Merge remote-tracking branch 'upstream/master' into int-flags

18 files changed, 674 insertions, 21 deletions

diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt
index aa62363a7..4c7399d5a 100644
--- a/src/video_core/host_shaders/CMakeLists.txt
+++ b/src/video_core/host_shaders/CMakeLists.txt
@@ -1,18 +1,29 @@
 set(SHADER_FILES
+    block_linear_unswizzle_2d.comp
+    block_linear_unswizzle_3d.comp
+    convert_depth_to_float.frag
+    convert_float_to_depth.frag
+    full_screen_triangle.vert
+    opengl_copy_bc4.comp
     opengl_present.frag
     opengl_present.vert
+    pitch_unswizzle.comp
+    vulkan_blit_color_float.frag
+    vulkan_blit_depth_stencil.frag
+    vulkan_present.frag
+    vulkan_present.vert
+    vulkan_quad_array.comp
+    vulkan_quad_indexed.comp
+    vulkan_uint8.comp
 )
 
-set(SHADER_INCLUDE ${CMAKE_CURRENT_BINARY_DIR}/include)
-set(HOST_SHADERS_INCLUDE ${SHADER_INCLUDE} PARENT_SCOPE)
+find_program(GLSLANGVALIDATOR "glslangValidator" REQUIRED)
+
+set(GLSL_FLAGS "")
 
+set(SHADER_INCLUDE ${CMAKE_CURRENT_BINARY_DIR}/include)
 set(SHADER_DIR ${SHADER_INCLUDE}/video_core/host_shaders)
-add_custom_command(
-    OUTPUT
-        ${SHADER_DIR}
-    COMMAND
-        ${CMAKE_COMMAND} -E make_directory ${SHADER_DIR}
-)
+set(HOST_SHADERS_INCLUDE ${SHADER_INCLUDE} PARENT_SCOPE)
 
 set(INPUT_FILE ${CMAKE_CURRENT_SOURCE_DIR}/source_shader.h.in)
 set(HEADER_GENERATOR ${CMAKE_CURRENT_SOURCE_DIR}/StringShaderHeader.cmake)
@@ -20,19 +31,36 @@ set(HEADER_GENERATOR ${CMAKE_CURRENT_SOURCE_DIR}/StringShaderHeader.cmake)
 foreach(FILENAME IN ITEMS ${SHADER_FILES})
     string(REPLACE "." "_" SHADER_NAME ${FILENAME})
     set(SOURCE_FILE ${CMAKE_CURRENT_SOURCE_DIR}/${FILENAME})
-    set(HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}.h)
-    add_custom_command(
-        OUTPUT
-            ${HEADER_FILE}
-        COMMAND
-            ${CMAKE_COMMAND} -P ${HEADER_GENERATOR} ${SOURCE_FILE} ${HEADER_FILE} ${INPUT_FILE}
-        MAIN_DEPENDENCY
-            ${SOURCE_FILE}
-        DEPENDS
-            ${HEADER_GENERATOR}
-            ${INPUT_FILE}
-    )
-    set(SHADER_HEADERS ${SHADER_HEADERS} ${HEADER_FILE})
+    # Skip generating source headers on Vulkan exclusive files
+    if (NOT ${FILENAME} MATCHES "vulkan.*")
+        set(SOURCE_HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}.h)
+        add_custom_command(
+            OUTPUT
+                ${SOURCE_HEADER_FILE}
+            COMMAND
+                ${CMAKE_COMMAND} -P ${HEADER_GENERATOR} ${SOURCE_FILE} ${SOURCE_HEADER_FILE} ${INPUT_FILE}
+            MAIN_DEPENDENCY
+                ${SOURCE_FILE}
+            DEPENDS
+                ${INPUT_FILE}
+                # HEADER_GENERATOR should be included here but msbuild seems to assume it's always modified
+        )
+        set(SHADER_HEADERS ${SHADER_HEADERS} ${SOURCE_HEADER_FILE})
+    endif()
+    # Skip compiling to SPIR-V OpenGL exclusive files
+    if (NOT ${FILENAME} MATCHES "opengl.*")
+        string(TOUPPER ${SHADER_NAME}_SPV SPIRV_VARIABLE_NAME)
+        set(SPIRV_HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}_spv.h)
+        add_custom_command(
+            OUTPUT
+                ${SPIRV_HEADER_FILE}
+            COMMAND
+                ${GLSLANGVALIDATOR} -V ${GLSL_FLAGS} --variable-name ${SPIRV_VARIABLE_NAME} -o ${SPIRV_HEADER_FILE} ${SOURCE_FILE}
+            MAIN_DEPENDENCY
+                ${SOURCE_FILE}
+        )
+        set(SHADER_HEADERS ${SHADER_HEADERS} ${SPIRV_HEADER_FILE})
+    endif()
 endforeach()
 
 add_custom_target(host_shaders
