From c96177090077a3cf4f73f83920bb067ef873bfca Mon Sep 17 00:00:00 2001
From: ReinUsesLisp <reinuseslisp@airmail.cc>
Date: Tue, 14 Apr 2020 17:54:45 -0300
Subject: [PATCH] vk_compute_pass: Implement indexed quads
Implement indexed quads (GL_QUADS used with glDrawElements*) with a
compute pass conversion.
The compute shader converts from uint8/uint16/uint32 indices to uint32.
The format is passed through push constants to avoid having different
variants of the same shader.
- Used by Fast RMX
- Used by Xenoblade Chronicles 2 (it still has graphical due to
synchronization issues on Vulkan)
---
.../renderer_vulkan/shaders/quad_indexed.comp | 50 +++++
.../renderer_vulkan/vk_compute_pass.cpp | 205 +++++++++++++++++-
.../renderer_vulkan/vk_compute_pass.h | 19 ++
.../renderer_vulkan/vk_rasterizer.cpp | 17 +-
.../renderer_vulkan/vk_rasterizer.h | 1 +
5 files changed, 280 insertions(+), 12 deletions(-)
create mode 100644 src/video_core/renderer_vulkan/shaders/quad_indexed.comp
diff --git a/src/video_core/renderer_vulkan/shaders/quad_indexed.comp b/src/video_core/renderer_vulkan/shaders/quad_indexed.comp
new file mode 100644
index 0000000000..5a472ba9b8
--- /dev/null
+++ b/src/video_core/renderer_vulkan/shaders/quad_indexed.comp
@@ -0,0 +1,50 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+/*
+ * Build instructions:
+ * $ glslangValidator -V quad_indexed.comp -o output.spv
+ * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
+ * $ xxd -i optimized.spv
+ *
+ * Then copy that bytecode to the C++ file
+ */
+
+#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/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
index 878a78755f..7b0268033f 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
@@ -135,11 +135,11 @@ VkDescriptorUpdateTemplateEntryKHR BuildQuadArrayPassDescriptorUpdateTemplateEnt
return entry;
}
-VkPushConstantRange BuildQuadArrayPassPushConstantRange() {
+VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
VkPushConstantRange range;
range.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT;
range.offset = 0;
- range.size = sizeof(u32);
+ range.size = static_cast<u32>(size);
return range;
}
@@ -220,7 +220,130 @@ constexpr u8 uint8_pass[] = {
0xf9, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00,
0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00};
-std::array<VkDescriptorSetLayoutBinding, 2> BuildUint8PassDescriptorSetBindings() {
+// Quad indexed SPIR-V module. Generated from the "shaders/" directory.
+constexpr u8 QUAD_INDEXED_SPV[] = {
+ 0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x7c, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
+ 0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
+ 0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+ 0x0f, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
+ 0x00, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00,
+ 0x11, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+ 0x47, 0x00, 0x04, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
+ 0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
+ 0x48, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
+ 0x48, 0x00, 0x05, 0x00, 0x16, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x16, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+ 0x47, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x47, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+ 0x48, 0x00, 0x05, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x22, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+ 0x23, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x22, 0x00, 0x00, 0x00,
+ 0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x56, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x04, 0x00, 0x00, 0x00, 0x48, 0x00, 0x04, 0x00, 0x57, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x18, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x57, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x57, 0x00, 0x00, 0x00,
+ 0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x59, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x59, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x72, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
+ 0x19, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00,
+ 0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00,
+ 0x07, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
+ 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
+ 0x01, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
+ 0x0c, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
+ 0x0d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0e, 0x00, 0x00, 0x00,
+ 0x01, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x13, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x15, 0x00, 0x00, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x16, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
