.0f;
g_fProjectionMatrix[2] = rawProjection[1] * g_ActiveConfig.fAspectRatioHackW;
g_fProjectionMatrix[3] = 0.0f;
g_fProjectionMatrix[4] = 0.0f;
g_fProjectionMatrix[5] = rawProjection[2] * g_ActiveConfig.fAspectRatioHackH;
g_fProjectionMatrix[6] = rawProjection[3] * g_ActiveConfig.fAspectRatioHackH;
g_fProjectionMatrix[7] = 0.0f;
g_fProjectionMatrix[8] = 0.0f;
g_fProjectionMatrix[9] = 0.0f;
g_fProjectionMatrix[10] = rawProjection[4];
g_fProjectionMatrix[11] = rawProjection[5];
g_fProjectionMatrix[12] = 0.0f;
g_fProjectionMatrix[13] = 0.0f;
g_fProjectionMatrix[14] = -1.0f;
g_fProjectionMatrix[15] = 0.0f;
SETSTAT_FT(stats.gproj_0, g_fProjectionMatrix[0]);
SETSTAT_FT(stats.gproj_1, g_fProjectionMatrix[1]);
SETSTAT_FT(stats.gproj_2, g_fProjectionMatrix[2]);
SETSTAT_FT(stats.gproj_3, g_fProjectionMatrix[3]);
SETSTAT_FT(stats.gproj_4, g_fProjectionMatrix[4]);
SETSTAT_FT(stats.gproj_5, g_fProjectionMatrix[5]);
SETSTAT_FT(stats.gproj_6, g_fProjectionMatrix[6]);
SETSTAT_FT(stats.gproj_7, g_fProjectionMatrix[7]);
SETSTAT_FT(stats.gproj_8, g_fProjectionMatrix[8]);
SETSTAT_FT(stats.gproj_9, g_fProjectionMatrix[9]);
SETSTAT_FT(stats.gproj_10, g_fProjectionMatrix[10]);
SETSTAT_FT(stats.gproj_11, g_fProjectionMatrix[11]);
SETSTAT_FT(stats.gproj_12, g_fProjectionMatrix[12]);
SETSTAT_FT(stats.gproj_13, g_fProjectionMatrix[13]);
SETSTAT_FT(stats.gproj_14, g_fProjectionMatrix[14]);
SETSTAT_FT(stats.gproj_15, g_fProjectionMatrix[15]);
break;
case GX_ORTHOGRAPHIC:
g_fProjectionMatrix[0] = rawProjection[0];
g_fProjectionMatrix[1] = 0.0f;
g_fProjectionMatrix[2] = 0.0f;
g_fProjectionMatrix[3] = rawProjection[1];
g_fProjectionMatrix[4] = 0.0f;
g_fProjectionMatrix[5] = rawProjection[2];
g_fProjectionMatrix[6] = 0.0f;
g_fProjectionMatrix[7] = rawProjection[3];
g_fProjectionMatrix[8] = 0.0f;
g_fProjectionMatrix[9] = 0.0f;
g_fProjectionMatrix[10] = rawProjection[4];
g_fProjectionMatrix[11] = rawProjection[5];
g_fProjectionMatrix[12] = 0.0f;
g_fProjectionMatrix[13] = 0.0f;
g_fProjectionMatrix[14] = 0.0f;
g_fProjectionMatrix[15] = 1.0f;
SETSTAT_FT(stats.g2proj_0, g_fProjectionMatrix[0]);
SETSTAT_FT(stats.g2proj_1, g_fProjectionMatrix[1]);
SETSTAT_FT(stats.g2proj_2, g_fProjectionMatrix[2]);
SETSTAT_FT(stats.g2proj_3, g_fProjectionMatrix[3]);
SETSTAT_FT(stats.g2proj_4, g_fProjectionMatrix[4]);
SETSTAT_FT(stats.g2proj_5, g_fProjectionMatrix[5]);
SETSTAT_FT(stats.g2proj_6, g_fProjectionMatrix[6]);
SETSTAT_FT(stats.g2proj_7, g_fProjectionMatrix[7]);
SETSTAT_FT(stats.g2proj_8, g_fProjectionMatrix[8]);
SETSTAT_FT(stats.g2proj_9, g_fProjectionMatrix[9]);
SETSTAT_FT(stats.g2proj_10, g_fProjectionMatrix[10]);
SETSTAT_FT(stats.g2proj_11, g_fProjectionMatrix[11]);
SETSTAT_FT(stats.g2proj_12, g_fProjectionMatrix[12]);
SETSTAT_FT(stats.g2proj_13, g_fProjectionMatrix[13]);
SETSTAT_FT(stats.g2proj_14, g_fProjectionMatrix[14]);
SETSTAT_FT(stats.g2proj_15, g_fProjectionMatrix[15]);
SETSTAT_FT(stats.proj_0, rawProjection[0]);
SETSTAT_FT(stats.proj_1, rawProjection[1]);
SETSTAT_FT(stats.proj_2, rawProjection[2]);
SETSTAT_FT(stats.proj_3, rawProjection[3]);
SETSTAT_FT(stats.proj_4, rawProjection[4]);
SETSTAT_FT(stats.proj_5, rawProjection[5]);
break;
default:
ERROR_LOG(VIDEO, "Unknown projection type: %d", xfmem.projection.type);
}
PRIM_LOG("Projection: %f %f %f %f %f %f", rawProjection[0], rawProjection[1], rawProjection[2],
rawProjection[3], rawProjection[4], rawProjection[5]);
if (g_ActiveConfig.bFreeLook && xfmem.projection.type == GX_PERSPECTIVE)