diff --git a/src/video_core/host_shaders/StringShaderHeader.cmake b/src/video_core/host_shaders/StringShaderHeader.cmake
index 368bce0ed..c0fc49768 100644
--- a/src/video_core/host_shaders/StringShaderHeader.cmake
+++ b/src/video_core/host_shaders/StringShaderHeader.cmake
@@ -8,4 +8,6 @@ string(TOUPPER ${CONTENTS_NAME} CONTENTS_NAME)
 
 file(READ ${SOURCE_FILE} CONTENTS)
 
+get_filename_component(OUTPUT_DIR ${HEADER_FILE} DIRECTORY)
+make_directory(${OUTPUT_DIR})
 configure_file(${INPUT_FILE} ${HEADER_FILE} @ONLY)
diff --git a/src/video_core/host_shaders/block_linear_unswizzle_2d.comp b/src/video_core/host_shaders/block_linear_unswizzle_2d.comp
new file mode 100644
index 000000000..a131be79e
--- /dev/null
+++ b/src/video_core/host_shaders/block_linear_unswizzle_2d.comp
@@ -0,0 +1,122 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 430
+
+#ifdef VULKAN
+
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+#define HAS_EXTENDED_TYPES 1
+#define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
+#define END_PUSH_CONSTANTS };
+#define UNIFORM(n)
+#define BINDING_SWIZZLE_BUFFER 0
+#define BINDING_INPUT_BUFFER 1
+#define BINDING_OUTPUT_IMAGE 2
+
+#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
+
+#extension GL_NV_gpu_shader5 : enable
+#ifdef GL_NV_gpu_shader5
+#define HAS_EXTENDED_TYPES 1
+#else
+#define HAS_EXTENDED_TYPES 0
+#endif
+#define BEGIN_PUSH_CONSTANTS
+#define END_PUSH_CONSTANTS
+#define UNIFORM(n) layout (location = n) uniform
+#define BINDING_SWIZZLE_BUFFER 0
+#define BINDING_INPUT_BUFFER 1
+#define BINDING_OUTPUT_IMAGE 0
+
+#endif
+
+BEGIN_PUSH_CONSTANTS
+UNIFORM(0) uvec3 origin;
+UNIFORM(1) ivec3 destination;
+UNIFORM(2) uint bytes_per_block_log2;
+UNIFORM(3) uint layer_stride;
+UNIFORM(4) uint block_size;
+UNIFORM(5) uint x_shift;
+UNIFORM(6) uint block_height;
+UNIFORM(7) uint block_height_mask;
+END_PUSH_CONSTANTS
+
+layout(binding = BINDING_SWIZZLE_BUFFER, std430) readonly buffer SwizzleTable {
+    uint swizzle_table[];
+};
+
+#if HAS_EXTENDED_TYPES
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU8 { uint8_t u8data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU16 { uint16_t u16data[]; };
+#endif
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32 { uint u32data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU64 { uvec2 u64data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u128data[]; };
+
+layout(binding = BINDING_OUTPUT_IMAGE) uniform writeonly uimage2DArray output_image;
+
+layout(local_size_x = 32, local_size_y = 32, local_size_z = 1) in;
+
+const uint GOB_SIZE_X = 64;
+const uint GOB_SIZE_Y = 8;
+const uint GOB_SIZE_Z = 1;
+const uint GOB_SIZE = GOB_SIZE_X * GOB_SIZE_Y * GOB_SIZE_Z;
+
+const uint GOB_SIZE_X_SHIFT = 6;
+const uint GOB_SIZE_Y_SHIFT = 3;
+const uint GOB_SIZE_Z_SHIFT = 0;
+const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHIFT;
+
+const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1, GOB_SIZE_Y - 1);
+
+uint SwizzleOffset(uvec2 pos) {
+    pos = pos & SWIZZLE_MASK;
+    return swizzle_table[pos.y * 64 + pos.x];
+}
+
+uvec4 ReadTexel(uint offset) {
+    switch (bytes_per_block_log2) {
+#if HAS_EXTENDED_TYPES
+    case 0:
+        return uvec4(u8data[offset], 0, 0, 0);
+    case 1:
+        return uvec4(u16data[offset / 2], 0, 0, 0);
+#else
+    case 0:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 24), 8), 0, 0, 0);
+    case 1:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 16), 16), 0, 0, 0);
+#endif
+    case 2:
+        return uvec4(u32data[offset / 4], 0, 0, 0);
+    case 3:
+        return uvec4(u64data[offset / 8], 0, 0);
+    case 4:
