video_core: Implement maxwell3d draw texture method

src/video_core/engines/draw_manager.cpp

@@ -51,6 +51,10 @@ void DrawManager::ProcessMethodCall(u32 method, u32 argument) {
         LOG_WARNING(HW_GPU, "(STUBBED) called");
         break;
     }
+    case MAXWELL3D_REG_INDEX(draw_texture.src_y0): {
+        DrawTexture();
+        break;
+    }
     default:
         break;
     }
@@ -179,6 +183,33 @@ void DrawManager::DrawIndexSmall(u32 argument) {
     ProcessDraw(true, 1);
 }
 
+void DrawManager::DrawTexture() {
+    const auto& regs{maxwell3d->regs};
+    draw_texture_state.dst_x0 = static_cast<float>(regs.draw_texture.dst_x0) / 4096.f;
+    draw_texture_state.dst_y0 = static_cast<float>(regs.draw_texture.dst_y0) / 4096.f;
+    const auto dst_width = static_cast<float>(regs.draw_texture.dst_width) / 4096.f;
+    const auto dst_height = static_cast<float>(regs.draw_texture.dst_height) / 4096.f;
+    const bool lower_left{regs.window_origin.mode !=
+                          Maxwell3D::Regs::WindowOrigin::Mode::UpperLeft};
+    if (lower_left) {
+        draw_texture_state.dst_y0 -= dst_height;
+    }
+    draw_texture_state.dst_x1 = draw_texture_state.dst_x0 + dst_width;
+    draw_texture_state.dst_y1 = draw_texture_state.dst_y0 + dst_height;
+    draw_texture_state.src_x0 = static_cast<float>(regs.draw_texture.src_x0) / 4096.f;
+    draw_texture_state.src_y0 = static_cast<float>(regs.draw_texture.src_y0) / 4096.f;
+    draw_texture_state.src_x1 =
+        (static_cast<float>(regs.draw_texture.dx_du) / 4294967295.f) * dst_width +
+        draw_texture_state.src_x0;
+    draw_texture_state.src_y1 =
+        (static_cast<float>(regs.draw_texture.dy_dv) / 4294967295.f) * dst_height +
+        draw_texture_state.src_y0;
+    draw_texture_state.src_sampler = regs.draw_texture.src_sampler;
+    draw_texture_state.src_texture = regs.draw_texture.src_texture;
+
+    maxwell3d->rasterizer->DrawTexture();
+}
+
 void DrawManager::UpdateTopology() {
     const auto& regs{maxwell3d->regs};
     switch (regs.primitive_topology_control) {

src/video_core/engines/draw_manager.h

@@ -32,6 +32,19 @@ public:
         std::vector<u8> inline_index_draw_indexes;
     };
 
+    struct DrawTextureState {
+        f32 dst_x0;
+        f32 dst_y0;
+        f32 dst_x1;
+        f32 dst_y1;
+        f32 src_x0;
+        f32 src_y0;
+        f32 src_x1;
+        f32 src_y1;
+        u32 src_sampler;
+        u32 src_texture;
+    };
+
     struct IndirectParams {
         bool is_indexed;
         bool include_count;
@@ -64,6 +77,10 @@ public:
         return draw_state;
     }
 
+    const DrawTextureState& GetDrawTextureState() const {
+        return draw_texture_state;
+    }
+
     IndirectParams& GetIndirectParams() {
         return indirect_state;
     }
@@ -81,6 +98,8 @@ private:
 
     void DrawIndexSmall(u32 argument);
 
+    void DrawTexture();
+
     void UpdateTopology();
 
     void ProcessDraw(bool draw_indexed, u32 instance_count);
@@ -89,6 +108,7 @@ private:
 
     Maxwell3D* maxwell3d{};
     State draw_state{};
+    DrawTextureState draw_texture_state{};
     IndirectParams indirect_state{};
 };
 } // namespace Tegra::Engines

src/video_core/engines/maxwell_3d.cpp

@@ -149,6 +149,7 @@ bool Maxwell3D::IsMethodExecutable(u32 method) {
     case MAXWELL3D_REG_INDEX(inline_index_4x8.index0):
     case MAXWELL3D_REG_INDEX(vertex_array_instance_first):
     case MAXWELL3D_REG_INDEX(vertex_array_instance_subsequent):
+    case MAXWELL3D_REG_INDEX(draw_texture.src_y0):
     case MAXWELL3D_REG_INDEX(wait_for_idle):
     case MAXWELL3D_REG_INDEX(shadow_ram_control):
     case MAXWELL3D_REG_INDEX(load_mme.instruction_ptr):

src/video_core/engines/maxwell_3d.h

@@ -1599,6 +1599,20 @@ public:
         };
         static_assert(sizeof(TIRModulationCoeff) == 0x4);
 
