8 files changed, 290 insertions, 113 deletions

diff --git a/src/video_core/renderer_opengl/blit_image.cpp b/src/video_core/renderer_opengl/blit_image.cpp
new file mode 100644
index 000000000..9a560a73b
--- /dev/null
+++ b/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
diff --git a/src/video_core/renderer_opengl/blit_image.h b/src/video_core/renderer_opengl/blit_image.h
new file mode 100644
index 000000000..5a2b12d16
--- /dev/null
+++ b/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
diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp
index cee5c3247..22ed16ebf 100644
--- a/src/video_core/renderer_opengl/gl_device.cpp
+++ b/src/video_core/renderer_opengl/gl_device.cpp
@@ -166,6 +166,7 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) {
     has_shader_int64 = HasExtension(extensions, "GL_ARB_gpu_shader_int64");
     has_amd_shader_half_float = GLAD_GL_AMD_gpu_shader_half_float;
     has_sparse_texture_2 = GLAD_GL_ARB_sparse_texture2;
+    has_draw_texture = GLAD_GL_NV_draw_texture;
     warp_size_potentially_larger_than_guest = !is_nvidia && !is_intel;
     need_fastmath_off = is_nvidia;
     can_report_memory = GLAD_GL_NVX_gpu_memory_info;
diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h
index 2a72d84be..3ff8cad83 100644
--- a/src/video_core/renderer_opengl/gl_device.h
+++ b/src/video_core/renderer_opengl/gl_device.h
@@ -4,6 +4,8 @@
 #pragma once
 
 #include <cstddef>
+#include <string>
+
 #include "common/common_types.h"
 #include "core/frontend/emu_window.h"
 #include "shader_recompiler/stage.h"
@@ -146,6 +148,10 @@ public:
         return has_sparse_texture_2;
     }
 
+    bool HasDrawTexture() const {
+        return has_draw_texture;
+    }
+
     bool IsWarpSizePotentiallyLargerThanGuest() const {
         return warp_size_potentially_larger_than_guest;
     }
@@ -216,6 +222,7 @@ private:
     bool has_shader_int64{};
     bool has_amd_shader_half_float{};
     bool has_sparse_texture_2{};
+    bool has_draw_texture{};
     bool warp_size_potentially_larger_than_guest{};
     bool need_fastmath_off{};
     bool has_cbuf_ftou_bug{};
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 181857d9c..7bced675c 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -64,7 +64,8 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra
       shader_cache(*this, emu_window_, device, texture_cache, buffer_cache, program_manager,
                    state_tracker, gpu.ShaderNotify()),
       query_cache(*this), accelerate_dma(buffer_cache),
-      fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache) {}
+      fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache),
+      blit_image(program_manager_) {}
 
 RasterizerOpenGL::~RasterizerOpenGL() = default;
 
@@ -320,6 +321,47 @@ void RasterizerOpenGL::DrawIndirect() {
     buffer_cache.SetDrawIndirect(nullptr);
 }
 