{
Matrix44 mtxA;
Matrix44 mtxB;
Matrix44 viewMtx;
Matrix44::Translate(mtxA, s_fViewTranslationVector);
Matrix44::LoadMatrix33(mtxB, s_viewRotationMatrix);
Matrix44::Multiply(mtxB, mtxA, viewMtx); // view = rotation x translation
Matrix44::Set(mtxB, g_fProjectionMatrix);
Matrix44::Multiply(mtxB, viewMtx, mtxA); // mtxA = projection x view
Matrix44::Multiply(s_viewportCorrection, mtxA, mtxB); // mtxB = viewportCorrection x mtxA
memcpy(constants.projection.data(), mtxB.data, 4 * sizeof(float4));
}
else
{
Matrix44 projMtx;
Matrix44::Set(projMtx, g_fProjectionMatrix);
Matrix44 correctedMtx;
Matrix44::Multiply(s_viewportCorrection, projMtx, correctedMtx);
memcpy(constants.projection.data(), correctedMtx.data, 4 * sizeof(float4));
}
dirty = true;
}
if (bTexMtxInfoChanged)
{
bTexMtxInfoChanged = false;
constants.xfmem_dualTexInfo = xfmem.dualTexTrans.enabled;
for (size_t i = 0; i < ArraySize(xfmem.texMtxInfo); i++)
constants.xfmem_pack1[i][0] = xfmem.texMtxInfo[i].hex;
for (size_t i = 0; i < ArraySize(xfmem.postMtxInfo); i++)
constants.xfmem_pack1[i][1] = xfmem.postMtxInfo[i].hex;
dirty = true;
}
if (bLightingConfigChanged)
{
bLightingConfigChanged = false;
for (size_t i = 0; i < 2; i++)
{
constants.xfmem_pack1[i][2] = xfmem.color[i].hex;
constants.xfmem_pack1[i][3] = xfmem.alpha[i].hex;
}
constants.xfmem_numColorChans = xfmem.numChan.numColorChans;
dirty = true;
}
}
// Syncs the shader constant buffers with xfmem
// TODO: A cleaner way to control the matrices without making a mess in the parameters field
void VertexShaderManager::SetConstants()
{
if (nTransformMatricesChanged[0] >= 0)
{
int startn = nTransformMatricesChanged[0] / 4;
int endn = (nTransformMatricesChanged[1] + 3) / 4;
memcpy(constants.transformmatrices[startn], &xfmem.posMatrices[startn * 4], (endn - startn) * 16);
dirty = true;
nTransformMatricesChanged[0] = nTransformMatricesChanged[1] = -1;
}
if (nNormalMatricesChanged[0] >= 0)
{
int startn = nNormalMatricesChanged[0] / 3;
int endn = (nNormalMatricesChanged[1] + 2) / 3;
for (int i=startn; i<endn; i++)
{
memcpy(constants.normalmatrices[i], &xfmem.normalMatrices[3*i], 12);
}
dirty = true;
nNormalMatricesChanged[0] = nNormalMatricesChanged[1] = -1;
}
if (nPostTransformMatricesChanged[0] >= 0)
{
int startn = nPostTransformMatricesChanged[0] / 4;
int endn = (nPostTransformMatricesChanged[1] + 3 ) / 4;
memcpy(constants.posttransformmatrices[startn], &xfmem.postMatrices[startn * 4], (endn - startn) * 16);
dirty = true;
nPostTransformMatricesChanged[0] = nPostTransformMatricesChanged[1] = -1;
}
if (nLightsChanged[0] >= 0)
{
// TODO: Outdated comment
// lights don't have a 1 to 1 mapping, the color component needs to be converted to 4 floats
int istart = nLightsChanged[0] / 0x10;
int iend = (nLightsChanged[1] + 15) / 0x10;
for (int i = istart; i < iend; ++i)
{
const Light& light = xfmem.lights[i];
VertexShaderConstants::Light& dstlight = constants.lights[i];
// xfmem.light.color is packed as abgr in u8[4], so we have to swap the order
dstlight.color[0] = light.color[3];
dstlight.color[1] = light.color[2];
dstlight.color[2] = light.color[1];
dstlight.color[3] = light.color[0];
dstlight.cosatt[0] = light.cosatt[0];
dstlight.cosatt[1] = light.cosatt[1];
dstlight.cosatt[2] = light.cosatt[2];
if (fabs(light.distatt[0]) < 0.00001f &&
fabs(light.distatt[1]) < 0.00001f &&
fabs(light.distatt[2]) < 0.00001f)
{
// dist attenuation, make sure not equal to 0!!!