+        return u128data[offset / 16];
+    }
+    return uvec4(0);
+}
+
+void main() {
+    uvec3 pos = gl_GlobalInvocationID + origin;
+    pos.x <<= bytes_per_block_log2;
+
+    // Read as soon as possible due to its latency
+    const uint swizzle = SwizzleOffset(pos.xy);
+
+    const uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
+
+    uint offset = 0;
+    offset += pos.z * layer_stride;
+    offset += (block_y >> block_height) * block_size;
+    offset += (block_y & block_height_mask) << GOB_SIZE_SHIFT;
+    offset += (pos.x >> GOB_SIZE_X_SHIFT) << x_shift;
+    offset += swizzle;
+
+    const uvec4 texel = ReadTexel(offset);
+    const ivec3 coord = ivec3(gl_GlobalInvocationID) + destination;
+    imageStore(output_image, coord, texel);
+}
diff --git a/src/video_core/host_shaders/block_linear_unswizzle_3d.comp b/src/video_core/host_shaders/block_linear_unswizzle_3d.comp
new file mode 100644
index 000000000..bb6872e6b
--- /dev/null
+++ b/src/video_core/host_shaders/block_linear_unswizzle_3d.comp
@@ -0,0 +1,125 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 430
+
+#ifdef VULKAN
+
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+#define HAS_EXTENDED_TYPES 1
+#define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
+#define END_PUSH_CONSTANTS };
+#define UNIFORM(n)
+#define BINDING_SWIZZLE_BUFFER 0
+#define BINDING_INPUT_BUFFER 1
+#define BINDING_OUTPUT_IMAGE 2
+
+#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
+
+#extension GL_NV_gpu_shader5 : enable
+#ifdef GL_NV_gpu_shader5
+#define HAS_EXTENDED_TYPES 1
+#else
+#define HAS_EXTENDED_TYPES 0
+#endif
+#define BEGIN_PUSH_CONSTANTS
+#define END_PUSH_CONSTANTS
+#define UNIFORM(n) layout (location = n) uniform
+#define BINDING_SWIZZLE_BUFFER 0
+#define BINDING_INPUT_BUFFER 1
+#define BINDING_OUTPUT_IMAGE 0
+
+#endif
+
+BEGIN_PUSH_CONSTANTS
+UNIFORM(0) uvec3 origin;
+UNIFORM(1) ivec3 destination;
+UNIFORM(2) uint bytes_per_block_log2;
+UNIFORM(3) uint slice_size;
+UNIFORM(4) uint block_size;
+UNIFORM(5) uint x_shift;
+UNIFORM(6) uint block_height;
+UNIFORM(7) uint block_height_mask;
+UNIFORM(8) uint block_depth;
+UNIFORM(9) uint block_depth_mask;
+END_PUSH_CONSTANTS
+
+layout(binding = BINDING_SWIZZLE_BUFFER, std430) readonly buffer SwizzleTable {
+    uint swizzle_table[];
+};
+
+#if HAS_EXTENDED_TYPES
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU8 { uint8_t u8data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU16 { uint16_t u16data[]; };
+#endif
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32 { uint u32data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU64 { uvec2 u64data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u128data[]; };
+
+layout(binding = BINDING_OUTPUT_IMAGE) uniform writeonly uimage3D output_image;
+
+layout(local_size_x = 16, local_size_y = 8, local_size_z = 8) in;
+
+const uint GOB_SIZE_X = 64;
+const uint GOB_SIZE_Y = 8;
+const uint GOB_SIZE_Z = 1;
+const uint GOB_SIZE = GOB_SIZE_X * GOB_SIZE_Y * GOB_SIZE_Z;
+
+const uint GOB_SIZE_X_SHIFT = 6;
+const uint GOB_SIZE_Y_SHIFT = 3;
+const uint GOB_SIZE_Z_SHIFT = 0;
+const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHIFT;
+
+const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1, GOB_SIZE_Y - 1);
+
+uint SwizzleOffset(uvec2 pos) {
+    pos = pos & SWIZZLE_MASK;
+    return swizzle_table[pos.y * 64 + pos.x];
+}
+
+uvec4 ReadTexel(uint offset) {
+    switch (bytes_per_block_log2) {
+#if HAS_EXTENDED_TYPES
+    case 0:
+        return uvec4(u8data[offset], 0, 0, 0);
+    case 1:
+        return uvec4(u16data[offset / 2], 0, 0, 0);
+#else
