aboutsummaryrefslogtreecommitdiff
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/CMakeLists.txt1
-rw-r--r--src/video_core/clipper.cpp84
-rw-r--r--src/video_core/color.h132
-rw-r--r--src/video_core/command_processor.cpp36
-rw-r--r--src/video_core/debug_utils/debug_utils.cpp88
-rw-r--r--src/video_core/gpu_debugger.h4
-rw-r--r--src/video_core/pica.h42
-rw-r--r--src/video_core/rasterizer.cpp328
-rw-r--r--src/video_core/renderer_opengl/renderer_opengl.cpp60
-rw-r--r--src/video_core/utils.h50
-rw-r--r--src/video_core/vertex_shader.cpp53
-rw-r--r--src/video_core/video_core.cpp3
-rw-r--r--src/video_core/video_core.h1
13 files changed, 561 insertions, 321 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index 13c3f7b22..4c1e6449a 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -19,6 +19,7 @@ set(HEADERS
renderer_opengl/gl_shaders.h
renderer_opengl/renderer_opengl.h
clipper.h
+ color.h
command_processor.h
gpu_debugger.h
math.h
diff --git a/src/video_core/clipper.cpp b/src/video_core/clipper.cpp
index 1744066ba..ba3876a76 100644
--- a/src/video_core/clipper.cpp
+++ b/src/video_core/clipper.cpp
@@ -15,30 +15,18 @@ namespace Clipper {
struct ClippingEdge {
public:
- enum Type {
- POS_X = 0,
- NEG_X = 1,
- POS_Y = 2,
- NEG_Y = 3,
- POS_Z = 4,
- NEG_Z = 5,
- };
-
- ClippingEdge(Type type, float24 position) : type(type), pos(position) {}
+ ClippingEdge(Math::Vec4<float24> coeffs,
+ Math::Vec4<float24> bias = Math::Vec4<float24>(float24::FromFloat32(0),
+ float24::FromFloat32(0),
+ float24::FromFloat32(0),
+ float24::FromFloat32(0)))
+ : coeffs(coeffs),
+ bias(bias)
+ {
+ }
bool IsInside(const OutputVertex& vertex) const {
- switch (type) {
- case POS_X: return vertex.pos.x <= pos * vertex.pos.w;
- case NEG_X: return vertex.pos.x >= pos * vertex.pos.w;
- case POS_Y: return vertex.pos.y <= pos * vertex.pos.w;
- case NEG_Y: return vertex.pos.y >= pos * vertex.pos.w;
-
- // TODO: Check z compares ... should be 0..1 instead?
- case POS_Z: return vertex.pos.z <= pos * vertex.pos.w;
-
- default:
- case NEG_Z: return vertex.pos.z >= pos * vertex.pos.w;
- }
+ return Math::Dot(vertex.pos + bias, coeffs) <= float24::FromFloat32(0);
}
bool IsOutSide(const OutputVertex& vertex) const {
@@ -46,31 +34,17 @@ public:
}
OutputVertex GetIntersection(const OutputVertex& v0, const OutputVertex& v1) const {
- auto dotpr = [this](const OutputVertex& vtx) {
- switch (type) {
- case POS_X: return vtx.pos.x - vtx.pos.w;
- case NEG_X: return -vtx.pos.x - vtx.pos.w;
- case POS_Y: return vtx.pos.y - vtx.pos.w;
- case NEG_Y: return -vtx.pos.y - vtx.pos.w;
-
- // TODO: Verify z clipping
- case POS_Z: return vtx.pos.z - vtx.pos.w;
-
- default:
- case NEG_Z: return -vtx.pos.w;
- }
- };
-
- float24 dp = dotpr(v0);
- float24 dp_prev = dotpr(v1);
+ float24 dp = Math::Dot(v0.pos + bias, coeffs);
+ float24 dp_prev = Math::Dot(v1.pos + bias, coeffs);
float24 factor = dp_prev / (dp_prev - dp);
return OutputVertex::Lerp(factor, v0, v1);
}
private:
- Type type;
float24 pos;
+ Math::Vec4<float24> coeffs;
+ Math::Vec4<float24> bias;
};
static void InitScreenCoordinates(OutputVertex& vtx)
@@ -98,10 +72,9 @@ static void InitScreenCoordinates(OutputVertex& vtx)
vtx.tc2 *= inv_w;
vtx.pos.w = inv_w;
- // TODO: Not sure why the viewport width needs to be divided by 2 but the viewport height does not
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[2] = viewport.offset_z - vtx.pos.z * inv_w * viewport.zscale;
+ vtx.screenpos[2] = viewport.offset_z + vtx.pos.z * inv_w * viewport.zscale;
}
void ProcessTriangle(OutputVertex &v0, OutputVertex &v1, OutputVertex &v2) {
@@ -117,14 +90,29 @@ void ProcessTriangle(OutputVertex &v0, OutputVertex &v1, OutputVertex &v2) {
auto* output_list = &buffer_a;
auto* input_list = &buffer_b;
+ // NOTE: We clip against a w=epsilon plane to guarantee that the output has a positive w value.
+ // TODO: Not sure if this is a valid approach. Also should probably instead use the smallest
+ // epsilon possible within float24 accuracy.
+ static const float24 EPSILON = float24::FromFloat32(0.00001);
+ static const float24 f0 = float24::FromFloat32(0.0);
+ static const float24 f1 = float24::FromFloat32(1.0);
+ static const std::array<ClippingEdge, 7> clipping_edges = {{
+ { Math::MakeVec( f1, f0, f0, -f1) }, // x = +w
+ { Math::MakeVec(-f1, f0, f0, -f1) }, // x = -w
+ { Math::MakeVec( f0, f1, f0, -f1) }, // y = +w
+ { Math::MakeVec( f0, -f1, f0, -f1) }, // y = -w
+ { Math::MakeVec( f0, f0, f1, f0) }, // z = 0
+ { Math::MakeVec( f0, f0, -f1, -f1) }, // z = -w
+ { Math::MakeVec( f0, f0, f0, -f1), Math::Vec4<float24>(f0, f0, f0, EPSILON) }, // w = EPSILON
+ }};
+
+ // TODO: If one vertex lies outside one of the depth clipping planes, some platforms (e.g. Wii)
+ // drop the whole primitive instead of clipping the primitive properly. We should test if
+ // this happens on the 3DS, too.
+
// Simple implementation of the Sutherland-Hodgman clipping algorithm.