dstlight.distatt[0] = .00001f;
}
else
{
dstlight.distatt[0] = light.distatt[0];
}
dstlight.distatt[1] = light.distatt[1];
dstlight.distatt[2] = light.distatt[2];
dstlight.pos[0] = light.dpos[0];
dstlight.pos[1] = light.dpos[1];
dstlight.pos[2] = light.dpos[2];
double norm = double(light.ddir[0]) * double(light.ddir[0]) +
double(light.ddir[1]) * double(light.ddir[1]) +
double(light.ddir[2]) * double(light.ddir[2]);
norm = 1.0 / sqrt(norm);
float norm_float = static_cast<float>(norm);
dstlight.dir[0] = light.ddir[0] * norm_float;
dstlight.dir[1] = light.ddir[1] * norm_float;
dstlight.dir[2] = light.ddir[2] * norm_float;
}
dirty = true;
nLightsChanged[0] = nLightsChanged[1] = -1;
}
for (int i : nMaterialsChanged)
{
u32 data = i >= 2 ? xfmem.matColor[i - 2] : xfmem.ambColor[i];
constants.materials[i][0] = (data >> 24) & 0xFF;
constants.materials[i][1] = (data >> 16) & 0xFF;
constants.materials[i][2] = (data >> 8) & 0xFF;
constants.materials[i][3] = data & 0xFF;
dirty = true;
}
nMaterialsChanged = BitSet32(0);
if (bPosNormalMatrixChanged)
{
bPosNormalMatrixChanged = false;
const float *pos = (const float *)xfmem.posMatrices + g_main_cp_state.matrix_index_a.PosNormalMtxIdx * 4;
const float *norm = (const float *)xfmem.normalMatrices + 3 * (g_main_cp_state.matrix_index_a.PosNormalMtxIdx & 31);
memcpy(constants.posnormalmatrix, pos, 3*16);
memcpy(constants.posnormalmatrix[3], norm, 12);
memcpy(constants.posnormalmatrix[4], norm+3, 12);
memcpy(constants.posnormalmatrix[5], norm+6, 12);
dirty = true;
}
if (bTexMatricesChanged[0])
{
bTexMatricesChanged[0] = false;
const float *fptrs[] =
{
(const float *)&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex0MtxIdx * 4],
(const float *)&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex1MtxIdx * 4],
(const float *)&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex2MtxIdx * 4],
(const float *)&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex3MtxIdx * 4]
};
for (int i = 0; i < 4; ++i)
{
memcpy(constants.texmatrices[3*i], fptrs[i], 3*16);
}
dirty = true;
}
if (bTexMatricesChanged[1])
{
bTexMatricesChanged[1] = false;
const float *fptrs[] = {
(const float *)&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex4MtxIdx * 4],
(const float *)&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex5MtxIdx * 4],
(const float *)&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex6MtxIdx * 4],
(const float *)&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex7MtxIdx * 4]
};
for (int i = 0; i < 4; ++i)
{
memcpy(constants.texmatrices[3*i+12], fptrs[i], 3*16);
}
dirty = true;
}
if (bViewportChanged)
{
bViewportChanged = false;
// The console GPU places the pixel center at 7/12 unless antialiasing
// is enabled, while D3D and OpenGL place it at 0.5. See the comment
// in VertexShaderGen.cpp for details.
// NOTE: If we ever emulate antialiasing, the sample locations set by
// BP registers 0x01-0x04 need to be considered here.
const float pixel_center_correction = 7.0f / 12.0f - 0.5f;
const float pixel_size_x = 2.f / Renderer::EFBToScaledXf(2.f * xfmem.viewport.wd);
const float pixel_size_y = 2.f / Renderer::EFBToScaledXf(2.f * xfmem.viewport.ht);
constants.pixelcentercorrection[0] = pixel_center_correction * pixel_size_x;
constants.pixelcentercorrection[1] = pixel_center_correction * pixel_size_y;
dirty = true;
// This is so implementation-dependent that we can't have it here.