+ 0x20, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00,
+ 0x3b, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+ 0x14, 0x00, 0x02, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x21, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
+ 0x24, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x25, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x26, 0x00, 0x00, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x2b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
+ 0x3b, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x3f, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x04, 0x00, 0x41, 0x00, 0x00, 0x00,
+ 0x06, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x42, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x43, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x09, 0x00, 0x41, 0x00, 0x00, 0x00,
+ 0x44, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00,
+ 0x42, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x43, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
+ 0x46, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x41, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
+ 0x56, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x57, 0x00, 0x00, 0x00,
+ 0x56, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x58, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+ 0x57, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x58, 0x00, 0x00, 0x00, 0x59, 0x00, 0x00, 0x00,
+ 0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x5b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x69, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x70, 0x00, 0x00, 0x00,
+ 0x00, 0x04, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00,
+ 0x01, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x72, 0x00, 0x00, 0x00,
+ 0x70, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00,
+ 0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+ 0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x46, 0x00, 0x00, 0x00,
+ 0x47, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00,
+ 0xf8, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00, 0xf6, 0x00, 0x04, 0x00, 0x73, 0x00, 0x00, 0x00,
+ 0x76, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x75, 0x00, 0x00, 0x00,
+ 0xf8, 0x00, 0x02, 0x00, 0x75, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0e, 0x00, 0x00, 0x00,
+ 0x0f, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
+ 0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00,
+ 0x06, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00,
+ 0x44, 0x00, 0x05, 0x00, 0x09, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
+ 0x19, 0x00, 0x00, 0x00, 0xaf, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
+ 0x14, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0xf7, 0x00, 0x03, 0x00, 0x1e, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
+ 0x1e, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00,
+ 0x73, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00,
+ 0x26, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
+ 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
+ 0xc4, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+ 0x28, 0x00, 0x00, 0x00, 0x82, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00,
+ 0x2b, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0xc4, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x31, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00, 0x82, 0x00, 0x05, 0x00,
+ 0x06, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
+ 0xf9, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00,
+ 0xf5, 0x00, 0x07, 0x00, 0x09, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00,
+ 0x1e, 0x00, 0x00, 0x00, 0x6f, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x05, 0x00,
+ 0x1b, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00,
+ 0xf6, 0x00, 0x04, 0x00, 0x37, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0xfa, 0x00, 0x04, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x37, 0x00, 0x00, 0x00,
+ 0xf8, 0x00, 0x02, 0x00, 0x36, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x40, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x3f, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00,
+ 0x47, 0x00, 0x00, 0x00, 0x44, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x07, 0x00, 0x00, 0x00,
+ 0x48, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
+ 0x06, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
+ 0x06, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00,
+ 0xc3, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x4e, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00,
+ 0x2e, 0x00, 0x00, 0x00, 0xc7, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x52, 0x00, 0x00, 0x00,
+ 0x4a, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
+ 0x54, 0x00, 0x00, 0x00, 0x52, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00,