// TODO: Make this less inefficient (currently lots of useless buffering overhead happens here)
- for (auto edge : { ClippingEdge(ClippingEdge::POS_X, float24::FromFloat32(+1.0)),
- ClippingEdge(ClippingEdge::NEG_X, float24::FromFloat32(-1.0)),
- ClippingEdge(ClippingEdge::POS_Y, float24::FromFloat32(+1.0)),
- ClippingEdge(ClippingEdge::NEG_Y, float24::FromFloat32(-1.0)),
- ClippingEdge(ClippingEdge::POS_Z, float24::FromFloat32(+1.0)),
- ClippingEdge(ClippingEdge::NEG_Z, float24::FromFloat32(-1.0)) }) {
+ for (auto edge : clipping_edges) {
std::swap(input_list, output_list);
output_list->clear();
diff --git a/src/video_core/color.h b/src/video_core/color.h
index e86ac1265..35da901f2 100644
--- a/src/video_core/color.h
+++ b/src/video_core/color.h
@@ -5,28 +5,152 @@
#pragma once
#include "common/common_types.h"
+#include "video_core/math.h"
namespace Color {
/// Convert a 1-bit color component to 8 bit
-static inline u8 Convert1To8(u8 value) {
+inline u8 Convert1To8(u8 value) {
return value * 255;
}
/// Convert a 4-bit color component to 8 bit
-static inline u8 Convert4To8(u8 value) {
+inline u8 Convert4To8(u8 value) {
return (value << 4) | value;
}
/// Convert a 5-bit color component to 8 bit
-static inline u8 Convert5To8(u8 value) {
+inline u8 Convert5To8(u8 value) {
return (value << 3) | (value >> 2);
}
/// Convert a 6-bit color component to 8 bit
-static inline u8 Convert6To8(u8 value) {
+inline u8 Convert6To8(u8 value) {
return (value << 2) | (value >> 4);
}
+/// Convert a 8-bit color component to 1 bit
+inline u8 Convert8To1(u8 value) {
+ return value >> 7;
+}
+
+/// Convert a 8-bit color component to 4 bit
+inline u8 Convert8To4(u8 value) {
+ return value >> 4;
+}
+
+/// Convert a 8-bit color component to 5 bit
+inline u8 Convert8To5(u8 value) {
+ return value >> 3;
+}
+
+/// Convert a 8-bit color component to 6 bit
+inline u8 Convert8To6(u8 value) {
+ return value >> 2;
+}
+
+/**
+ * Decode a color stored in RGBA8 format
+ * @param bytes Pointer to encoded source color
+ * @return Result color decoded as Math::Vec4<u8>
+ */
+inline const Math::Vec4<u8> DecodeRGBA8(const u8* bytes) {
+ return { bytes[3], bytes[2], bytes[1], bytes[0] };
+}
+
+/**
+ * Decode a color stored in RGB8 format
+ * @param bytes Pointer to encoded source color
+ * @return Result color decoded as Math::Vec4<u8>
+ */
+inline const Math::Vec4<u8> DecodeRGB8(const u8* bytes) {
+ return { bytes[2], bytes[1], bytes[0], 255 };
+}
+
+/**
+ * Decode a color stored in RGB565 format
+ * @param bytes Pointer to encoded source color
+ * @return Result color decoded as Math::Vec4<u8>
+ */
+inline const Math::Vec4<u8> DecodeRGB565(const u8* bytes) {
+ const u16_le pixel = *reinterpret_cast<const u16_le*>(bytes);
+ return { Convert5To8((pixel >> 11) & 0x1F), Convert6To8((pixel >> 5) & 0x3F),
+ Convert5To8(pixel & 0x1F), 255 };
+}
+
+/**
+ * Decode a color stored in RGB5A1 format
+ * @param bytes Pointer to encoded source color
+ * @return Result color decoded as Math::Vec4<u8>
+ */
+inline const Math::Vec4<u8> DecodeRGB5A1(const u8* bytes) {
+ const u16_le pixel = *reinterpret_cast<const u16_le*>(bytes);
+ return { Convert5To8((pixel >> 11) & 0x1F), Convert5To8((pixel >> 6) & 0x1F),
+ Convert5To8((pixel >> 1) & 0x1F), Convert1To8(pixel & 0x1) };
+}
+
+/**
+ * Decode a color stored in RGBA4 format
+ * @param bytes Pointer to encoded source color
+ * @return Result color decoded as Math::Vec4<u8>
+ */
+inline const Math::Vec4<u8> DecodeRGBA4(const u8* bytes) {
+ const u16_le pixel = *reinterpret_cast<const u16_le*>(bytes);
+ return { Convert4To8((pixel >> 12) & 0xF), Convert4To8((pixel >> 8) & 0xF),
+ Convert4To8((pixel >> 4) & 0xF), Convert4To8(pixel & 0xF) };
+}
+
+/**
+ * Encode a color as RGBA8 format
+ * @param color Source color to encode
+ * @param bytes Destination pointer to store encoded color
+ */
+inline void EncodeRGBA8(const Math::Vec4<u8>& color, u8* bytes) {
+ bytes[3] = color.r();
+ bytes[2] = color.g();
+ bytes[1] = color.b();
+ bytes[0] = color.a();
+}
+
+/**
+ * Encode a color as RGB8 format
+ * @param color Source color to encode
+ * @param bytes Destination pointer to store encoded color
+ */
+inline void EncodeRGB8(const Math::Vec4<u8>& color, u8* bytes) {
+ bytes[2] = color.r();
+ bytes[1] = color.g();
+ bytes[0] = color.b();
+}
+
+/**
+ * Encode a color as RGB565 format
+ * @param color Source color to encode
+ * @param bytes Destination pointer to store encoded color
+ */
+inline void EncodeRGB565(const Math::Vec4<u8>& color, u8* bytes) {
+ *reinterpret_cast<u16_le*>(bytes) = (Convert8To5(color.r()) << 11) |
+ (Convert8To6(color.g()) << 5) | Convert8To5(color.b());
+}
+
+/**
+ * Encode a color as RGB5A1 format
+ * @param color Source color to encode
+ * @param bytes Destination pointer to store encoded color
+ */
+inline void EncodeRGB5A1(const Math::Vec4<u8>& color, u8* bytes) {
+ *reinterpret_cast<u16_le*>(bytes) = (Convert8To5(color.r()) << 11) |
+ (Convert8To5(color.g()) << 6) | (Convert8To5(color.b()) << 1) | Convert8To1(color.a());
+}
+
+/**
+ * Encode a color as RGBA4 format
+ * @param color Source color to encode
+ * @param bytes Destination pointer to store encoded color
+ */
+inline void EncodeRGBA4(const Math::Vec4<u8>& color, u8* bytes) {
+ *reinterpret_cast<u16_le*>(bytes) = (Convert8To4(color.r()) << 12) |
+ (Convert8To4(color.g()) << 8) | (Convert8To4(color.b()) << 4) | Convert8To4(color.a());
+}
} // namespace
diff --git a/src/video_core/command_processor.cpp b/src/video_core/command_processor.cpp
index 0d9f4ba66..e031871e8 100644
--- a/src/video_core/command_processor.cpp
+++ b/src/video_core/command_processor.cpp
@@ -2,6 +2,10 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
+#include <boost/range/algorithm/fill.hpp>
+
+#include "common/profiler.h"
+
#include "clipper.h"
#include "command_processor.h"
#include "math.h"
@@ -23,9 +27,7 @@ static int float_regs_counter = 0;
static u32 uniform_write_buffer[4];
-// Used for VSLoadProgramData and VSLoadSwizzleData
-static u32 vs_binary_write_offset = 0;
-static u32 vs_swizzle_write_offset = 0;
+Common::Profiling::TimingCategory category_drawing("Drawing");
static inline void WritePicaReg(u32 id, u32 value, u32 mask) {
@@ -55,6 +57,8 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) {
case PICA_REG_INDEX(trigger_draw):
case PICA_REG_INDEX(trigger_draw_indexed):
{
+ Common::Profiling::ScopeTimer scope_timer(category_drawing);
+
DebugUtils::DumpTevStageConfig(registers.GetTevStages());
if (g_debug_context)
@@ -65,10 +69,14 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) {
// Information about internal vertex attributes
u32 vertex_attribute_sources[16];
- std::fill(vertex_attribute_sources, &vertex_attribute_sources[16], 0xdeadbeef);
+ boost::fill(vertex_attribute_sources, 0xdeadbeef);
u32 vertex_attribute_strides[16];
u32 vertex_attribute_formats[16];
- u32 vertex_attribute_elements[16];
+
+ // HACK: Initialize vertex_attribute_elements to zero to prevent infinite loops below.
+ // This is one of the hacks required to deal with uninitalized vertex attributes.
+ // TODO: Fix this properly.
