Pica: Implement W-Buffer in SW rasterizer

This commit is contained in:
Jannik Vogel 2016-03-20 01:53:49 +01:00
parent 006fe5fc0f
commit 4c98113b57
4 changed files with 43 additions and 11 deletions

View file

@ -75,8 +75,6 @@ static void InitScreenCoordinates(OutputVertex& vtx)
viewport.halfsize_y = float24::FromRaw(regs.viewport_size_y); viewport.halfsize_y = float24::FromRaw(regs.viewport_size_y);
viewport.offset_x = float24::FromFloat32(static_cast<float>(regs.viewport_corner.x)); viewport.offset_x = float24::FromFloat32(static_cast<float>(regs.viewport_corner.x));
viewport.offset_y = float24::FromFloat32(static_cast<float>(regs.viewport_corner.y)); viewport.offset_y = float24::FromFloat32(static_cast<float>(regs.viewport_corner.y));
viewport.zscale = float24::FromRaw(regs.viewport_depth_range);
viewport.offset_z = float24::FromRaw(regs.viewport_depth_far_plane);
float24 inv_w = float24::FromFloat32(1.f) / vtx.pos.w; float24 inv_w = float24::FromFloat32(1.f) / vtx.pos.w;
vtx.color *= inv_w; vtx.color *= inv_w;
@ -89,7 +87,7 @@ static void InitScreenCoordinates(OutputVertex& vtx)
vtx.screenpos[0] = (vtx.pos.x * inv_w + float24::FromFloat32(1.0)) * viewport.halfsize_x + viewport.offset_x; vtx.screenpos[0] = (vtx.pos.x * inv_w + float24::FromFloat32(1.0)) * viewport.halfsize_x + viewport.offset_x;
vtx.screenpos[1] = (vtx.pos.y * inv_w + float24::FromFloat32(1.0)) * viewport.halfsize_y + viewport.offset_y; vtx.screenpos[1] = (vtx.pos.y * inv_w + float24::FromFloat32(1.0)) * viewport.halfsize_y + viewport.offset_y;
vtx.screenpos[2] = viewport.offset_z + vtx.pos.z * inv_w * viewport.zscale; vtx.screenpos[2] = vtx.pos.z * inv_w;
} }
void ProcessTriangle(const OutputVertex &v0, const OutputVertex &v1, const OutputVertex &v2) { void ProcessTriangle(const OutputVertex &v0, const OutputVertex &v1, const OutputVertex &v2) {

View file

@ -70,7 +70,7 @@ struct Regs {
INSERT_PADDING_WORDS(0x9); INSERT_PADDING_WORDS(0x9);
BitField<0, 24, u32> viewport_depth_range; // float24 BitField<0, 24, u32> viewport_depth_range; // float24
BitField<0, 24, u32> viewport_depth_far_plane; // float24 BitField<0, 24, u32> viewport_depth_near_plane; // float24
BitField<0, 3, u32> vs_output_total; BitField<0, 3, u32> vs_output_total;
@ -122,7 +122,20 @@ struct Regs {
BitField<16, 10, s32> y; BitField<16, 10, s32> y;
} viewport_corner; } viewport_corner;
INSERT_PADDING_WORDS(0x17); INSERT_PADDING_WORDS(0x1);
//TODO: early depth
INSERT_PADDING_WORDS(0x1);
INSERT_PADDING_WORDS(0x2);
enum DepthBuffering : u32 {
WBuffering = 0,
ZBuffering = 1,
};
BitField< 0, 1, DepthBuffering> depthmap_enable;
INSERT_PADDING_WORDS(0x12);
struct TextureConfig { struct TextureConfig {
enum WrapMode : u32 { enum WrapMode : u32 {
@ -1279,10 +1292,11 @@ ASSERT_REG_POSITION(cull_mode, 0x40);
ASSERT_REG_POSITION(viewport_size_x, 0x41); ASSERT_REG_POSITION(viewport_size_x, 0x41);
ASSERT_REG_POSITION(viewport_size_y, 0x43); ASSERT_REG_POSITION(viewport_size_y, 0x43);
ASSERT_REG_POSITION(viewport_depth_range, 0x4d); ASSERT_REG_POSITION(viewport_depth_range, 0x4d);
ASSERT_REG_POSITION(viewport_depth_far_plane, 0x4e); ASSERT_REG_POSITION(viewport_depth_near_plane, 0x4e);
ASSERT_REG_POSITION(vs_output_attributes[0], 0x50); ASSERT_REG_POSITION(vs_output_attributes[0], 0x50);
ASSERT_REG_POSITION(vs_output_attributes[1], 0x51); ASSERT_REG_POSITION(vs_output_attributes[1], 0x51);
ASSERT_REG_POSITION(viewport_corner, 0x68); ASSERT_REG_POSITION(viewport_corner, 0x68);
ASSERT_REG_POSITION(depthmap_enable, 0x6D);
ASSERT_REG_POSITION(texture0_enable, 0x80); ASSERT_REG_POSITION(texture0_enable, 0x80);
ASSERT_REG_POSITION(texture0, 0x81); ASSERT_REG_POSITION(texture0, 0x81);
ASSERT_REG_POSITION(texture0_format, 0x8e); ASSERT_REG_POSITION(texture0_format, 0x8e);