+    case 0:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 24), 8), 0, 0, 0);
+    case 1:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 16), 16), 0, 0, 0);
+#endif
+    case 2:
+        return uvec4(u32data[offset / 4], 0, 0, 0);
+    case 3:
+        return uvec4(u64data[offset / 8], 0, 0);
+    case 4:
+        return u128data[offset / 16];
+    }
+    return uvec4(0);
+}
+
+void main() {
+    uvec3 pos = gl_GlobalInvocationID + origin;
+    pos.x <<= bytes_per_block_log2;
+
+    // Read as soon as possible due to its latency
+    const uint swizzle = SwizzleOffset(pos.xy);
+
+    const uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
+
+    uint offset = 0;
+    offset += (pos.z >> block_depth) * slice_size;
+    offset += (pos.z & block_depth_mask) << (GOB_SIZE_SHIFT + block_height);
+    offset += (block_y >> block_height) * block_size;
+    offset += (block_y & block_height_mask) << GOB_SIZE_SHIFT;
+    offset += (pos.x >> GOB_SIZE_X_SHIFT) << x_shift;
+    offset += swizzle;
+
+    const uvec4 texel = ReadTexel(offset);
+    const ivec3 coord = ivec3(gl_GlobalInvocationID) + destination;
+    imageStore(output_image, coord, texel);
+}
diff --git a/src/video_core/host_shaders/convert_depth_to_float.frag b/src/video_core/host_shaders/convert_depth_to_float.frag
new file mode 100644
index 000000000..624c58509
--- /dev/null
+++ b/src/video_core/host_shaders/convert_depth_to_float.frag
@@ -0,0 +1,13 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 450
+
+layout(binding = 0) uniform sampler2D depth_texture;
+layout(location = 0) out float output_color;
+
+void main() {
+    ivec2 coord = ivec2(gl_FragCoord.xy);
+    output_color = texelFetch(depth_texture, coord, 0).r;
+}
diff --git a/src/video_core/host_shaders/convert_float_to_depth.frag b/src/video_core/host_shaders/convert_float_to_depth.frag
new file mode 100644
index 000000000..d86c795f4
--- /dev/null
+++ b/src/video_core/host_shaders/convert_float_to_depth.frag
@@ -0,0 +1,13 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 450
+
+layout(binding = 0) uniform sampler2D color_texture;
+
+void main() {
+    ivec2 coord = ivec2(gl_FragCoord.xy);
+    float color = texelFetch(color_texture, coord, 0).r;
+    gl_FragDepth = color;
+}
diff --git a/src/video_core/host_shaders/full_screen_triangle.vert b/src/video_core/host_shaders/full_screen_triangle.vert
new file mode 100644
index 000000000..452ad6502
--- /dev/null
+++ b/src/video_core/host_shaders/full_screen_triangle.vert
@@ -0,0 +1,29 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 450
+
+#ifdef VULKAN
+#define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
+#define END_PUSH_CONSTANTS };
+#define UNIFORM(n)
+#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
+#define BEGIN_PUSH_CONSTANTS
+#define END_PUSH_CONSTANTS
+#define UNIFORM(n) layout (location = n) uniform
+#endif
+
+BEGIN_PUSH_CONSTANTS
+UNIFORM(0) vec2 tex_scale;
+UNIFORM(1) vec2 tex_offset;
+END_PUSH_CONSTANTS
+
+layout(location = 0) out vec2 texcoord;
+
+void main() {
+    float x = float((gl_VertexIndex & 1) << 2);
+    float y = float((gl_VertexIndex & 2) << 1);
+    gl_Position = vec4(x - 1.0, y - 1.0, 0.0, 1.0);
+    texcoord = fma(vec2(x, y) / 2.0, tex_scale, tex_offset);
+}
diff --git a/src/video_core/host_shaders/opengl_copy_bc4.comp b/src/video_core/host_shaders/opengl_copy_bc4.comp
new file mode 100644
index 000000000..7b8e20fbe
--- /dev/null
+++ b/src/video_core/host_shaders/opengl_copy_bc4.comp
@@ -0,0 +1,70 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 430 core
+#extension GL_ARB_gpu_shader_int64 : require