+ 0x5b, 0x00, 0x00, 0x00, 0x5c, 0x00, 0x00, 0x00, 0x59, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00,
+ 0x4e, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x5d, 0x00, 0x00, 0x00,
+ 0x5c, 0x00, 0x00, 0x00, 0xcb, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00, 0x62, 0x00, 0x00, 0x00,
+ 0x5d, 0x00, 0x00, 0x00, 0x54, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00,
+ 0x41, 0x00, 0x05, 0x00, 0x69, 0x00, 0x00, 0x00, 0x6a, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
+ 0x42, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x6b, 0x00, 0x00, 0x00,
+ 0x6a, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x09, 0x00, 0x00, 0x00, 0x6c, 0x00, 0x00, 0x00,
+ 0x62, 0x00, 0x00, 0x00, 0x6b, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x5b, 0x00, 0x00, 0x00,
+ 0x6d, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00,
+ 0x3e, 0x00, 0x03, 0x00, 0x6d, 0x00, 0x00, 0x00, 0x6c, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
+ 0x09, 0x00, 0x00, 0x00, 0x6f, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
+ 0xf9, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x37, 0x00, 0x00, 0x00,
+ 0xf9, 0x00, 0x02, 0x00, 0x73, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x76, 0x00, 0x00, 0x00,
+ 0xf9, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x73, 0x00, 0x00, 0x00,
+ 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00};
+
+std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBindings() {
std::array<VkDescriptorSetLayoutBinding, 2> bindings;
bindings[0].binding = 0;
bindings[0].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
@@ -235,7 +358,7 @@ std::array<VkDescriptorSetLayoutBinding, 2> BuildUint8PassDescriptorSetBindings(
return bindings;
}
-VkDescriptorUpdateTemplateEntryKHR BuildUint8PassDescriptorUpdateTemplateEntry() {
+VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
VkDescriptorUpdateTemplateEntryKHR entry;
entry.dstBinding = 0;
entry.dstArrayElement = 0;
@@ -337,14 +460,14 @@ QuadArrayPass::QuadArrayPass(const VKDevice& device, VKScheduler& scheduler,
VKUpdateDescriptorQueue& update_descriptor_queue)
: VKComputePass(device, descriptor_pool, BuildQuadArrayPassDescriptorSetLayoutBinding(),
BuildQuadArrayPassDescriptorUpdateTemplateEntry(),
- BuildQuadArrayPassPushConstantRange(), std::size(quad_array), quad_array),
+ BuildComputePushConstantRange(sizeof(u32)), std::size(quad_array), quad_array),
scheduler{scheduler}, staging_buffer_pool{staging_buffer_pool},
update_descriptor_queue{update_descriptor_queue} {}
QuadArrayPass::~QuadArrayPass() = default;
std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) {
- const u32 num_triangle_vertices = num_vertices * 6 / 4;
+ const u32 num_triangle_vertices = (num_vertices / 4) * 6;
const std::size_t staging_size = num_triangle_vertices * sizeof(u32);
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
@@ -383,8 +506,8 @@ std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32
Uint8Pass::Uint8Pass(const VKDevice& device, VKScheduler& scheduler,
VKDescriptorPool& descriptor_pool, VKStagingBufferPool& staging_buffer_pool,
VKUpdateDescriptorQueue& update_descriptor_queue)
- : VKComputePass(device, descriptor_pool, BuildUint8PassDescriptorSetBindings(),
- BuildUint8PassDescriptorUpdateTemplateEntry(), {}, std::size(uint8_pass),
+ : VKComputePass(device, descriptor_pool, BuildInputOutputDescriptorSetBindings(),
+ BuildInputOutputDescriptorUpdateTemplate(), {}, std::size(uint8_pass),
uint8_pass),
scheduler{scheduler}, staging_buffer_pool{staging_buffer_pool},
update_descriptor_queue{update_descriptor_queue} {}
@@ -425,4 +548,70 @@ std::pair<VkBuffer, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buff
return {*buffer.handle, 0};
}
+QuadIndexedPass::QuadIndexedPass(const VKDevice& device, VKScheduler& scheduler,
+ VKDescriptorPool& descriptor_pool,
+ VKStagingBufferPool& staging_buffer_pool,
+ VKUpdateDescriptorQueue& update_descriptor_queue)
+ : VKComputePass(device, descriptor_pool, BuildInputOutputDescriptorSetBindings(),
+ BuildInputOutputDescriptorUpdateTemplate(),
+ BuildComputePushConstantRange(sizeof(u32) * 2), std::size(QUAD_INDEXED_SPV),
+ QUAD_INDEXED_SPV),
+ scheduler{scheduler}, staging_buffer_pool{staging_buffer_pool},
+ update_descriptor_queue{update_descriptor_queue} {}
+
+QuadIndexedPass::~QuadIndexedPass() = default;
+
+std::pair<VkBuffer, u64> QuadIndexedPass::Assemble(
+ Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, u32 num_vertices, u32 base_vertex,
+ VkBuffer src_buffer, u64 src_offset) {
+ const u32 index_shift = [index_format] {
+ switch (index_format) {
+ case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedByte:
+ return 0;
+ case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedShort:
+ return 1;
+ case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedInt:
+ return 2;
+ }
+ UNREACHABLE();