g_renderer->SetViewport();
// Update projection if the viewport isn't 1:1 useable
if (!g_ActiveConfig.backend_info.bSupportsOversizedViewports)
{
ViewportCorrectionMatrix(s_viewportCorrection);
bProjectionChanged = true;
}
}
if (bProjectionChanged)
{
bProjectionChanged = false;
float *rawProjection = xfmem.projection.rawProjection;
switch (xfmem.projection.type)
{
case GX_PERSPECTIVE:
g_fProjectionMatrix[0] = rawProjection[0] * g_ActiveConfig.fAspectRatioHackW;
g_fProjectionMatrix[1] = 0.0f;
g_fProjectionMatrix[2] = rawProjection[1];
g_fProjectionMatrix[3] = 0.0f;
g_fProjectionMatrix[4] = 0.0f;
g_fProjectionMatrix[5] = rawProjection[2] * g_ActiveConfig.fAspectRatioHackH;
g_fProjectionMatrix[6] = rawProjection[3];
g_fProjectionMatrix[7] = 0.0f;
g_fProjectionMatrix[8] = 0.0f;
g_fProjectionMatrix[9] = 0.0f;
g_fProjectionMatrix[10] = rawProjection[4];
g_fProjectionMatrix[11] = rawProjection[5];
g_fProjectionMatrix[12] = 0.0f;
g_fProjectionMatrix[13] = 0.0f;
// donkopunchstania suggested the GC GPU might round differently
// He had thus changed this to -(1 + epsilon) to fix clipping issues.
// I (neobrain) don't think his conjecture is true and thus reverted his change.
g_fProjectionMatrix[14] = -1.0f;
g_fProjectionMatrix[15] = 0.0f;
SETSTAT_FT(stats.gproj_0, g_fProjectionMatrix[0]);
SETSTAT_FT(stats.gproj_1, g_fProjectionMatrix[1]);
SETSTAT_FT(stats.gproj_2, g_fProjectionMatrix[2]);
SETSTAT_FT(stats.gproj_3, g_fProjectionMatrix[3]);
SETSTAT_FT(stats.gproj_4, g_fProjectionMatrix[4]);
SETSTAT_FT(stats.gproj_5, g_fProjectionMatrix[5]);
SETSTAT_FT(stats.gproj_6, g_fProjectionMatrix[6]);
SETSTAT_FT(stats.gproj_7, g_fProjectionMatrix[7]);
SETSTAT_FT(stats.gproj_8, g_fProjectionMatrix[8]);
SETSTAT_FT(stats.gproj_9, g_fProjectionMatrix[9]);
SETSTAT_FT(stats.gproj_10, g_fProjectionMatrix[10]);
SETSTAT_FT(stats.gproj_11, g_fProjectionMatrix[11]);
SETSTAT_FT(stats.gproj_12, g_fProjectionMatrix[12]);
SETSTAT_FT(stats.gproj_13, g_fProjectionMatrix[13]);
SETSTAT_FT(stats.gproj_14, g_fProjectionMatrix[14]);
SETSTAT_FT(stats.gproj_15, g_fProjectionMatrix[15]);
break;
case GX_ORTHOGRAPHIC:
g_fProjectionMatrix[0] = rawProjection[0];
g_fProjectionMatrix[1] = 0.0f;
g_fProjectionMatrix[2] = 0.0f;
g_fProjectionMatrix[3] = rawProjection[1];
g_fProjectionMatrix[4] = 0.0f;
g_fProjectionMatrix[5] = rawProjection[2];
g_fProjectionMatrix[6] = 0.0f;
g_fProjectionMatrix[7] = rawProjection[3];
g_fProjectionMatrix[8] = 0.0f;
g_fProjectionMatrix[9] = 0.0f;
g_fProjectionMatrix[10] = (g_ProjHack1.value + rawProjection[4]) * ((g_ProjHack1.sign == 0) ? 1.0f : g_ProjHack1.sign);
g_fProjectionMatrix[11] = (g_ProjHack2.value + rawProjection[5]) * ((g_ProjHack2.sign == 0) ? 1.0f : g_ProjHack2.sign);
g_fProjectionMatrix[12] = 0.0f;
g_fProjectionMatrix[13] = 0.0f;
g_fProjectionMatrix[14] = 0.0f;
g_fProjectionMatrix[15] = 1.0f + FLT_EPSILON; // hack to fix depth clipping precision issues (such as Sonic Unleashed UI)
SETSTAT_FT(stats.g2proj_0, g_fProjectionMatrix[0]);
SETSTAT_FT(stats.g2proj_1, g_fProjectionMatrix[1]);
SETSTAT_FT(stats.g2proj_2, g_fProjectionMatrix[2]);
SETSTAT_FT(stats.g2proj_3, g_fProjectionMatrix[3]);
SETSTAT_FT(stats.g2proj_4, g_fProjectionMatrix[4]);
SETSTAT_FT(stats.g2proj_5, g_fProjectionMatrix[5]);
SETSTAT_FT(stats.g2proj_6, g_fProjectionMatrix[6]);
SETSTAT_FT(stats.g2proj_7, g_fProjectionMatrix[7]);
SETSTAT_FT(stats.g2proj_8, g_fProjectionMatrix[8]);
SETSTAT_FT(stats.g2proj_9, g_fProjectionMatrix[9]);
SETSTAT_FT(stats.g2proj_10, g_fProjectionMatrix[10]);
SETSTAT_FT(stats.g2proj_11, g_fProjectionMatrix[11]);