+void RasterizerOpenGL::DrawTexture() {
+    MICROPROFILE_SCOPE(OpenGL_Drawing);
+
+    SCOPE_EXIT({ gpu.TickWork(); });
+    query_cache.UpdateCounters();
+
+    texture_cache.SynchronizeGraphicsDescriptors();
+    texture_cache.UpdateRenderTargets(false);
+
+    SyncState();
+
+    const auto& draw_texture_state = maxwell3d->draw_manager->GetDrawTextureState();
+    const auto& sampler = texture_cache.GetGraphicsSampler(draw_texture_state.src_sampler);
+    const auto& texture = texture_cache.GetImageView(draw_texture_state.src_texture);
+
+    if (device.HasDrawTexture()) {
+        state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle());
+
+        glDrawTextureNV(texture.DefaultHandle(), sampler->Handle(), draw_texture_state.dst_x0,
+                        draw_texture_state.dst_y0, draw_texture_state.dst_x1,
+                        draw_texture_state.dst_y1, 0,
+                        draw_texture_state.src_x0 / static_cast<float>(texture.size.width),
+                        draw_texture_state.src_y0 / static_cast<float>(texture.size.height),
+                        draw_texture_state.src_x1 / static_cast<float>(texture.size.width),
+                        draw_texture_state.src_y1 / static_cast<float>(texture.size.height));
+    } else {
+        Region2D dst_region = {Offset2D{.x = static_cast<s32>(draw_texture_state.dst_x0),
+                                        .y = static_cast<s32>(draw_texture_state.dst_y0)},
+                               Offset2D{.x = static_cast<s32>(draw_texture_state.dst_x1),
+                                        .y = static_cast<s32>(draw_texture_state.dst_y1)}};
+        Region2D src_region = {Offset2D{.x = static_cast<s32>(draw_texture_state.src_x0),
+                                        .y = static_cast<s32>(draw_texture_state.src_y0)},
+                               Offset2D{.x = static_cast<s32>(draw_texture_state.src_x1),
+                                        .y = static_cast<s32>(draw_texture_state.src_y1)}};
+        blit_image.BlitColor(texture_cache.GetFramebuffer()->Handle(), texture.DefaultHandle(),
+                             sampler->Handle(), dst_region, src_region, texture.size);
+    }
+
+    ++num_queued_commands;
+}
+
 void RasterizerOpenGL::DispatchCompute() {
     gpu_memory->FlushCaching();
     ComputePipeline* const pipeline{shader_cache.CurrentComputePipeline()};
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index be4f76c18..0c45832ae 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -16,6 +16,7 @@
 #include "video_core/engines/maxwell_dma.h"
 #include "video_core/rasterizer_accelerated.h"
 #include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_opengl/blit_image.h"
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_fence_manager.h"
@@ -70,6 +71,7 @@ public:
 
     void Draw(bool is_indexed, u32 instance_count) override;
     void DrawIndirect() override;
+    void DrawTexture() override;
     void Clear(u32 layer_count) override;
     void DispatchCompute() override;
     void ResetCounter(VideoCore::QueryType type) override;
@@ -224,6 +226,8 @@ private:
     AccelerateDMA accelerate_dma;
     FenceManagerOpenGL fence_manager;
 