+ u32 vertex_attribute_elements[16] = {};
u32 vertex_attribute_element_size[16];
// Setup attribute data from loaders
@@ -252,11 +260,6 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) {
break;
}
- // Seems to be used to reset the write pointer for VSLoadProgramData
- case PICA_REG_INDEX(vs_program.begin_load):
- vs_binary_write_offset = 0;
- break;
-
// Load shader program code
case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[0], 0x2cc):
case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[1], 0x2cd):
@@ -267,16 +270,11 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) {
case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[6], 0x2d2):
case PICA_REG_INDEX_WORKAROUND(vs_program.set_word[7], 0x2d3):
{
- VertexShader::SubmitShaderMemoryChange(vs_binary_write_offset, value);
- vs_binary_write_offset++;
+ VertexShader::SubmitShaderMemoryChange(registers.vs_program.offset, value);
+ registers.vs_program.offset++;
break;
}
- // Seems to be used to reset the write pointer for VSLoadSwizzleData
- case PICA_REG_INDEX(vs_swizzle_patterns.begin_load):
- vs_swizzle_write_offset = 0;
- break;
-
// Load swizzle pattern data
case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[0], 0x2d6):
case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[1], 0x2d7):
@@ -287,8 +285,8 @@ static inline void WritePicaReg(u32 id, u32 value, u32 mask) {
case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[6], 0x2dc):
case PICA_REG_INDEX_WORKAROUND(vs_swizzle_patterns.set_word[7], 0x2dd):
{
- VertexShader::SubmitSwizzleDataChange(vs_swizzle_write_offset, value);
- vs_swizzle_write_offset++;
+ VertexShader::SubmitSwizzleDataChange(registers.vs_swizzle_patterns.offset, value);
+ registers.vs_swizzle_patterns.offset++;
break;
}
diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index 8c4ec1044..745c4f4ed 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -23,6 +23,7 @@
#include "video_core/color.h"
#include "video_core/math.h"
#include "video_core/pica.h"
+#include "video_core/utils.h"
#include "debug_utils.h"
@@ -189,7 +190,7 @@ void DumpShader(const u32* binary_data, u32 binary_size, const u32* swizzle_data
);
if (it == output_info_table.end()) {
- output_info_table.push_back({});
+ output_info_table.emplace_back();
output_info_table.back().type = type;
output_info_table.back().component_mask = component_mask;
output_info_table.back().id = i;
@@ -285,7 +286,7 @@ void OnPicaRegWrite(u32 id, u32 value)
if (!is_pica_tracing)
return;
- pica_trace->writes.push_back({id, value});
+ pica_trace->writes.emplace_back(id, value);
}
std::unique_ptr<PicaTrace> FinishPicaTracing()
@@ -306,111 +307,69 @@ std::unique_ptr<PicaTrace> FinishPicaTracing()
}
const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const TextureInfo& info, bool disable_alpha) {
- // Images are split into 8x8 tiles. Each tile is composed of four 4x4 subtiles each
- // of which is composed of four 2x2 subtiles each of which is composed of four texels.
- // Each structure is embedded into the next-bigger one in a diagonal pattern, e.g.
- // texels are laid out in a 2x2 subtile like this:
- // 2 3
- // 0 1
- //
- // The full 8x8 tile has the texels arranged like this:
- //
- // 42 43 46 47 58 59 62 63
- // 40 41 44 45 56 57 60 61
- // 34 35 38 39 50 51 54 55
- // 32 33 36 37 48 49 52 53
- // 10 11 14 15 26 27 30 31
- // 08 09 12 13 24 25 28 29
- // 02 03 06 07 18 19 22 23
- // 00 01 04 05 16 17 20 21
-
- const unsigned int block_width = 8;
- const unsigned int block_height = 8;
-
const unsigned int coarse_x = x & ~7;
const unsigned int coarse_y = y & ~7;
- // Interleave the lower 3 bits of each coordinate to get the intra-block offsets, which are
- // arranged in a Z-order curve. More details on the bit manipulation at:
- // https://fgiesen.wordpress.com/2009/12/13/decoding-morton-codes/
- unsigned int i = (x | (y << 8)) & 0x0707; // ---- -210
- i = (i ^ (i << 2)) & 0x1313; // ---2 --10
- i = (i ^ (i << 1)) & 0x1515; // ---2 -1-0
- i = (i | (i >> 7)) & 0x3F;
-
if (info.format != Regs::TextureFormat::ETC1 &&
info.format != Regs::TextureFormat::ETC1A4) {
// TODO(neobrain): Fix code design to unify vertical block offsets!
source += coarse_y * info.stride;
}
- const unsigned int offset = coarse_x * block_height;
-
+
// TODO: Assert that width/height are multiples of block dimensions
switch (info.format) {
case Regs::TextureFormat::RGBA8:
{
- const u8* source_ptr = source + offset * 4 + i * 4;
- return { source_ptr[3], source_ptr[2], source_ptr[1], disable_alpha ? (u8)255 : source_ptr[0] };
+ auto res = Color::DecodeRGBA8(source + VideoCore::GetMortonOffset(x, y, 4));
+ return { res.r(), res.g(), res.b(), disable_alpha ? 255 : res.a() };
}
case Regs::TextureFormat::RGB8:
{
- const u8* source_ptr = source + offset * 3 + i * 3;
- return { source_ptr[2], source_ptr[1], source_ptr[0], 255 };
+ auto res = Color::DecodeRGB8(source + VideoCore::GetMortonOffset(x, y, 3));
+ return { res.r(), res.g(), res.b(), 255 };
}
- case Regs::TextureFormat::RGBA5551:
+ case Regs::TextureFormat::RGB5A1:
{
- const u16 source_ptr = *(const u16*)(source + offset * 2 + i * 2);
- u8 r = (source_ptr >> 11) & 0x1F;
- u8 g = ((source_ptr) >> 6) & 0x1F;
- u8 b = (source_ptr >> 1) & 0x1F;
- u8 a = source_ptr & 1;
- return Math::MakeVec<u8>(Color::Convert5To8(r), Color::Convert5To8(g),
- Color::Convert5To8(b), disable_alpha ? 255 : Color::Convert1To8(a));
+ auto res = Color::DecodeRGB5A1(source + VideoCore::GetMortonOffset(x, y, 2));
+ return { res.r(), res.g(), res.b(), disable_alpha ? 255 : res.a() };
}
case Regs::TextureFormat::RGB565:
{
- const u16 source_ptr = *(const u16*)(source + offset * 2 + i * 2);
- u8 r = Color::Convert5To8((source_ptr >> 11) & 0x1F);
- u8 g = Color::Convert6To8(((source_ptr) >> 5) & 0x3F);
- u8 b = Color::Convert5To8((source_ptr) & 0x1F);
- return Math::MakeVec<u8>(r, g, b, 255);
+ auto res = Color::DecodeRGB565(source + VideoCore::GetMortonOffset(x, y, 2));
+ return { res.r(), res.g(), res.b(), 255 };
}
case Regs::TextureFormat::RGBA4:
{
- const u8* source_ptr = source + offset * 2 + i * 2;
- u8 r = Color::Convert4To8(source_ptr[1] >> 4);
- u8 g = Color::Convert4To8(source_ptr[1] & 0xF);
- u8 b = Color::Convert4To8(source_ptr[0] >> 4);
- u8 a = Color::Convert4To8(source_ptr[0] & 0xF);
- return { r, g, b, disable_alpha ? (u8)255 : a };
+ auto res = Color::DecodeRGBA4(source + VideoCore::GetMortonOffset(x, y, 2));
+ return { res.r(), res.g(), res.b(), disable_alpha ? 255 : res.a() };
}
case Regs::TextureFormat::IA8:
{
- const u8* source_ptr = source + offset * 2 + i * 2;
+ const u8* source_ptr = source + VideoCore::GetMortonOffset(x, y, 2);
if (disable_alpha) {
// Show intensity as red, alpha as green
return { source_ptr[1], source_ptr[0], 0, 255 };
} else {
- return { source_ptr[1], source_ptr[1], source_ptr[1], source_ptr[0]};
+ return { source_ptr[1], source_ptr[1], source_ptr[1], source_ptr[0] };
}
}
case Regs::TextureFormat::I8:
{
- const u8* source_ptr = source + offset + i;
+ const u8* source_ptr = source + VideoCore::GetMortonOffset(x, y, 1);
return { *source_ptr, *source_ptr, *source_ptr, 255 };
}
case Regs::TextureFormat::A8:
{
- const u8* source_ptr = source + offset + i;
+ const u8* source_ptr = source + VideoCore::GetMortonOffset(x, y, 1);
if (disable_alpha) {
return { *source_ptr, *source_ptr, *source_ptr, 255 };
@@ -421,7 +380,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
case Regs::TextureFormat::IA4:
{
- const u8* source_ptr = source + offset + i;
+ const u8* source_ptr = source + VideoCore::GetMortonOffset(x, y, 1);
u8 i = Color::Convert4To8(((*source_ptr) & 0xF0) >> 4);
u8 a = Color::Convert4To8((*source_ptr) & 0xF);
@@ -436,9 +395,10 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
case Regs::TextureFormat::A4:
{
- const u8* source_ptr = source + offset / 2 + i / 2;
+ u32 morton_offset = VideoCore::GetMortonOffset(x, y, 1);
+ const u8* source_ptr = source + morton_offset / 2;
- u8 a = (coarse_x % 2) ? ((*source_ptr)&0xF) : (((*source_ptr) & 0xF0) >> 4);
+ u8 a = (morton_offset % 2) ? ((*source_ptr & 0xF0) >> 4) : (*source_ptr & 0xF);
a = Color::Convert4To8(a);
if (disable_alpha) {
@@ -545,7 +505,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
}
// Add modifier
- unsigned table_index = (x < 2) ? table_index_2.Value() : table_index_1.Value();
+ unsigned table_index = (x < 2) ? table_index_1.Value() : table_index_2.Value();
static const auto etc1_modifier_table = std::array<std::array<u8, 2>, 8>{{
{ 2, 8 }, { 5, 17 }, { 9, 29 }, { 13, 42 },
diff --git a/src/video_core/gpu_debugger.h b/src/video_core/gpu_debugger.h
index 03641d93b..48ac269e3 100644
--- a/src/video_core/gpu_debugger.h
+++ b/src/video_core/gpu_debugger.h
@@ -58,8 +58,8 @@ public:
if (observers.empty())
return;
- gx_command_history.push_back(GSP_GPU::Command());
- GSP_GPU::Command& cmd = gx_command_history[gx_command_history.size()-1];
+ gx_command_history.emplace_back();
+ GSP_GPU::Command& cmd = gx_command_history.back();
memcpy(&cmd, command_data, sizeof(GSP_GPU::Command));
diff --git a/src/video_core/pica.h b/src/video_core/pica.h
index 9c1a12dc8..b14de9278 100644
--- a/src/video_core/pica.h
+++ b/src/video_core/pica.h
@@ -39,13 +39,6 @@ namespace Pica {
struct Regs {
-// helper macro to properly align structure members.