+
+layout (local_size_x = 4, local_size_y = 4) in;
+
+layout(binding = 0, rg32ui) readonly uniform uimage3D bc4_input;
+layout(binding = 1, rgba8ui) writeonly uniform uimage3D bc4_output;
+
+layout(location = 0) uniform uvec3 src_offset;
+layout(location = 1) uniform uvec3 dst_offset;
+
+// https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_texture_compression_rgtc.txt
+uint DecompressBlock(uint64_t bits, uvec2 coord) {
+    const uint code_offset = 16 + 3 * (4 * coord.y + coord.x);
+    const uint code = uint(bits >> code_offset) & 7;
+    const uint red0 = uint(bits >> 0) & 0xff;
+    const uint red1 = uint(bits >> 8) & 0xff;
+    if (red0 > red1) {
+        switch (code) {
+        case 0:
+            return red0;
+        case 1:
+            return red1;
+        case 2:
+            return (6 * red0 + 1 * red1) / 7;
+        case 3:
+            return (5 * red0 + 2 * red1) / 7;
+        case 4:
+            return (4 * red0 + 3 * red1) / 7;
+        case 5:
+            return (3 * red0 + 4 * red1) / 7;
+        case 6:
+            return (2 * red0 + 5 * red1) / 7;
+        case 7:
+            return (1 * red0 + 6 * red1) / 7;
+        }
+    } else {
+        switch (code) {
+        case 0:
+            return red0;
+        case 1:
+            return red1;
+        case 2:
+            return (4 * red0 + 1 * red1) / 5;
+        case 3:
+            return (3 * red0 + 2 * red1) / 5;
+        case 4:
+            return (2 * red0 + 3 * red1) / 5;
+        case 5:
+            return (1 * red0 + 4 * red1) / 5;
+        case 6:
+            return 0;
+        case 7:
+            return 0xff;
+        }
+    }
+    return 0;
+}
+
+void main() {
+    uvec2 packed_bits = imageLoad(bc4_input, ivec3(gl_WorkGroupID + src_offset)).rg;
+    uint64_t bits = packUint2x32(packed_bits);
+    uint red = DecompressBlock(bits, gl_LocalInvocationID.xy);
+    uvec4 color = uvec4(red & 0xff, 0, 0, 0xff);
+    imageStore(bc4_output, ivec3(gl_GlobalInvocationID + dst_offset), color);
+}
diff --git a/src/video_core/host_shaders/opengl_present.frag b/src/video_core/host_shaders/opengl_present.frag
index 8a4cb024b..84b818227 100644
--- a/src/video_core/host_shaders/opengl_present.frag
+++ b/src/video_core/host_shaders/opengl_present.frag
@@ -1,3 +1,7 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
 #version 430 core
 
 layout (location = 0) in vec2 frag_tex_coord;
diff --git a/src/video_core/host_shaders/opengl_present.vert b/src/video_core/host_shaders/opengl_present.vert
index 2235d31a4..c3b5adbba 100644
--- a/src/video_core/host_shaders/opengl_present.vert
+++ b/src/video_core/host_shaders/opengl_present.vert
@@ -1,3 +1,7 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
 #version 430 core
 
 out gl_PerVertex {
diff --git a/src/video_core/host_shaders/pitch_unswizzle.comp b/src/video_core/host_shaders/pitch_unswizzle.comp
new file mode 100644
index 000000000..cb48ec170
--- /dev/null
+++ b/src/video_core/host_shaders/pitch_unswizzle.comp
@@ -0,0 +1,86 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 430
+
+#ifdef VULKAN
+
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+#define HAS_EXTENDED_TYPES 1
+#define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
+#define END_PUSH_CONSTANTS };
+#define UNIFORM(n)
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_IMAGE 1
+
+#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
+
+#extension GL_NV_gpu_shader5 : enable
+#ifdef GL_NV_gpu_shader5
+#define HAS_EXTENDED_TYPES 1
+#else
+#define HAS_EXTENDED_TYPES 0
+#endif
+#define BEGIN_PUSH_CONSTANTS