+    BlitImageHelper blit_image;
+
     boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices;
     std::array<ImageViewId, MAX_IMAGE_VIEWS> image_view_ids;
     boost::container::static_vector<GLuint, MAX_TEXTURES> sampler_handles;
diff --git a/src/video_core/renderer_opengl/gl_shader_manager.cpp b/src/video_core/renderer_opengl/gl_shader_manager.cpp
index d9c29d8b7..98841ae65 100644
--- a/src/video_core/renderer_opengl/gl_shader_manager.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp
@@ -1,2 +1,123 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <glad/glad.h>
+
+#include "video_core/renderer_opengl/gl_shader_manager.h"
+
+namespace OpenGL {
+
+static constexpr std::array ASSEMBLY_PROGRAM_ENUMS{
+    GL_VERTEX_PROGRAM_NV,   GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV,
+    GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV,
+};
+
+ProgramManager::ProgramManager(const Device& device) {
+    glCreateProgramPipelines(1, &pipeline.handle);
+    if (device.UseAssemblyShaders()) {
+        glEnable(GL_COMPUTE_PROGRAM_NV);
+    }
+}
+
+void ProgramManager::BindComputeProgram(GLuint program) {
+    glUseProgram(program);
+    is_compute_bound = true;
+}
+
+void ProgramManager::BindComputeAssemblyProgram(GLuint program) {
+    if (current_assembly_compute_program != program) {
+        current_assembly_compute_program = program;
+        glBindProgramARB(GL_COMPUTE_PROGRAM_NV, program);
+    }
+    UnbindPipeline();
+}
+
+void ProgramManager::BindSourcePrograms(std::span<const OGLProgram, NUM_STAGES> programs) {
+    static constexpr std::array<GLenum, 5> stage_enums{
+        GL_VERTEX_SHADER_BIT,   GL_TESS_CONTROL_SHADER_BIT, GL_TESS_EVALUATION_SHADER_BIT,
+        GL_GEOMETRY_SHADER_BIT, GL_FRAGMENT_SHADER_BIT,
+    };
+    for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
+        if (current_programs[stage] != programs[stage].handle) {
+            current_programs[stage] = programs[stage].handle;
+            glUseProgramStages(pipeline.handle, stage_enums[stage], programs[stage].handle);
+        }
+    }
+    BindPipeline();
+}
+
+void ProgramManager::BindPresentPrograms(GLuint vertex, GLuint fragment) {
+    if (current_programs[0] != vertex) {
+        current_programs[0] = vertex;
+        glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, vertex);
+    }
+    if (current_programs[4] != fragment) {
+        current_programs[4] = fragment;
+        glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, fragment);
+    }
+    glUseProgramStages(
+        pipeline.handle,
+        GL_TESS_CONTROL_SHADER_BIT | GL_TESS_EVALUATION_SHADER_BIT | GL_GEOMETRY_SHADER_BIT, 0);
+    current_programs[1] = 0;
+    current_programs[2] = 0;
+    current_programs[3] = 0;
+
+    if (current_stage_mask != 0) {
+        current_stage_mask = 0;
+        for (const GLenum program_type : ASSEMBLY_PROGRAM_ENUMS) {
+            glDisable(program_type);
+        }
+    }
+    BindPipeline();
+}
+
+void ProgramManager::BindAssemblyPrograms(std::span<const OGLAssemblyProgram, NUM_STAGES> programs,
+                                          u32 stage_mask) {
+    const u32 changed_mask = current_stage_mask ^ stage_mask;
+    current_stage_mask = stage_mask;
+
+    if (changed_mask != 0) {
+        for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
+            if (((changed_mask >> stage) & 1) != 0) {
+                if (((stage_mask >> stage) & 1) != 0) {
+                    glEnable(ASSEMBLY_PROGRAM_ENUMS[stage]);
+                } else {
+                    glDisable(ASSEMBLY_PROGRAM_ENUMS[stage]);
+                }
+            }
+        }
+    }
+    for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
+        if (current_programs[stage] != programs[stage].handle) {
+            current_programs[stage] = programs[stage].handle;
+            glBindProgramARB(ASSEMBLY_PROGRAM_ENUMS[stage], programs[stage].handle);
+        }
+    }
+    UnbindPipeline();
+}
+
+void ProgramManager::RestoreGuestCompute() {}
+
+void ProgramManager::BindPipeline() {
+    if (!is_pipeline_bound) {
+        is_pipeline_bound = true;
+        glBindProgramPipeline(pipeline.handle);
+    }
+    UnbindCompute();
+}
+
+void ProgramManager::UnbindPipeline() {
+    if (is_pipeline_bound) {
+        is_pipeline_bound = false;
+        glBindProgramPipeline(0);
+    }
+    UnbindCompute();
+}
+
+void ProgramManager::UnbindCompute() {
+    if (is_compute_bound) {
+        is_compute_bound = false;
+        glUseProgram(0);
+    }
+}
+} // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h
index a84f5aeb3..07ffab77f 100644
--- a/src/video_core/renderer_opengl/gl_shader_manager.h
+++ b/src/video_core/renderer_opengl/gl_shader_manager.h
@@ -6,8 +6,6 @@
 #include <array>
 #include <span>
 
-#include <glad/glad.h>
-
 #include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 
@@ -16,121 +14,28 @@ namespace OpenGL {
 class ProgramManager {
     static constexpr size_t NUM_STAGES = 5;
 