-// Calling INSERT_PADDING_WORDS will add a new member variable with a name like "pad121",
-// depending on the current source line to make sure variable names are unique.
-#define INSERT_PADDING_WORDS_HELPER1(x, y) x ## y
-#define INSERT_PADDING_WORDS_HELPER2(x, y) INSERT_PADDING_WORDS_HELPER1(x, y)
-#define INSERT_PADDING_WORDS(num_words) u32 INSERT_PADDING_WORDS_HELPER2(pad, __LINE__)[(num_words)];
-
INSERT_PADDING_WORDS(0x10);
u32 trigger_irq;
@@ -118,8 +111,9 @@ struct Regs {
struct TextureConfig {
enum WrapMode : u32 {
- ClampToEdge = 0,
- Repeat = 2,
+ ClampToEdge = 0,
+ Repeat = 2,
+ MirroredRepeat = 3,
};
INSERT_PADDING_WORDS(0x1);
@@ -131,7 +125,7 @@ struct Regs {
union {
BitField< 8, 2, WrapMode> wrap_s;
- BitField<11, 2, WrapMode> wrap_t;
+ BitField<12, 2, WrapMode> wrap_t;
};
INSERT_PADDING_WORDS(0x1);
@@ -151,7 +145,7 @@ struct Regs {
enum class TextureFormat : u32 {
RGBA8 = 0,
RGB8 = 1,
- RGBA5551 = 2,
+ RGB5A1 = 2,
RGB565 = 3,
RGBA4 = 4,
IA8 = 5,
@@ -173,7 +167,7 @@ struct Regs {
case TextureFormat::RGB8:
return 6;
- case TextureFormat::RGBA5551:
+ case TextureFormat::RGB5A1:
case TextureFormat::RGB565:
case TextureFormat::RGBA4:
case TextureFormat::IA8:
@@ -223,6 +217,8 @@ struct Regs {
struct TevStageConfig {
enum class Source : u32 {
PrimaryColor = 0x0,
+ PrimaryFragmentColor = 0x1,
+
Texture0 = 0x3,
Texture1 = 0x4,
Texture2 = 0x5,
@@ -265,6 +261,9 @@ struct Regs {
AddSigned = 3,
Lerp = 4,
Subtract = 5,
+
+ MultiplyThenAdd = 8,
+ AddThenMultiply = 9,
};
union {
@@ -337,7 +336,7 @@ struct Regs {
};
union {
- enum BlendEquation : u32 {
+ enum class BlendEquation : u32 {
Add = 0,
Subtract = 1,
ReverseSubtract = 2,
@@ -410,10 +409,11 @@ struct Regs {
} output_merger;
struct {
+ // Components are laid out in reverse byte order, most significant bits first.
enum ColorFormat : u32 {
RGBA8 = 0,
RGB8 = 1,
- RGBA5551 = 2,
+ RGB5A1 = 2,
RGB565 = 3,
RGBA4 = 4,
};
@@ -421,7 +421,7 @@ struct Regs {
INSERT_PADDING_WORDS(0x6);
u32 depth_format;
- u32 color_format;
+ BitField<16, 3, u32> color_format;
INSERT_PADDING_WORDS(0x4);
@@ -678,7 +678,9 @@ struct Regs {
INSERT_PADDING_WORDS(0x2);
struct {
- u32 begin_load;
+ // Offset of the next instruction to write code to.
+ // Incremented with each instruction write.
+ u32 offset;
// Writing to these registers sets the "current" word in the shader program.
// TODO: It's not clear how the hardware stores what the "current" word is.
@@ -690,7 +692,9 @@ struct Regs {
// This register group is used to load an internal table of swizzling patterns,
// which are indexed by each shader instruction to specify vector component swizzling.
struct {
- u32 begin_load;
+ // Offset of the next swizzle pattern to write code to.
+ // Incremented with each instruction write.
+ u32 offset;
// Writing to these registers sets the "current" swizzle pattern in the table.
// TODO: It's not clear how the hardware stores what the "current" swizzle pattern is.
@@ -699,10 +703,6 @@ struct Regs {
INSERT_PADDING_WORDS(0x22);
-#undef INSERT_PADDING_WORDS_HELPER1
-#undef INSERT_PADDING_WORDS_HELPER2
-#undef INSERT_PADDING_WORDS
-
// Map register indices to names readable by humans
// Used for debugging purposes, so performance is not an issue here
static std::string GetCommandName(int index) {
diff --git a/src/video_core/rasterizer.cpp b/src/video_core/rasterizer.cpp
index 3faa10153..5861c1926 100644
--- a/src/video_core/rasterizer.cpp
+++ b/src/video_core/rasterizer.cpp
@@ -5,13 +5,16 @@
#include <algorithm>
#include "common/common_types.h"
+#include "common/math_util.h"
+#include "core/hw/gpu.h"
+#include "debug_utils/debug_utils.h"
#include "math.h"
+#include "color.h"
#include "pica.h"
#include "rasterizer.h"
#include "vertex_shader.h"
-
-#include "debug_utils/debug_utils.h"
+#include "video_core/utils.h"
namespace Pica {
@@ -19,40 +22,101 @@ namespace Rasterizer {
static void DrawPixel(int x, int y, const Math::Vec4<u8>& color) {
const PAddr addr = registers.framebuffer.GetColorBufferPhysicalAddress();
- u32* color_buffer = reinterpret_cast<u32*>(Memory::GetPointer(PAddrToVAddr(addr)));
- u32 value = (color.a() << 24) | (color.r() << 16) | (color.g() << 8) | color.b();
- // Assuming RGBA8 format until actual framebuffer format handling is implemented
- *(color_buffer + x + y * registers.framebuffer.GetWidth()) = value;
+ // Similarly to textures, the render framebuffer is laid out from bottom to top, too.
+ // NOTE: The framebuffer height register contains the actual FB height minus one.