+#define END_PUSH_CONSTANTS
+#define UNIFORM(n) layout (location = n) uniform
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_IMAGE 0
+
+#endif
+
+BEGIN_PUSH_CONSTANTS
+UNIFORM(0) uvec2 origin;
+UNIFORM(1) ivec2 destination;
+UNIFORM(2) uint bytes_per_block;
+UNIFORM(3) uint pitch;
+END_PUSH_CONSTANTS
+
+#if HAS_EXTENDED_TYPES
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU8 { uint8_t u8data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU16 { uint16_t u16data[]; };
+#endif
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU32 { uint u32data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU64 { uvec2 u64data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU128 { uvec4 u128data[]; };
+
+layout(binding = BINDING_OUTPUT_IMAGE) writeonly uniform uimage2D output_image;
+
+layout(local_size_x = 32, local_size_y = 32, local_size_z = 1) in;
+
+uvec4 ReadTexel(uint offset) {
+    switch (bytes_per_block) {
+#if HAS_EXTENDED_TYPES
+    case 1:
+        return uvec4(u8data[offset], 0, 0, 0);
+    case 2:
+        return uvec4(u16data[offset / 2], 0, 0, 0);
+#else
+    case 1:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 24), 8), 0, 0, 0);
+    case 2:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 16), 16), 0, 0, 0);
+#endif
+    case 4:
+        return uvec4(u32data[offset / 4], 0, 0, 0);
+    case 8:
+        return uvec4(u64data[offset / 8], 0, 0);
+    case 16:
+        return u128data[offset / 16];
+    }
+    return uvec4(0);
+}
+
+void main() {
+    uvec2 pos = gl_GlobalInvocationID.xy + origin;
+
+    uint offset = 0;
+    offset += pos.x * bytes_per_block;
+    offset += pos.y * pitch;
+
+    const uvec4 texel = ReadTexel(offset);
+    const ivec2 coord = ivec2(gl_GlobalInvocationID.xy) + destination;
+    imageStore(output_image, coord, texel);
+}
diff --git a/src/video_core/host_shaders/vulkan_blit_color_float.frag b/src/video_core/host_shaders/vulkan_blit_color_float.frag
new file mode 100644
index 000000000..4a6aae410
--- /dev/null
+++ b/src/video_core/host_shaders/vulkan_blit_color_float.frag
@@ -0,0 +1,14 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#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);
+}
diff --git a/src/video_core/host_shaders/vulkan_blit_depth_stencil.frag b/src/video_core/host_shaders/vulkan_blit_depth_stencil.frag
new file mode 100644
index 000000000..19bb23a5a
--- /dev/null
+++ b/src/video_core/host_shaders/vulkan_blit_depth_stencil.frag
@@ -0,0 +1,16 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 450
+#extension GL_ARB_shader_stencil_export : require
+
+layout(binding = 0) uniform sampler2D depth_tex;
+layout(binding = 1) uniform isampler2D stencil_tex;
+
+layout(location = 0) in vec2 texcoord;
+
+void main() {
+    gl_FragDepth = textureLod(depth_tex, texcoord, 0).r;
+    gl_FragStencilRefARB = textureLod(stencil_tex, texcoord, 0).r;
+}
diff --git a/src/video_core/host_shaders/vulkan_present.frag b/src/video_core/host_shaders/vulkan_present.frag
new file mode 100644
index 000000000..0979ff3e6
--- /dev/null
+++ b/src/video_core/host_shaders/vulkan_present.frag
@@ -0,0 +1,15 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 460 core
+
+layout (location = 0) in vec2 frag_tex_coord;
+
+layout (location = 0) out vec4 color;
+
+layout (binding = 1) uniform sampler2D color_texture;
+
+void main() {
+    color = texture(color_texture, frag_tex_coord);
+}
diff --git a/src/video_core/host_shaders/vulkan_present.vert b/src/video_core/host_shaders/vulkan_present.vert
new file mode 100644
index 000000000..00b868958
--- /dev/null