-    static constexpr std::array ASSEMBLY_PROGRAM_ENUMS{
-        GL_VERTEX_PROGRAM_NV,   GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV,
-        GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV,
-    };
-
 public:
-    explicit ProgramManager(const Device& device) {
-        glCreateProgramPipelines(1, &pipeline.handle);
-        if (device.UseAssemblyShaders()) {
-            glEnable(GL_COMPUTE_PROGRAM_NV);
-        }
-    }
-
-    void BindComputeProgram(GLuint program) {
-        glUseProgram(program);
-        is_compute_bound = true;
-    }
-
-    void BindComputeAssemblyProgram(GLuint program) {
-        if (current_assembly_compute_program != program) {
-            current_assembly_compute_program = program;
-            glBindProgramARB(GL_COMPUTE_PROGRAM_NV, program);
-        }
-        UnbindPipeline();
-    }
-
-    void BindSourcePrograms(std::span<const OGLProgram, NUM_STAGES> programs) {
-        static constexpr std::array<GLenum, 5> stage_enums{
-            GL_VERTEX_SHADER_BIT,   GL_TESS_CONTROL_SHADER_BIT, GL_TESS_EVALUATION_SHADER_BIT,
-            GL_GEOMETRY_SHADER_BIT, GL_FRAGMENT_SHADER_BIT,
-        };
-        for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
-            if (current_programs[stage] != programs[stage].handle) {
-                current_programs[stage] = programs[stage].handle;
-                glUseProgramStages(pipeline.handle, stage_enums[stage], programs[stage].handle);
-            }
-        }
-        BindPipeline();
-    }
-
-    void BindPresentPrograms(GLuint vertex, GLuint fragment) {
-        if (current_programs[0] != vertex) {
-            current_programs[0] = vertex;
-            glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, vertex);
-        }
-        if (current_programs[4] != fragment) {
-            current_programs[4] = fragment;
-            glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, fragment);
-        }
-        glUseProgramStages(
-            pipeline.handle,
-            GL_TESS_CONTROL_SHADER_BIT | GL_TESS_EVALUATION_SHADER_BIT | GL_GEOMETRY_SHADER_BIT, 0);
-        current_programs[1] = 0;
-        current_programs[2] = 0;
-        current_programs[3] = 0;
-
-        if (current_stage_mask != 0) {
-            current_stage_mask = 0;
-            for (const GLenum program_type : ASSEMBLY_PROGRAM_ENUMS) {
-                glDisable(program_type);
-            }
-        }
-        BindPipeline();
-    }
+    explicit ProgramManager(const Device& device);
+
+    void BindComputeProgram(GLuint program);
+
+    void BindComputeAssemblyProgram(GLuint program);
+
+    void BindSourcePrograms(std::span<const OGLProgram, NUM_STAGES> programs);
+
+    void BindPresentPrograms(GLuint vertex, GLuint fragment);
 
     void BindAssemblyPrograms(std::span<const OGLAssemblyProgram, NUM_STAGES> programs,
-                              u32 stage_mask) {
-        const u32 changed_mask = current_stage_mask ^ stage_mask;
-        current_stage_mask = stage_mask;
-
-        if (changed_mask != 0) {
-            for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
-                if (((changed_mask >> stage) & 1) != 0) {
-                    if (((stage_mask >> stage) & 1) != 0) {
-                        glEnable(ASSEMBLY_PROGRAM_ENUMS[stage]);
-                    } else {
-                        glDisable(ASSEMBLY_PROGRAM_ENUMS[stage]);
-                    }
-                }
-            }
-        }
-        for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
-            if (current_programs[stage] != programs[stage].handle) {
-                current_programs[stage] = programs[stage].handle;
-                glBindProgramARB(ASSEMBLY_PROGRAM_ENUMS[stage], programs[stage].handle);
-            }
-        }
-        UnbindPipeline();
-    }
-
-    void RestoreGuestCompute() {}
+                              u32 stage_mask);
+
+    void RestoreGuestCompute();
 
 private:
-    void BindPipeline() {
-        if (!is_pipeline_bound) {
-            is_pipeline_bound = true;
-            glBindProgramPipeline(pipeline.handle);
-        }
-        UnbindCompute();
-    }
-
-    void UnbindPipeline() {
-        if (is_pipeline_bound) {
-            is_pipeline_bound = false;
-            glBindProgramPipeline(0);
-        }
-        UnbindCompute();
-    }
-
-    void UnbindCompute() {
-        if (is_compute_bound) {
-            is_compute_bound = false;
-            glUseProgram(0);
-        }
-    }
+    void BindPipeline();
+
+    void UnbindPipeline();
+
+    void UnbindCompute();
 
     OGLPipeline pipeline;
     bool is_pipeline_bound{};