+ y = (registers.framebuffer.height - y);
+
+ const u32 coarse_y = y & ~7;
+ u32 bytes_per_pixel = GPU::Regs::BytesPerPixel(GPU::Regs::PixelFormat(registers.framebuffer.color_format.Value()));
+ u32 dst_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) + coarse_y * registers.framebuffer.width * bytes_per_pixel;
+ u8* dst_pixel = Memory::GetPointer(PAddrToVAddr(addr)) + dst_offset;
+
+ switch (registers.framebuffer.color_format) {
+ case registers.framebuffer.RGBA8:
+ Color::EncodeRGBA8(color, dst_pixel);
+ break;
+
+ case registers.framebuffer.RGB8:
+ Color::EncodeRGB8(color, dst_pixel);
+ break;
+
+ case registers.framebuffer.RGB5A1:
+ Color::EncodeRGB5A1(color, dst_pixel);
+ break;
+
+ case registers.framebuffer.RGB565:
+ Color::EncodeRGB565(color, dst_pixel);
+ break;
+
+ case registers.framebuffer.RGBA4:
+ Color::EncodeRGBA4(color, dst_pixel);
+ break;
+
+ default:
+ LOG_CRITICAL(Render_Software, "Unknown framebuffer color format %x", registers.framebuffer.color_format.Value());
+ UNIMPLEMENTED();
+ }
}
static const Math::Vec4<u8> GetPixel(int x, int y) {
const PAddr addr = registers.framebuffer.GetColorBufferPhysicalAddress();
- u32* color_buffer_u32 = reinterpret_cast<u32*>(Memory::GetPointer(PAddrToVAddr(addr)));
-
- u32 value = *(color_buffer_u32 + x + y * registers.framebuffer.GetWidth());
- Math::Vec4<u8> ret;
- ret.a() = value >> 24;
- ret.r() = (value >> 16) & 0xFF;
- ret.g() = (value >> 8) & 0xFF;
- ret.b() = value & 0xFF;
- return ret;
+
+ y = (registers.framebuffer.height - y);
+
+ const u32 coarse_y = y & ~7;
+ u32 bytes_per_pixel = GPU::Regs::BytesPerPixel(GPU::Regs::PixelFormat(registers.framebuffer.color_format.Value()));
+ u32 src_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) + coarse_y * registers.framebuffer.width * bytes_per_pixel;
+ u8* src_pixel = Memory::GetPointer(PAddrToVAddr(addr)) + src_offset;
+
+ switch (registers.framebuffer.color_format) {
+ case registers.framebuffer.RGBA8:
+ return Color::DecodeRGBA8(src_pixel);
+
+ case registers.framebuffer.RGB8:
+ return Color::DecodeRGB8(src_pixel);
+
+ case registers.framebuffer.RGB5A1:
+ return Color::DecodeRGB5A1(src_pixel);
+
+ case registers.framebuffer.RGB565:
+ return Color::DecodeRGB565(src_pixel);
+
+ case registers.framebuffer.RGBA4:
+ return Color::DecodeRGBA4(src_pixel);
+
+ default:
+ LOG_CRITICAL(Render_Software, "Unknown framebuffer color format %x", registers.framebuffer.color_format.Value());
+ UNIMPLEMENTED();
+ }
+
+ return {};
}
static u32 GetDepth(int x, int y) {
const PAddr addr = registers.framebuffer.GetDepthBufferPhysicalAddress();
- u16* depth_buffer = reinterpret_cast<u16*>(Memory::GetPointer(PAddrToVAddr(addr)));
+ u8* depth_buffer = Memory::GetPointer(PAddrToVAddr(addr));
+
+ y = (registers.framebuffer.height - y);
+
+ const u32 coarse_y = y & ~7;
+ u32 stride = registers.framebuffer.width * 2;
// Assuming 16-bit depth buffer format until actual format handling is implemented
- return *(depth_buffer + x + y * registers.framebuffer.GetWidth());
+ return *(u16*)(depth_buffer + VideoCore::GetMortonOffset(x, y, 2) + coarse_y * stride);
}
static void SetDepth(int x, int y, u16 value) {
const PAddr addr = registers.framebuffer.GetDepthBufferPhysicalAddress();
- u16* depth_buffer = reinterpret_cast<u16*>(Memory::GetPointer(PAddrToVAddr(addr)));
+ u8* depth_buffer = Memory::GetPointer(PAddrToVAddr(addr));
+
+ y = (registers.framebuffer.height - y);
+
+ const u32 coarse_y = y & ~7;
+ u32 stride = registers.framebuffer.width * 2;
// Assuming 16-bit depth buffer format until actual format handling is implemented
- *(depth_buffer + x + y * registers.framebuffer.GetWidth()) = value;
+ *(u16*)(depth_buffer + VideoCore::GetMortonOffset(x, y, 2) + coarse_y * stride) = value;
}
// NOTE: Assuming that rasterizer coordinates are 12.4 fixed-point values
@@ -90,30 +154,43 @@ static int SignedArea (const Math::Vec2<Fix12P4>& vtx1,
return Math::Cross(vec1, vec2).z;
};
-void ProcessTriangle(const VertexShader::OutputVertex& v0,
- const VertexShader::OutputVertex& v1,
- const VertexShader::OutputVertex& v2)
+/**
+ * Helper function for ProcessTriangle with the "reversed" flag to allow for implementing
+ * culling via recursion.
+ */
+static void ProcessTriangleInternal(const VertexShader::OutputVertex& v0,
+ const VertexShader::OutputVertex& v1,
+ const VertexShader::OutputVertex& v2,
+ bool reversed = false)
{
// vertex positions in rasterizer coordinates
- auto FloatToFix = [](float24 flt) {
- return Fix12P4(static_cast<unsigned short>(flt.ToFloat32() * 16.0f));
- };
- auto ScreenToRasterizerCoordinates = [FloatToFix](const Math::Vec3<float24> vec) {
- return Math::Vec3<Fix12P4>{FloatToFix(vec.x), FloatToFix(vec.y), FloatToFix(vec.z)};
- };
+ static auto FloatToFix = [](float24 flt) {
+ // TODO: Rounding here is necessary to prevent garbage pixels at
+ // triangle borders. Is it that the correct solution, though?
+ return Fix12P4(static_cast<unsigned short>(round(flt.ToFloat32() * 16.0f)));
+ };
+ static auto ScreenToRasterizerCoordinates = [](const Math::Vec3<float24>& vec) {
+ return Math::Vec3<Fix12P4>{FloatToFix(vec.x), FloatToFix(vec.y), FloatToFix(vec.z)};
+ };
Math::Vec3<Fix12P4> vtxpos[3]{ ScreenToRasterizerCoordinates(v0.screenpos),
ScreenToRasterizerCoordinates(v1.screenpos),
ScreenToRasterizerCoordinates(v2.screenpos) };
- if (registers.cull_mode == Regs::CullMode::KeepClockWise) {
- // Reverse vertex order and use the CCW code path.
- std::swap(vtxpos[1], vtxpos[2]);
- }
+ if (registers.cull_mode == Regs::CullMode::KeepAll) {
+ // Make sure we always end up with a triangle wound counter-clockwise
+ if (!reversed && SignedArea(vtxpos[0].xy(), vtxpos[1].xy(), vtxpos[2].xy()) <= 0) {
+ ProcessTriangleInternal(v0, v2, v1, true);
+ return;
+ }
+ } else {
+ if (!reversed && registers.cull_mode == Regs::CullMode::KeepClockWise) {
+ // Reverse vertex order and use the CCW code path.
+ ProcessTriangleInternal(v0, v2, v1, true);
+ return;
+ }
- if (registers.cull_mode != Regs::CullMode::KeepAll) {
// Cull away triangles which are wound clockwise.
- // TODO: A check for degenerate triangles ("== 0") should be considered for CullMode::KeepAll
if (SignedArea(vtxpos[0].xy(), vtxpos[1].xy(), vtxpos[2].xy()) <= 0)
return;
}
@@ -155,9 +232,10 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
auto textures = registers.GetTextures();
auto tev_stages = registers.GetTevStages();
+ // Enter rasterization loop, starting at the center of the topleft bounding box corner.