SETSTAT_FT(stats.g2proj_12, g_fProjectionMatrix[12]);
SETSTAT_FT(stats.g2proj_13, g_fProjectionMatrix[13]);
SETSTAT_FT(stats.g2proj_14, g_fProjectionMatrix[14]);
SETSTAT_FT(stats.g2proj_15, g_fProjectionMatrix[15]);
SETSTAT_FT(stats.proj_0, rawProjection[0]);
SETSTAT_FT(stats.proj_1, rawProjection[1]);
SETSTAT_FT(stats.proj_2, rawProjection[2]);
SETSTAT_FT(stats.proj_3, rawProjection[3]);
SETSTAT_FT(stats.proj_4, rawProjection[4]);
SETSTAT_FT(stats.proj_5, rawProjection[5]);
break;
default:
ERROR_LOG(VIDEO, "Unknown projection type: %d", xfmem.projection.type);
}
PRIM_LOG("Projection: %f %f %f %f %f %f\n", rawProjection[0], rawProjection[1], rawProjection[2], rawProjection[3], rawProjection[4], rawProjection[5]);
if (g_ActiveConfig.bFreeLook && xfmem.projection.type == GX_PERSPECTIVE)
{
Matrix44 mtxA;
Matrix44 mtxB;
Matrix44 viewMtx;
Matrix44::Translate(mtxA, s_fViewTranslationVector);
Matrix44::LoadMatrix33(mtxB, s_viewRotationMatrix);
Matrix44::Multiply(mtxB, mtxA, viewMtx); // view = rotation x translation
Matrix44::Set(mtxB, g_fProjectionMatrix);
Matrix44::Multiply(mtxB, viewMtx, mtxA); // mtxA = projection x view
Matrix44::Multiply(s_viewportCorrection, mtxA, mtxB); // mtxB = viewportCorrection x mtxA
memcpy(constants.projection, mtxB.data, 4*16);
}
else
{
Matrix44 projMtx;
Matrix44::Set(projMtx, g_fProjectionMatrix);
Matrix44 correctedMtx;
Matrix44::Multiply(s_viewportCorrection, projMtx, correctedMtx);
memcpy(constants.projection, correctedMtx.data, 4*16);
}
dirty = true;
}
}
// Syncs the shader constant buffers with xfmem
// TODO: A cleaner way to control the matrices without making a mess in the parameters field
void VertexShaderManager::SetConstants()
{
if (nTransformMatricesChanged[0] >= 0)
{
int startn = nTransformMatricesChanged[0] / 4;
int endn = (nTransformMatricesChanged[1] + 3) / 4;
memcpy(constants.transformmatrices[startn], &xfmem.posMatrices[startn * 4],
(endn - startn) * sizeof(float4));
dirty = true;
nTransformMatricesChanged[0] = nTransformMatricesChanged[1] = -1;
}
if (nNormalMatricesChanged[0] >= 0)
{
int startn = nNormalMatricesChanged[0] / 3;
int endn = (nNormalMatricesChanged[1] + 2) / 3;
for (int i = startn; i < endn; i++)
{
memcpy(constants.normalmatrices[i], &xfmem.normalMatrices[3 * i], 12);
}
dirty = true;
nNormalMatricesChanged[0] = nNormalMatricesChanged[1] = -1;
}
if (nPostTransformMatricesChanged[0] >= 0)
{
int startn = nPostTransformMatricesChanged[0] / 4;
int endn = (nPostTransformMatricesChanged[1] + 3) / 4;
memcpy(constants.posttransformmatrices[startn], &xfmem.postMatrices[startn * 4],
(endn - startn) * sizeof(float4));
dirty = true;
nPostTransformMatricesChanged[0] = nPostTransformMatricesChanged[1] = -1;
}
if (nLightsChanged[0] >= 0)
{
// TODO: Outdated comment
// lights don't have a 1 to 1 mapping, the color component needs to be converted to 4 floats
int istart = nLightsChanged[0] / 0x10;
int iend = (nLightsChanged[1] + 15) / 0x10;
for (int i = istart; i < iend; ++i)
{
const Light& light = xfmem.lights[i];
VertexShaderConstants::Light& dstlight = constants.lights[i];
// xfmem.light.color is packed as abgr in u8[4], so we have to swap the order
dstlight.color[0] = light.color[3];
dstlight.color[1] = light.color[2];
dstlight.color[2] = light.color[1];
dstlight.color[3] = light.color[0];
dstlight.cosatt[0] = light.cosatt[0];
dstlight.cosatt[1] = light.cosatt[1];
dstlight.cosatt[2] = light.cosatt[2];
if (fabs(light.distatt[0]) < 0.00001f && fabs(light.distatt[1]) < 0.00001f &&
fabs(light.distatt[2]) < 0.00001f)
{
// dist attenuation, make sure not equal to 0!!!