+ return 2;
+ }();
+ const u32 input_size = num_vertices << index_shift;
+ const u32 num_tri_vertices = (num_vertices / 4) * 6;
+
+ const std::size_t staging_size = num_tri_vertices * sizeof(u32);
+ auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
+
+ update_descriptor_queue.Acquire();
+ update_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size);
+ update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size);
+ const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence());
+
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, set,
+ num_tri_vertices, base_vertex, index_shift](vk::CommandBuffer cmdbuf) {
+ static constexpr u32 dispatch_size = 1024;
+ const std::array push_constants = {base_vertex, index_shift};
+ cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
+ cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
+ cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants),
+ &push_constants);
+ cmdbuf.Dispatch(Common::AlignUp(num_tri_vertices, dispatch_size) / dispatch_size, 1, 1);
+
+ VkBufferMemoryBarrier barrier;
+ barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
+ barrier.pNext = nullptr;
+ barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
+ barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
+ barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
+ barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
+ barrier.buffer = buffer;
+ barrier.offset = 0;
+ barrier.size = static_cast<VkDeviceSize>(num_tri_vertices * sizeof(u32));
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+ VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {});
+ });
+ return {*buffer.handle, 0};
+}
+
} // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h
index ec80c86838..26bf834de2 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.h
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.h
@@ -8,6 +8,7 @@
#include <utility>
#include <vector>
#include "common/common_types.h"
+#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/wrapper.h"
@@ -73,4 +74,22 @@ private:
VKUpdateDescriptorQueue& update_descriptor_queue;
};
+class QuadIndexedPass final : public VKComputePass {
+public:
+ explicit QuadIndexedPass(const VKDevice& device, VKScheduler& scheduler,
+ VKDescriptorPool& descriptor_pool,
+ VKStagingBufferPool& staging_buffer_pool,
+ VKUpdateDescriptorQueue& update_descriptor_queue);
+ ~QuadIndexedPass();
+
+ std::pair<VkBuffer, u64> Assemble(Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format,
+ u32 num_vertices, u32 base_vertex, VkBuffer src_buffer,
+ u64 src_offset);
+
+private:
+ VKScheduler& scheduler;
+ VKStagingBufferPool& staging_buffer_pool;
+ VKUpdateDescriptorQueue& update_descriptor_queue;
+};
+
} // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index 4ca0febb86..857bea19f6 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -293,6 +293,7 @@ RasterizerVulkan::RasterizerVulkan(Core::System& system, Core::Frontend::EmuWind
update_descriptor_queue(device, scheduler), renderpass_cache(device),
quad_array_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
+ quad_indexed_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
texture_cache(system, *this, device, resource_manager, memory_manager, scheduler,
staging_pool),
pipeline_cache(system, *this, device, scheduler, descriptor_pool, update_descriptor_queue,
@@ -844,18 +845,26 @@ void RasterizerVulkan::SetupIndexBuffer(BufferBindings& buffer_bindings, DrawPar
bool is_indexed) {
const auto& regs = system.GPU().Maxwell3D().regs;
switch (regs.draw.topology) {
- case Maxwell::PrimitiveTopology::Quads:
- if (params.is_indexed) {
- UNIMPLEMENTED();
- } else {
+ case Maxwell::PrimitiveTopology::Quads: {
+ if (!params.is_indexed) {
const auto [buffer, offset] =
quad_array_pass.Assemble(params.num_vertices, params.base_vertex);
buffer_bindings.SetIndexBinding(buffer, offset, VK_INDEX_TYPE_UINT32);
params.base_vertex = 0;
params.num_vertices = params.num_vertices * 6 / 4;
params.is_indexed = true;
+ break;
}
+ const GPUVAddr gpu_addr = regs.index_array.IndexStart();
+ auto [buffer, offset] = buffer_cache.UploadMemory(gpu_addr, CalculateIndexBufferSize());
+ std::tie(buffer, offset) = quad_indexed_pass.Assemble(
+ regs.index_array.format, params.num_vertices, params.base_vertex, buffer, offset);
+
+ buffer_bindings.SetIndexBinding(buffer, offset, VK_INDEX_TYPE_UINT32);
+ params.num_vertices = (params.num_vertices / 4) * 6;
+ params.base_vertex = 0;
break;
+ }
default: {
if (!is_indexed) {
break;
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index 46037860a9..d9108f862a 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -254,6 +254,7 @@ private:
VKUpdateDescriptorQueue update_descriptor_queue;
VKRenderPassCache renderpass_cache;
QuadArrayPass quad_array_pass;
+ QuadIndexedPass quad_indexed_pass;
Uint8Pass uint8_pass;
VKTextureCache texture_cache;