// TODO: Not sure if looping through x first might be faster
- for (u16 y = min_y; y < max_y; y += 0x10) {
- for (u16 x = min_x; x < max_x; x += 0x10) {
+ for (u16 y = min_y + 8; y < max_y; y += 0x10) {
+ for (u16 x = min_x + 8; x < max_x; x += 0x10) {
// Calculate the barycentric coordinates w0, w1 and w2
int w0 = bias0 + SignedArea(vtxpos[1].xy(), vtxpos[2].xy(), {x, y});
@@ -220,7 +298,7 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
int s = (int)(uv[i].u() * float24::FromFloat32(static_cast<float>(texture.config.width))).ToFloat32();
int t = (int)(uv[i].v() * float24::FromFloat32(static_cast<float>(texture.config.height))).ToFloat32();
- auto GetWrappedTexCoord = [](Regs::TextureConfig::WrapMode mode, int val, unsigned size) {
+ static auto GetWrappedTexCoord = [](Regs::TextureConfig::WrapMode mode, int val, unsigned size) {
switch (mode) {
case Regs::TextureConfig::ClampToEdge:
val = std::max(val, 0);
@@ -228,7 +306,15 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
return val;
case Regs::TextureConfig::Repeat:
- return (int)(((unsigned)val) % size);
+ return (int)((unsigned)val % size);
+
+ case Regs::TextureConfig::MirroredRepeat:
+ {
+ int coord = (int)((unsigned)val % (2 * size));
+ if (coord >= size)
+ coord = 2 * size - 1 - coord;
+ return coord;
+ }
default:
LOG_ERROR(HW_GPU, "Unknown texture coordinate wrapping mode %x\n", (int)mode);
@@ -236,6 +322,10 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
return 0;
}
};
+
+ // Textures are laid out from bottom to top, hence we invert the t coordinate.
+ // NOTE: This may not be the right place for the inversion.
+ // TODO: Check if this applies to ETC textures, too.
s = GetWrappedTexCoord(texture.config.wrap_s, s, texture.config.width);
t = texture.config.height - 1 - GetWrappedTexCoord(texture.config.wrap_t, t, texture.config.height);
@@ -262,7 +352,9 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
auto GetSource = [&](Source source) -> Math::Vec4<u8> {
switch (source) {
+ // TODO: What's the difference between these two?
case Source::PrimaryColor:
+ case Source::PrimaryFragmentColor:
return primary_color;
case Source::Texture0:
@@ -378,6 +470,25 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
return result.Cast<u8>();
}
+ case Operation::MultiplyThenAdd:
+ {
+ auto result = (input[0] * input[1] + 255 * input[2].Cast<int>()) / 255;
+ result.r() = std::min(255, result.r());
+ result.g() = std::min(255, result.g());
+ result.b() = std::min(255, result.b());
+ return result.Cast<u8>();
+ }
+
+ case Operation::AddThenMultiply:
+ {
+ auto result = input[0] + input[1];
+ result.r() = std::min(255, result.r());
+ result.g() = std::min(255, result.g());
+ result.b() = std::min(255, result.b());
+ result = (result * input[2].Cast<int>()) / 255;
+ return result.Cast<u8>();
+ }
+
default:
LOG_ERROR(HW_GPU, "Unknown color combiner operation %d\n", (int)op);
UNIMPLEMENTED();
@@ -402,6 +513,12 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
case Operation::Subtract:
return std::max(0, (int)input[0] - (int)input[1]);
+ case Operation::MultiplyThenAdd:
+ return std::min(255, (input[0] * input[1] + 255 * input[2]) / 255);
+
+ case Operation::AddThenMultiply:
+ return (std::min(255, (input[0] + input[1])) * input[2]) / 255;
+
default:
LOG_ERROR(HW_GPU, "Unknown alpha combiner operation %d\n", (int)op);
UNIMPLEMENTED();
@@ -475,7 +592,7 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
// TODO: Does depth indeed only get written even if depth testing is enabled?
if (registers.output_merger.depth_test_enable) {
- u16 z = (u16)(-(v0.screenpos[2].ToFloat32() * w0 +
+ u16 z = (u16)((v0.screenpos[2].ToFloat32() * w0 +
v1.screenpos[2].ToFloat32() * w1 +
v2.screenpos[2].ToFloat32() * w2) * 65535.f / wsum);
u16 ref_z = GetDepth(x >> 4, y >> 4);
@@ -524,6 +641,7 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
}
auto dest = GetPixel(x >> 4, y >> 4);
+ Math::Vec4<u8> blend_output = combiner_output;
if (registers.output_merger.alphablend_enable) {
auto params = registers.output_merger.alpha_blending;
@@ -574,7 +692,7 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
default:
LOG_CRITICAL(HW_GPU, "Unknown color blend factor %x", factor);
- exit(0);
+ UNIMPLEMENTED();
break;
}
};
@@ -607,86 +725,78 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
default:
LOG_CRITICAL(HW_GPU, "Unknown alpha blend factor %x", factor);
- exit(0);
+ UNIMPLEMENTED();
+ break;
+ }
+ };
+
+ using BlendEquation = decltype(params)::BlendEquation;
+ static auto EvaluateBlendEquation = [](const Math::Vec4<u8>& src, const Math::Vec4<u8>& srcfactor,
+ const Math::Vec4<u8>& dest, const Math::Vec4<u8>& destfactor,
+ BlendEquation equation) {
+ Math::Vec4<int> result;
+
+ auto src_result = (src * srcfactor).Cast<int>();
+ auto dst_result = (dest * destfactor).Cast<int>();
+
+ switch (equation) {
+ case BlendEquation::Add:
+ result = (src_result + dst_result) / 255;
+ break;
+
+ case BlendEquation::Subtract:
+ result = (src_result - dst_result) / 255;
+ break;
+
+ case BlendEquation::ReverseSubtract:
+ result = (dst_result - src_result) / 255;
+ break;
+
+ // TODO: How do these two actually work?
+ // OpenGL doesn't include the blend factors in the min/max computations,
+ // but is this what the 3DS actually does?
+ case BlendEquation::Min:
+ result.r() = std::min(src.r(), dest.r());
+ result.g() = std::min(src.g(), dest.g());
+ result.b() = std::min(src.b(), dest.b());
+ result.a() = std::min(src.a(), dest.a());
break;
+
+ case BlendEquation::Max:
+ result.r() = std::max(src.r(), dest.r());
+ result.g() = std::max(src.g(), dest.g());
+ result.b() = std::max(src.b(), dest.b());
+ result.a() = std::max(src.a(), dest.a());
+ break;
+
+ default:
+ LOG_CRITICAL(HW_GPU, "Unknown RGB blend equation %x", equation);
+ UNIMPLEMENTED();
}
+
+ return Math::Vec4<u8>(MathUtil::Clamp(result.r(), 0, 255),
+ MathUtil::Clamp(result.g(), 0, 255),
+ MathUtil::Clamp(result.b(), 0, 255),
+ MathUtil::Clamp(result.a(), 0, 255));
};
auto srcfactor = Math::MakeVec(LookupFactorRGB(params.factor_source_rgb),
LookupFactorA(params.factor_source_a));
auto dstfactor = Math::MakeVec(LookupFactorRGB(params.factor_dest_rgb),
LookupFactorA(params.factor_dest_a));
-
- auto src_result = (combiner_output * srcfactor).Cast<int>();
- auto dst_result = (dest * dstfactor).Cast<int>();
-
- switch (params.blend_equation_rgb) {
- case params.Add:
- {
- auto result = (src_result + dst_result) / 255;
- result.r() = std::min(255, result.r());
- result.g() = std::min(255, result.g());
- result.b() = std::min(255, result.b());
- combiner_output = result.Cast<u8>();
- break;
- }
-
- case params.Subtract:
- {
- auto result = (src_result - dst_result) / 255;
- result.r() = std::max(0, result.r());
- result.g() = std::max(0, result.g());
- result.b() = std::max(0, result.b());
- combiner_output = result.Cast<u8>();
- break;
- }
-
- case params.ReverseSubtract:
- {
- auto result = (dst_result - src_result) / 255;
- result.r() = std::max(0, result.r());
- result.g() = std::max(0, result.g());
- result.b() = std::max(0, result.b());
- combiner_output = result.Cast<u8>();
- break;
- }
-
- case params.Min:
- {
- // TODO: GL spec says to do it without the factors, but is this what the 3DS does?