dstlight.distatt[0] = .00001f;
}
else
{
dstlight.distatt[0] = light.distatt[0];
}
dstlight.distatt[1] = light.distatt[1];
dstlight.distatt[2] = light.distatt[2];
dstlight.pos[0] = light.dpos[0];
dstlight.pos[1] = light.dpos[1];
dstlight.pos[2] = light.dpos[2];
double norm = double(light.ddir[0]) * double(light.ddir[0]) +
double(light.ddir[1]) * double(light.ddir[1]) +
double(light.ddir[2]) * double(light.ddir[2]);
norm = 1.0 / sqrt(norm);
float norm_float = static_cast<float>(norm);
dstlight.dir[0] = light.ddir[0] * norm_float;
dstlight.dir[1] = light.ddir[1] * norm_float;
dstlight.dir[2] = light.ddir[2] * norm_float;
}
dirty = true;
nLightsChanged[0] = nLightsChanged[1] = -1;
}
for (int i : nMaterialsChanged)
{
u32 data = i >= 2 ? xfmem.matColor[i - 2] : xfmem.ambColor[i];
constants.materials[i][0] = (data >> 24) & 0xFF;
constants.materials[i][1] = (data >> 16) & 0xFF;
constants.materials[i][2] = (data >> 8) & 0xFF;
constants.materials[i][3] = data & 0xFF;
dirty = true;
}
nMaterialsChanged = BitSet32(0);
if (bPosNormalMatrixChanged)
{
bPosNormalMatrixChanged = false;
const float* pos = &xfmem.posMatrices[g_main_cp_state.matrix_index_a.PosNormalMtxIdx * 4];
const float* norm =
&xfmem.normalMatrices[3 * (g_main_cp_state.matrix_index_a.PosNormalMtxIdx & 31)];
memcpy(constants.posnormalmatrix, pos, 3 * sizeof(float4));
memcpy(constants.posnormalmatrix[3], norm, 3 * sizeof(float));
memcpy(constants.posnormalmatrix[4], norm + 3, 3 * sizeof(float));
memcpy(constants.posnormalmatrix[5], norm + 6, 3 * sizeof(float));
dirty = true;
}
if (bTexMatricesChanged[0])
{
bTexMatricesChanged[0] = false;
const float* pos_matrix_ptrs[] = {
&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex0MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex1MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex2MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_a.Tex3MtxIdx * 4]};
for (size_t i = 0; i < ArraySize(pos_matrix_ptrs); ++i)
{
memcpy(constants.texmatrices[3 * i], pos_matrix_ptrs[i], 3 * sizeof(float4));
}
dirty = true;
}
if (bTexMatricesChanged[1])
{
bTexMatricesChanged[1] = false;
const float* pos_matrix_ptrs[] = {
&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex4MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex5MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex6MtxIdx * 4],
&xfmem.posMatrices[g_main_cp_state.matrix_index_b.Tex7MtxIdx * 4]};
for (size_t i = 0; i < ArraySize(pos_matrix_ptrs); ++i)
{
memcpy(constants.texmatrices[3 * i + 12], pos_matrix_ptrs[i], 3 * sizeof(float4));
}
dirty = true;
}
if (bViewportChanged)
{
bViewportChanged = false;
// The console GPU places the pixel center at 7/12 unless antialiasing
// is enabled, while D3D and OpenGL place it at 0.5. See the comment
// in VertexShaderGen.cpp for details.
// NOTE: If we ever emulate antialiasing, the sample locations set by
// BP registers 0x01-0x04 need to be considered here.
const float pixel_center_correction = 7.0f / 12.0f - 0.5f;
const float pixel_size_x = 2.f / Renderer::EFBToScaledXf(2.f * xfmem.viewport.wd);
const float pixel_size_y = 2.f / Renderer::EFBToScaledXf(2.f * xfmem.viewport.ht);
constants.pixelcentercorrection[0] = pixel_center_correction * pixel_size_x;
constants.pixelcentercorrection[1] = pixel_center_correction * pixel_size_y;
// By default we don't change the depth value at all in the vertex shader.
constants.pixelcentercorrection[2] = 1.0f;
constants.pixelcentercorrection[3] = 0.0f;
if (g_ActiveConfig.backend_info.bSupportsDepthClamp)
{
// Oversized depth ranges are handled in the vertex shader. We need to reverse
// the far value to get a reversed depth range mapping. This is necessary
// because the standard depth range equation pushes all depth values towards
// the back of the depth buffer where conventionally depth buffers have the
// least precision.