View file

@ -862,10 +862,30 @@ static void ProcessTriangleInternal(const Shader::OutputVertex& v0,
} }
} }
unsigned num_bits = Regs::DepthBitsPerPixel(regs.framebuffer.depth_format); // interpolated_z = z / w
u32 z = (u32)((v0.screenpos[2].ToFloat32() * w0 + float interpolated_z_over_w = (v0.screenpos[2].ToFloat32() * w0 +
v1.screenpos[2].ToFloat32() * w1 + v1.screenpos[2].ToFloat32() * w1 +
v2.screenpos[2].ToFloat32() * w2) * ((1 << num_bits) - 1) / wsum); v2.screenpos[2].ToFloat32() * w2) / wsum;
// Not fully accurate. About 3 bits in precision are missing.
// Z-Buffer (z / w * scale + offset)
float depth_scale = float24::FromRaw(regs.viewport_depth_range).ToFloat32();
float depth_offset = float24::FromRaw(regs.viewport_depth_near_plane).ToFloat32();
float depth = interpolated_z_over_w * depth_scale + depth_offset;
// Potentially switch to W-Buffer
if (regs.depthmap_enable == Pica::Regs::DepthBuffering::WBuffering) {
// W-Buffer (z * scale + w * offset = (z / w * scale + offset) * w)
depth *= interpolated_w_inverse.ToFloat32() * wsum;
}
// Clamp the result
depth = MathUtil::Clamp(depth, 0.0f, 1.0f);
// Convert float to integer
unsigned num_bits = Regs::DepthBitsPerPixel(regs.framebuffer.depth_format);
u32 z = (u32)(depth * ((1 << num_bits) - 1));
if (output_merger.depth_test_enable) { if (output_merger.depth_test_enable) {
u32 ref_z = GetDepth(x >> 4, y >> 4); u32 ref_z = GetDepth(x >> 4, y >> 4);

View file

@ -256,7 +256,7 @@ void RasterizerOpenGL::NotifyPicaRegisterChanged(u32 id) {
// Depth modifiers // Depth modifiers
case PICA_REG_INDEX(viewport_depth_range): case PICA_REG_INDEX(viewport_depth_range):
case PICA_REG_INDEX(viewport_depth_far_plane): case PICA_REG_INDEX(viewport_depth_near_plane):
SyncDepthModifiers(); SyncDepthModifiers();
break; break;
@ -911,7 +911,7 @@ void RasterizerOpenGL::SyncCullMode() {
void RasterizerOpenGL::SyncDepthModifiers() { void RasterizerOpenGL::SyncDepthModifiers() {
float depth_scale = -Pica::float24::FromRaw(Pica::g_state.regs.viewport_depth_range).ToFloat32(); float depth_scale = -Pica::float24::FromRaw(Pica::g_state.regs.viewport_depth_range).ToFloat32();
float depth_offset = Pica::float24::FromRaw(Pica::g_state.regs.viewport_depth_far_plane).ToFloat32() / 2.0f; float depth_offset = Pica::float24::FromRaw(Pica::g_state.regs.viewport_depth_near_plane).ToFloat32() / 2.0f;
// TODO: Implement scale modifier // TODO: Implement scale modifier
uniform_block_data.data.depth_offset = depth_offset; uniform_block_data.data.depth_offset = depth_offset;