- Math::Vec4<int> result;
- result.r() = std::min(combiner_output.r(),dest.r());
- result.g() = std::min(combiner_output.g(),dest.g());
- result.b() = std::min(combiner_output.b(),dest.b());
- combiner_output = result.Cast<u8>();
- break;
- }
-
- case params.Max:
- {
- // TODO: GL spec says to do it without the factors, but is this what the 3DS does?
- Math::Vec4<int> result;
- result.r() = std::max(combiner_output.r(),dest.r());
- result.g() = std::max(combiner_output.g(),dest.g());
- result.b() = std::max(combiner_output.b(),dest.b());
- combiner_output = result.Cast<u8>();
- break;
- }
- default:
- LOG_CRITICAL(HW_GPU, "Unknown RGB blend equation %x", params.blend_equation_rgb.Value());
- exit(0);
- }
+ blend_output = EvaluateBlendEquation(combiner_output, srcfactor, dest, dstfactor, params.blend_equation_rgb);
+ blend_output.a() = EvaluateBlendEquation(combiner_output, srcfactor, dest, dstfactor, params.blend_equation_a).a();
} else {
LOG_CRITICAL(HW_GPU, "logic op: %x", registers.output_merger.logic_op);
- exit(0);
+ UNIMPLEMENTED();
}
const Math::Vec4<u8> result = {
- registers.output_merger.red_enable ? combiner_output.r() : dest.r(),
- registers.output_merger.green_enable ? combiner_output.g() : dest.g(),
- registers.output_merger.blue_enable ? combiner_output.b() : dest.b(),
- registers.output_merger.alpha_enable ? combiner_output.a() : dest.a()
+ registers.output_merger.red_enable ? blend_output.r() : dest.r(),
+ registers.output_merger.green_enable ? blend_output.g() : dest.g(),
+ registers.output_merger.blue_enable ? blend_output.b() : dest.b(),
+ registers.output_merger.alpha_enable ? blend_output.a() : dest.a()
};
DrawPixel(x >> 4, y >> 4, result);
@@ -694,6 +804,12 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
}
}
+void ProcessTriangle(const VertexShader::OutputVertex& v0,
+ const VertexShader::OutputVertex& v1,
+ const VertexShader::OutputVertex& v2) {
+ ProcessTriangleInternal(v0, v1, v2);
+}
+
} // namespace Rasterizer
} // namespace Pica
diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp
index 272695174..95ab96340 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.cpp
+++ b/src/video_core/renderer_opengl/renderer_opengl.cpp
@@ -4,7 +4,10 @@
#include "core/hw/gpu.h"
#include "core/mem_map.h"
+
#include "common/emu_window.h"
+#include "common/profiler_reporting.h"
+
#include "video_core/video_core.h"
#include "video_core/renderer_opengl/renderer_opengl.h"
#include "video_core/renderer_opengl/gl_shader_util.h"
@@ -75,9 +78,18 @@ void RendererOpenGL::SwapBuffers() {
DrawScreens();
+ auto& profiler = Common::Profiling::GetProfilingManager();
+ profiler.FinishFrame();
+ {
+ auto aggregator = Common::Profiling::GetTimingResultsAggregator();
+ aggregator->AddFrame(profiler.GetPreviousFrameResults());
+ }
+
// Swap buffers
render_window->PollEvents();
render_window->SwapBuffers();
+
+ profiler.BeginFrame();
}
/**
@@ -242,28 +254,26 @@ void RendererOpenGL::DrawSingleScreenRotated(const TextureInfo& texture, float x
* Draws the emulated screens to the emulator window.
*/
void RendererOpenGL::DrawScreens() {
- auto viewport_extent = GetViewportExtent();
- glViewport(viewport_extent.left, viewport_extent.top, viewport_extent.GetWidth(), viewport_extent.GetHeight()); // TODO: Or bottom?
+ auto layout = render_window->GetFramebufferLayout();
+
+ glViewport(0, 0, layout.width, layout.height);
glClear(GL_COLOR_BUFFER_BIT);
glUseProgram(program_id);
// Set projection matrix
- std::array<GLfloat, 3*2> ortho_matrix = MakeOrthographicMatrix((float)resolution_width, (float)resolution_height);
+ std::array<GLfloat, 3 * 2> ortho_matrix = MakeOrthographicMatrix((float)layout.width,
+ (float)layout.height);
glUniformMatrix3x2fv(uniform_modelview_matrix, 1, GL_FALSE, ortho_matrix.data());
// Bind texture in Texture Unit 0
glActiveTexture(GL_TEXTURE0);
glUniform1i(uniform_color_texture, 0);
- const float max_width = std::max((float)VideoCore::kScreenTopWidth, (float)VideoCore::kScreenBottomWidth);
- const float top_x = 0.5f * (max_width - VideoCore::kScreenTopWidth);
- const float bottom_x = 0.5f * (max_width - VideoCore::kScreenBottomWidth);
-
- DrawSingleScreenRotated(textures[0], top_x, 0,
- (float)VideoCore::kScreenTopWidth, (float)VideoCore::kScreenTopHeight);
- DrawSingleScreenRotated(textures[1], bottom_x, (float)VideoCore::kScreenTopHeight,
- (float)VideoCore::kScreenBottomWidth, (float)VideoCore::kScreenBottomHeight);
+ DrawSingleScreenRotated(textures[0], (float)layout.top_screen.left, (float)layout.top_screen.top,
+ (float)layout.top_screen.GetWidth(), (float)layout.top_screen.GetHeight());
+ DrawSingleScreenRotated(textures[1], (float)layout.bottom_screen.left,(float)layout.bottom_screen.top,
+ (float)layout.bottom_screen.GetWidth(), (float)layout.bottom_screen.GetHeight());
m_current_frame++;
}
@@ -280,34 +290,6 @@ void RendererOpenGL::SetWindow(EmuWindow* window) {
render_window = window;
}
-MathUtil::Rectangle<unsigned> RendererOpenGL::GetViewportExtent() {
- unsigned framebuffer_width;
- unsigned framebuffer_height;
- std::tie(framebuffer_width, framebuffer_height) = render_window->GetFramebufferSize();
-
- float window_aspect_ratio = static_cast<float>(framebuffer_height) / framebuffer_width;
- float emulation_aspect_ratio = static_cast<float>(resolution_height) / resolution_width;
-
- MathUtil::Rectangle<unsigned> viewport_extent;
- if (window_aspect_ratio > emulation_aspect_ratio) {
- // Window is narrower than the emulation content => apply borders to the top and bottom
- unsigned viewport_height = static_cast<unsigned>(std::round(emulation_aspect_ratio * framebuffer_width));
- viewport_extent.left = 0;
- viewport_extent.top = (framebuffer_height - viewport_height) / 2;
- viewport_extent.right = viewport_extent.left + framebuffer_width;
- viewport_extent.bottom = viewport_extent.top + viewport_height;
- } else {
- // Otherwise, apply borders to the left and right sides of the window.