if (g_ActiveConfig.backend_info.bSupportsReversedDepthRange)
{
if (fabs(xfmem.viewport.zRange) > 16777215.0f || fabs(xfmem.viewport.farZ) > 16777215.0f)
{
// For backends that support reversing the depth range we also support cases
// where the console also uses reversed depth with the same accuracy. We need
// to make sure the depth range is positive here and then reverse the depth in
// the backend viewport.
constants.pixelcentercorrection[2] = fabs(xfmem.viewport.zRange) / 16777215.0f;
if (xfmem.viewport.zRange < 0.0f)
constants.pixelcentercorrection[3] = xfmem.viewport.farZ / 16777215.0f;
else
constants.pixelcentercorrection[3] = 1.0f - xfmem.viewport.farZ / 16777215.0f;
}
}
else
{
if (xfmem.viewport.zRange < 0.0f || xfmem.viewport.zRange > 16777215.0f ||
fabs(xfmem.viewport.farZ) > 16777215.0f)
{
// For backends that don't support reversing the depth range we can still render
// cases where the console uses reversed depth correctly. But we simply can't
// provide the same accuracy as the console.
constants.pixelcentercorrection[2] = xfmem.viewport.zRange / 16777215.0f;
constants.pixelcentercorrection[3] = 1.0f - xfmem.viewport.farZ / 16777215.0f;
}
}
}
dirty = true;
// This is so implementation-dependent that we can't have it here.
g_renderer->SetViewport();
// Update projection if the viewport isn't 1:1 useable
if (!g_ActiveConfig.backend_info.bSupportsOversizedViewports)
{
ViewportCorrectionMatrix(s_viewportCorrection);
bProjectionChanged = true;
}
}
if (bProjectionChanged)
{
bProjectionChanged = false;
float* rawProjection = xfmem.projection.rawProjection;
switch (xfmem.projection.type)
{
case GX_PERSPECTIVE:
g_fProjectionMatrix[0] = rawProjection[0] * g_ActiveConfig.fAspectRatioHackW;
g_fProjectionMatrix[1] = 0.0f;
g_fProjectionMatrix[2] = rawProjection[1];
g_fProjectionMatrix[3] = 0.0f;
g_fProjectionMatrix[4] = 0.0f;
g_fProjectionMatrix[5] = rawProjection[2] * g_ActiveConfig.fAspectRatioHackH;
g_fProjectionMatrix[6] = rawProjection[3];
g_fProjectionMatrix[7] = 0.0f;
g_fProjectionMatrix[8] = 0.0f;
g_fProjectionMatrix[9] = 0.0f;
g_fProjectionMatrix[10] = rawProjection[4];
g_fProjectionMatrix[11] = rawProjection[5];
g_fProjectionMatrix[12] = 0.0f;
g_fProjectionMatrix[13] = 0.0f;
g_fProjectionMatrix[14] = -1.0f;
g_fProjectionMatrix[15] = 0.0f;
// Heuristic to detect if a GameCube game is in 16:9 anamorphic widescreen mode.
if (!SConfig::GetInstance().bWii)
{
bool viewport_is_4_3 = AspectIs4_3(xfmem.viewport.wd, xfmem.viewport.ht);
if (AspectIs16_9(rawProjection[2], rawProjection[0]) && viewport_is_4_3)
Core::g_aspect_wide = true; // Projection is 16:9 and viewport is 4:3, we are rendering
// an anamorphic widescreen picture
else if (AspectIs4_3(rawProjection[2], rawProjection[0]) && viewport_is_4_3)
Core::g_aspect_wide =
false; // Project and viewports are both 4:3, we are rendering a normal image.
