summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorReinUsesLisp <reinuseslisp@airmail.cc>2020-12-30 05:39:05 +0100
committerReinUsesLisp <reinuseslisp@airmail.cc>2020-12-30 05:39:35 +0100
commit12d16248dd61d2998ee271e16d6980d6df963bbe (patch)
tree0717a3b516a58411d5860568c65381ae5c7fc4c6
parenthost_shaders: Add copyright headers to OpenGL present shaders (diff)
downloadyuzu-12d16248dd61d2998ee271e16d6980d6df963bbe.tar
yuzu-12d16248dd61d2998ee271e16d6980d6df963bbe.tar.gz
yuzu-12d16248dd61d2998ee271e16d6980d6df963bbe.tar.bz2
yuzu-12d16248dd61d2998ee271e16d6980d6df963bbe.tar.lz
yuzu-12d16248dd61d2998ee271e16d6980d6df963bbe.tar.xz
yuzu-12d16248dd61d2998ee271e16d6980d6df963bbe.tar.zst
yuzu-12d16248dd61d2998ee271e16d6980d6df963bbe.zip
-rw-r--r--src/video_core/host_shaders/CMakeLists.txt2
-rw-r--r--src/video_core/host_shaders/block_linear_unswizzle_2d.comp122
-rw-r--r--src/video_core/host_shaders/block_linear_unswizzle_3d.comp125
3 files changed, 249 insertions, 0 deletions
diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt
index ff20bc93b..7feb6df99 100644
--- a/src/video_core/host_shaders/CMakeLists.txt
+++ b/src/video_core/host_shaders/CMakeLists.txt
@@ -1,4 +1,6 @@
set(SHADER_FILES
+ block_linear_unswizzle_2d.comp
+ block_linear_unswizzle_3d.comp
opengl_present.frag
opengl_present.vert
)
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);
+}