- unsigned viewport_width = static_cast<unsigned>(std::round(framebuffer_height / emulation_aspect_ratio));
- viewport_extent.left = (framebuffer_width - viewport_width) / 2;
- viewport_extent.top = 0;
- viewport_extent.right = viewport_extent.left + viewport_width;
- viewport_extent.bottom = viewport_extent.top + framebuffer_height;
- }
-
- return viewport_extent;
-}
-
/// Initialize the renderer
void RendererOpenGL::Init() {
render_window->MakeCurrent();
diff --git a/src/video_core/utils.h b/src/video_core/utils.h
index 6fd640425..bda793fa5 100644
--- a/src/video_core/utils.h
+++ b/src/video_core/utils.h
@@ -35,4 +35,54 @@ struct TGAHeader {
*/
void DumpTGA(std::string filename, short width, short height, u8* raw_data);
+/**
+ * Interleave the lower 3 bits of each coordinate to get the intra-block offsets, which are
+ * arranged in a Z-order curve. More details on the bit manipulation at:
+ * https://fgiesen.wordpress.com/2009/12/13/decoding-morton-codes/
+ */
+static inline u32 MortonInterleave(u32 x, u32 y) {
+ u32 i = (x & 7) | ((y & 7) << 8); // ---- -210
+ i = (i ^ (i << 2)) & 0x1313; // ---2 --10
+ i = (i ^ (i << 1)) & 0x1515; // ---2 -1-0
+ i = (i | (i >> 7)) & 0x3F;
+ return i;
+}
+
+/**
+ * Calculates the offset of the position of the pixel in Morton order
+ */
+static inline u32 GetMortonOffset(u32 x, u32 y, u32 bytes_per_pixel) {
+ // Images are split into 8x8 tiles. Each tile is composed of four 4x4 subtiles each
+ // of which is composed of four 2x2 subtiles each of which is composed of four texels.
+ // Each structure is embedded into the next-bigger one in a diagonal pattern, e.g.
+ // texels are laid out in a 2x2 subtile like this:
+ // 2 3
+ // 0 1
+ //
+ // The full 8x8 tile has the texels arranged like this:
+ //
+ // 42 43 46 47 58 59 62 63
+ // 40 41 44 45 56 57 60 61
+ // 34 35 38 39 50 51 54 55
+ // 32 33 36 37 48 49 52 53
+ // 10 11 14 15 26 27 30 31
+ // 08 09 12 13 24 25 28 29
+ // 02 03 06 07 18 19 22 23
+ // 00 01 04 05 16 17 20 21
+ //
+ // This pattern is what's called Z-order curve, or Morton order.
+
+ const unsigned int block_width = 8;
+ const unsigned int block_height = 8;
+
+ const unsigned int coarse_x = x & ~7;
+ const unsigned int coarse_y = y & ~7;
+
+ u32 i = VideoCore::MortonInterleave(x, y);
+
+ const unsigned int offset = coarse_x * block_height;
+
+ return (i + offset) * bytes_per_pixel;
+}
+
} // namespace
diff --git a/src/video_core/vertex_shader.cpp b/src/video_core/vertex_shader.cpp
index 80935a50a..bc8c0041c 100644
--- a/src/video_core/vertex_shader.cpp
+++ b/src/video_core/vertex_shader.cpp
@@ -85,8 +85,12 @@ struct VertexShaderState {
};
struct CallStackElement {
- u32 final_address;
- u32 return_address;
+ u32 final_address; // Address upon which we jump to return_address
+ u32 return_address; // Where to jump when leaving scope
+ u8 repeat_counter; // How often to repeat until this call stack element is removed
+ u8 loop_increment; // Which value to add to the loop counter after an iteration
+ // TODO: Should this be a signed value? Does it even matter?
+ u32 loop_address; // The address where we'll return to after each loop iteration
};
// TODO: Is there a maximal size for this?
@@ -105,9 +109,16 @@ static void ProcessShaderCode(VertexShaderState& state) {
while (true) {
if (!state.call_stack.empty()) {
- if (state.program_counter - shader_memory.data() == state.call_stack.top().final_address) {
- state.program_counter = &shader_memory[state.call_stack.top().return_address];
- state.call_stack.pop();
+ auto& top = state.call_stack.top();
+ if (state.program_counter - shader_memory.data() == top.final_address) {
+ state.address_registers[2] += top.loop_increment;
+
+ if (top.repeat_counter-- == 0) {
+ state.program_counter = &shader_memory[top.return_address];
+ state.call_stack.pop();
+ } else {
+ state.program_counter = &shader_memory[top.loop_address];
+ }
// TODO: Is "trying again" accurate to hardware?
continue;
@@ -118,9 +129,10 @@ static void ProcessShaderCode(VertexShaderState& state) {
const Instruction& instr = *(const Instruction*)state.program_counter;
const SwizzlePattern& swizzle = *(SwizzlePattern*)&swizzle_data[instr.common.operand_desc_id];
- auto call = [&](VertexShaderState& state, u32 offset, u32 num_instructions, u32 return_offset) {
+ static auto call = [](VertexShaderState& state, u32 offset, u32 num_instructions,
+ u32 return_offset, u8 repeat_count, u8 loop_increment) {
state.program_counter = &shader_memory[offset] - 1; // -1 to make sure when incrementing the PC we end up at the correct offset
- state.call_stack.push({ offset + num_instructions, return_offset });
+ state.call_stack.push({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset });
};
u32 binary_offset = state.program_counter - shader_memory.data();
@@ -457,7 +469,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- binary_offset + 1);
+ binary_offset + 1, 0, 0);
break;
case Instruction::OpCode::CALLU:
@@ -465,7 +477,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- binary_offset + 1);
+ binary_offset + 1, 0, 0);
}
break;
@@ -474,7 +486,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- binary_offset + 1);
+ binary_offset + 1, 0, 0);
}
break;
@@ -486,12 +498,12 @@ static void ProcessShaderCode(VertexShaderState& state) {
call(state,
binary_offset + 1,
instr.flow_control.dest_offset - binary_offset - 1,
- instr.flow_control.dest_offset + instr.flow_control.num_instructions);
+ instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
} else {
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- instr.flow_control.dest_offset + instr.flow_control.num_instructions);
+ instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
}
break;
@@ -504,17 +516,30 @@ static void ProcessShaderCode(VertexShaderState& state) {
call(state,
binary_offset + 1,
instr.flow_control.dest_offset - binary_offset - 1,
- instr.flow_control.dest_offset + instr.flow_control.num_instructions);
+ instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
} else {
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- instr.flow_control.dest_offset + instr.flow_control.num_instructions);
+ instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
}
break;
}
+ case Instruction::OpCode::LOOP:
+ {
+ state.address_registers[2] = shader_uniforms.i[instr.flow_control.int_uniform_id].y;
+
+ call(state,
+ binary_offset + 1,
+ instr.flow_control.dest_offset - binary_offset + 1,
+ instr.flow_control.dest_offset + 1,
+ shader_uniforms.i[instr.flow_control.int_uniform_id].x,
+ shader_uniforms.i[instr.flow_control.int_uniform_id].z);
+ break;
+ }
+
default:
LOG_ERROR(HW_GPU, "Unhandled instruction: 0x%02x (%s): 0x%08x",
(int)instr.opcode.Value(), instr.opcode.GetInfo().name, instr.hex);
diff --git a/src/video_core/video_core.cpp b/src/video_core/video_core.cpp
index 0a236595c..b9d4ede3a 100644
--- a/src/video_core/video_core.cpp
+++ b/src/video_core/video_core.cpp
@@ -18,7 +18,6 @@ namespace VideoCore {
EmuWindow* g_emu_window = nullptr; ///< Frontend emulator window
RendererBase* g_renderer = nullptr; ///< Renderer plugin
-int g_current_frame = 0;
/// Initialize the video core
void Init(EmuWindow* emu_window) {
@@ -27,8 +26,6 @@ void Init(EmuWindow* emu_window) {
g_renderer->SetWindow(g_emu_window);
g_renderer->Init();
- g_current_frame = 0;
-
LOG_DEBUG(Render, "initialized OK");
}
diff --git a/src/video_core/video_core.h b/src/video_core/video_core.h
index b782f17bd..1b51d39bf 100644
--- a/src/video_core/video_core.h
+++ b/src/video_core/video_core.h
@@ -30,7 +30,6 @@ static const int kScreenBottomHeight = 240; ///< 3DS bottom screen height
// ---------------------
extern RendererBase* g_renderer; ///< Renderer plugin
-extern int g_current_frame; ///< Current frame
extern EmuWindow* g_emu_window; ///< Emu window
/// Start the video core
generated by cgit v1.2.3 (git 2.25.1) at 2025-12-26 21:18:30 +0000