+++ b/src/video_core/host_shaders/vulkan_present.vert
@@ -0,0 +1,19 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 460 core
+
+layout (location = 0) in vec2 vert_position;
+layout (location = 1) in vec2 vert_tex_coord;
+
+layout (location = 0) out vec2 frag_tex_coord;
+
+layout (set = 0, binding = 0) uniform MatrixBlock {
+    mat4 modelview_matrix;
+};
+
+void main() {
+    gl_Position = modelview_matrix * vec4(vert_position, 0.0, 1.0);
+    frag_tex_coord = vert_tex_coord;
+}
diff --git a/src/video_core/host_shaders/vulkan_quad_array.comp b/src/video_core/host_shaders/vulkan_quad_array.comp
new file mode 100644
index 000000000..212f4e998
--- /dev/null
+++ b/src/video_core/host_shaders/vulkan_quad_array.comp
@@ -0,0 +1,28 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 460 core
+
+layout (local_size_x = 1024) in;
+
+layout (std430, set = 0, binding = 0) buffer OutputBuffer {
+    uint output_indexes[];
+};
+
+layout (push_constant) uniform PushConstants {
+    uint first;
+};
+
+void main() {
+    uint primitive = gl_GlobalInvocationID.x;
+    if (primitive * 6 >= output_indexes.length()) {
+        return;
+    }
+
+    const uint quad_map[6] = uint[](0, 1, 2, 0, 2, 3);
+    for (uint vertex = 0; vertex < 6; ++vertex) {
+        uint index = first + primitive * 4 + quad_map[vertex];
+        output_indexes[primitive * 6 + vertex] = index;
+    }
+}
diff --git a/src/video_core/host_shaders/vulkan_quad_indexed.comp b/src/video_core/host_shaders/vulkan_quad_indexed.comp
new file mode 100644
index 000000000..8655591d0
--- /dev/null
+++ b/src/video_core/host_shaders/vulkan_quad_indexed.comp
@@ -0,0 +1,41 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 460 core
+
+layout (local_size_x = 1024) in;
+
+layout (std430, set = 0, binding = 0) readonly buffer InputBuffer {
+    uint input_indexes[];
+};
+
+layout (std430, set = 0, binding = 1) writeonly buffer OutputBuffer {
+    uint output_indexes[];
+};
+
+layout (push_constant) uniform PushConstants {
+    uint base_vertex;
+    int index_shift; // 0: uint8, 1: uint16, 2: uint32
+};
+
+void main() {
+    int primitive = int(gl_GlobalInvocationID.x);
+    if (primitive * 6 >= output_indexes.length()) {
+        return;
+    }
+
+    int index_size = 8 << index_shift;
+    int flipped_shift = 2 - index_shift;
+    int mask = (1 << flipped_shift) - 1;
+
+    const int quad_swizzle[6] = int[](0, 1, 2, 0, 2, 3);
+    for (uint vertex = 0; vertex < 6; ++vertex) {
+        int offset = primitive * 4 + quad_swizzle[vertex];
+        int int_offset = offset >> flipped_shift;
+        int bit_offset = (offset & mask) * index_size;
+        uint packed_input = input_indexes[int_offset];
+        uint index = bitfieldExtract(packed_input, bit_offset, index_size);
+        output_indexes[primitive * 6 + vertex] = index + base_vertex;
+    }
+}
diff --git a/src/video_core/host_shaders/vulkan_uint8.comp b/src/video_core/host_shaders/vulkan_uint8.comp
new file mode 100644
index 000000000..ad74d7af9
--- /dev/null
+++ b/src/video_core/host_shaders/vulkan_uint8.comp
@@ -0,0 +1,24 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 460 core
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+
+layout (local_size_x = 1024) in;
+
+layout (std430, set = 0, binding = 0) readonly buffer InputBuffer {
+    uint8_t input_indexes[];
+};
+
+layout (std430, set = 0, binding = 1) writeonly buffer OutputBuffer {
+    uint16_t output_indexes[];
+};
+
+void main() {
+    uint id = gl_GlobalInvocationID.x;
+    if (id < input_indexes.length()) {
+        output_indexes[id] = uint16_t(input_indexes[id]);
+    }
+}