}
SETSTAT_FT(stats.gproj_0, g_fProjectionMatrix[0]);
SETSTAT_FT(stats.gproj_1, g_fProjectionMatrix[1]);
SETSTAT_FT(stats.gproj_2, g_fProjectionMatrix[2]);
SETSTAT_FT(stats.gproj_3, g_fProjectionMatrix[3]);
SETSTAT_FT(stats.gproj_4, g_fProjectionMatrix[4]);
SETSTAT_FT(stats.gproj_5, g_fProjectionMatrix[5]);
SETSTAT_FT(stats.gproj_6, g_fProjectionMatrix[6]);
SETSTAT_FT(stats.gproj_7, g_fProjectionMatrix[7]);
SETSTAT_FT(stats.gproj_8, g_fProjectionMatrix[8]);
SETSTAT_FT(stats.gproj_9, g_fProjectionMatrix[9]);
SETSTAT_FT(stats.gproj_10, g_fProjectionMatrix[10]);
SETSTAT_FT(stats.gproj_11, g_fProjectionMatrix[11]);
SETSTAT_FT(stats.gproj_12, g_fProjectionMatrix[12]);
SETSTAT_FT(stats.gproj_13, g_fProjectionMatrix[13]);
SETSTAT_FT(stats.gproj_14, g_fProjectionMatrix[14]);
SETSTAT_FT(stats.gproj_15, g_fProjectionMatrix[15]);
break;
case GX_ORTHOGRAPHIC:
g_fProjectionMatrix[0] = rawProjection[0];
g_fProjectionMatrix[1] = 0.0f;
g_fProjectionMatrix[2] = 0.0f;
g_fProjectionMatrix[3] = rawProjection[1];
g_fProjectionMatrix[4] = 0.0f;
g_fProjectionMatrix[5] = rawProjection[2];
g_fProjectionMatrix[6] = 0.0f;
g_fProjectionMatrix[7] = rawProjection[3];
g_fProjectionMatrix[8] = 0.0f;
g_fProjectionMatrix[9] = 0.0f;
g_fProjectionMatrix[10] = (g_ProjHack1.value + rawProjection[4]) *
((g_ProjHack1.sign == 0) ? 1.0f : g_ProjHack1.sign);
g_fProjectionMatrix[11] = (g_ProjHack2.value + rawProjection[5]) *
((g_ProjHack2.sign == 0) ? 1.0f : g_ProjHack2.sign);
g_fProjectionMatrix[12] = 0.0f;
g_fProjectionMatrix[13] = 0.0f;
g_fProjectionMatrix[14] = 0.0f;
g_fProjectionMatrix[15] = 1.0f;
SETSTAT_FT(stats.g2proj_0, g_fProjectionMatrix[0]);
SETSTAT_FT(stats.g2proj_1, g_fProjectionMatrix[1]);
SETSTAT_FT(stats.g2proj_2, g_fProjectionMatrix[2]);
SETSTAT_FT(stats.g2proj_3, g_fProjectionMatrix[3]);
SETSTAT_FT(stats.g2proj_4, g_fProjectionMatrix[4]);
SETSTAT_FT(stats.g2proj_5, g_fProjectionMatrix[5]);
SETSTAT_FT(stats.g2proj_6, g_fProjectionMatrix[6]);
SETSTAT_FT(stats.g2proj_7, g_fProjectionMatrix[7]);
SETSTAT_FT(stats.g2proj_8, g_fProjectionMatrix[8]);
SETSTAT_FT(stats.g2proj_9, g_fProjectionMatrix[9]);
SETSTAT_FT(stats.g2proj_10, g_fProjectionMatrix[10]);
SETSTAT_FT(stats.g2proj_11, g_fProjectionMatrix[11]);
SETSTAT_FT(stats.g2proj_12, g_fProjectionMatrix[12]);
SETSTAT_FT(stats.g2proj_13, g_fProjectionMatrix[13]);
SETSTAT_FT(stats.g2proj_14, g_fProjectionMatrix[14]);
SETSTAT_FT(stats.g2proj_15, g_fProjectionMatrix[15]);
SETSTAT_FT(stats.proj_0, rawProjection[0]);
SETSTAT_FT(stats.proj_1, rawProjection[1]);
SETSTAT_FT(stats.proj_2, rawProjection[2]);
SETSTAT_FT(stats.proj_3, rawProjection[3]);
SETSTAT_FT(stats.proj_4, rawProjection[4]);
SETSTAT_FT(stats.proj_5, rawProjection[5]);
break;
default:
ERROR_LOG(VIDEO, "Unknown projection type: %d", xfmem.projection.type);
}
PRIM_LOG("Projection: %f %f %f %f %f %f", rawProjection[0], rawProjection[1], rawProjection[2],
rawProjection[3], rawProjection[4], rawProjection[5]);
if (g_ActiveConfig.bFreeLook && xfmem.projection.type == GX_PERSPECTIVE)
{
Matrix44 mtxA;
Matrix44 mtxB;
Matrix44 viewMtx;
Matrix44::Translate(mtxA, s_fViewTranslationVector);
Matrix44::LoadMatrix33(mtxB, s_viewRotationMatrix);
Matrix44::Multiply(mtxB, mtxA, viewMtx); // view = rotation x translation
Matrix44::Set(mtxB, g_fProjectionMatrix);
Matrix44::Multiply(mtxB, viewMtx, mtxA); // mtxA = projection x view
Matrix44::Multiply(s_viewportCorrection, mtxA, mtxB); // mtxB = viewportCorrection x mtxA
memcpy(constants.projection, mtxB.data, 4 * sizeof(float4));
}
else
{
Matrix44 projMtx;
Matrix44::Set(projMtx, g_fProjectionMatrix);
Matrix44 correctedMtx;
Matrix44::Multiply(s_viewportCorrection, projMtx, correctedMtx);
memcpy(constants.projection, correctedMtx.data, 4 * sizeof(float4));
}
dirty = true;
}
}