+        struct DrawTexture {
+            s32 dst_x0;
+            s32 dst_y0;
+            s32 dst_width;
+            s32 dst_height;
+            s64 dx_du;
+            s64 dy_dv;
+            u32 src_sampler;
+            u32 src_texture;
+            s32 src_x0;
+            s32 src_y0;
+        };
+        static_assert(sizeof(DrawTexture) == 0x30);
+
         struct ReduceColorThreshold {
             union {
                 BitField<0, 8, u32> all_hit_once;
@@ -2751,7 +2765,7 @@ public:
                 u32 reserved_sw_method2;                                               ///< 0x102C
                 std::array<TIRModulationCoeff, 5> tir_modulation_coeff;                ///< 0x1030
                 std::array<u32, 15> spare_nop;                                         ///< 0x1044
-                INSERT_PADDING_BYTES_NOINIT(0x30);
+                DrawTexture draw_texture;                                              ///< 0x1080
                 std::array<u32, 7> reserved_sw_method3_to_7;                           ///< 0x10B0
                 ReduceColorThreshold reduce_color_thresholds_unorm8;                   ///< 0x10CC
                 std::array<u32, 4> reserved_sw_method10_to_13;                         ///< 0x10D0

src/video_core/host_shaders/blit_color_float.frag

@@ -0,0 +1,13 @@
+// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#version 450
+
+layout(binding = 0) uniform sampler2D tex;
+
+layout(location = 0) in vec2 texcoord;
+layout(location = 0) out vec4 color;
+
+void main() {
+    color = textureLod(tex, texcoord, 0);
+}

src/video_core/renderer_opengl/blit_image.cpp

@@ -0,0 +1,59 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <algorithm>
+
+#include "video_core/host_shaders/blit_color_float_frag.h"
+#include "video_core/host_shaders/full_screen_triangle_vert.h"
+#include "video_core/renderer_opengl/blit_image.h"
+#include "video_core/renderer_opengl/gl_shader_manager.h"
+#include "video_core/renderer_opengl/gl_shader_util.h"
+
+namespace OpenGL {
+
+BlitImageHelper::BlitImageHelper(ProgramManager& program_manager_)
+    : program_manager(program_manager_),
+      full_screen_vert(CreateProgram(HostShaders::FULL_SCREEN_TRIANGLE_VERT, GL_VERTEX_SHADER)),
+      blit_color_to_color_frag(
+          CreateProgram(HostShaders::BLIT_COLOR_FLOAT_FRAG, GL_FRAGMENT_SHADER)) {}
+
+BlitImageHelper::~BlitImageHelper() = default;
+
+void BlitImageHelper::BlitColor(GLuint dst_framebuffer, GLuint src_image_view, GLuint src_sampler,
+                                const Region2D& dst_region, const Region2D& src_region,
+                                const Extent3D& src_size) {
+    glEnable(GL_CULL_FACE);
+    glDisable(GL_COLOR_LOGIC_OP);
+    glDisable(GL_DEPTH_TEST);
+    glDisable(GL_STENCIL_TEST);
+    glDisable(GL_POLYGON_OFFSET_FILL);
+    glDisable(GL_RASTERIZER_DISCARD);
+    glDisable(GL_ALPHA_TEST);
+    glDisablei(GL_BLEND, 0);
+    glPolygonMode(GL_FRONT_AND_BACK, GL_FILL);
+    glCullFace(GL_BACK);
+    glFrontFace(GL_CW);
+    glColorMaski(0, GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
+    glDepthRangeIndexed(0, 0.0, 0.0);
+
+    program_manager.BindPresentPrograms(full_screen_vert.handle, blit_color_to_color_frag.handle);
+    glProgramUniform2f(full_screen_vert.handle, 0,
+                       static_cast<float>(src_region.end.x - src_region.start.x) /
+                           static_cast<float>(src_size.width),
+                       static_cast<float>(src_region.end.y - src_region.start.y) /
+                           static_cast<float>(src_size.height));
+    glProgramUniform2f(full_screen_vert.handle, 1,
+                       static_cast<float>(src_region.start.x) / static_cast<float>(src_size.width),
+                       static_cast<float>(src_region.start.y) /
+                           static_cast<float>(src_size.height));
+    glViewport(std::min(dst_region.start.x, dst_region.end.x),
+               std::min(dst_region.start.y, dst_region.end.y),
+               std::abs(dst_region.end.x - dst_region.start.x),
+               std::abs(dst_region.end.y - dst_region.start.y));
+    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, dst_framebuffer);
+    glBindSampler(0, src_sampler);
+    glBindTextureUnit(0, src_image_view);
+    glClear(GL_COLOR_BUFFER_BIT);
+    glDrawArrays(GL_TRIANGLES, 0, 3);
+}
+} // namespace OpenGL

src/video_core/renderer_opengl/blit_image.h

@@ -0,0 +1,38 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <glad/glad.h>
+
+#include "video_core/engines/fermi_2d.h"
+#include "video_core/renderer_opengl/gl_resource_manager.h"
+#include "video_core/texture_cache/types.h"
+
+namespace OpenGL {
+
+using VideoCommon::Extent3D;
+using VideoCommon::Offset2D;
+using VideoCommon::Region2D;
+
+class ProgramManager;
+class Framebuffer;
+class ImageView;
+
+class BlitImageHelper {
+public:
+    explicit BlitImageHelper(ProgramManager& program_manager);
+    ~BlitImageHelper();
+
+    void BlitColor(GLuint dst_framebuffer, GLuint src_image_view, GLuint src_sampler,
+                   const Region2D& dst_region, const Region2D& src_region,
+                   const Extent3D& src_size);
+
+private:
+    ProgramManager& program_manager;
+
+    OGLProgram full_screen_vert;
+    OGLProgram blit_color_to_color_frag;
+};
+
+} // namespace OpenGL