vulkan: repack 69-byte vertices to 72 bytes (#281)

Splatoon Raiders draws with a 69-byte vertex stride and a packed 4-byte
attribute at byte 65. Those fetches are unaligned, which Vulkan leaves
undefined: NVIDIA reads them as meant, the AMD Windows driver does not,
and the outdoor scenes turn into a white haze with horizontal streaks.

When a binding uses that layout, a compute pass copies it to a 72-byte
stride with the tail moved to byte 68, and the vertex input offsets and
stride follow. EGLANTINE_TEST_VERTEX_REPACK=0 turns it off;
EGLANTINE_EXPERIMENTAL_STATIC_VERTEX_REPACK=1 also covers pipelines
without dynamic vertex input.

Based on an outside patch (02-portable-vertex-repacking). Changes from
it: a range over the device limits skips the repack instead of throwing,
and the pipeline cache version stays, so caches are kept.

2 files changed, 112 insertions, 0 deletions

diff --git a/src/tests/CMakeLists.txt b/src/tests/CMakeLists.txt
index dd6f5d97b5..194e90a243 100644
--- a/src/tests/CMakeLists.txt
+++ b/src/tests/CMakeLists.txt
@@ -19,6 +19,7 @@ add_executable(tests
     core/core_timing.cpp
     core/internal_network/network.cpp
     video_core/memory_tracker.cpp
+    video_core/vertex_repack.cpp
     video_core/ampere_returns.cpp
     input_common/calibration_configuration_job.cpp
 )
diff --git a/src/tests/video_core/vertex_repack.cpp b/src/tests/video_core/vertex_repack.cpp
new file mode 100644
index 0000000000..6409be3ca4
--- /dev/null
+++ b/src/tests/video_core/vertex_repack.cpp
@@ -0,0 +1,111 @@
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#include <catch2/catch_test_macros.hpp>
+#include "video_core/renderer_vulkan/vertex_repack.h"
+
+namespace {
+using Regs = Tegra::Engines::Maxwell3D::Regs;
+using Size = Regs::VertexAttribute::Size;
+
+Regs MakeLayout() {
+    Regs regs{};
+    for (auto& attr : regs.vertex_attrib_format) {
+        attr.constant.Assign(1);
+    }
+    regs.vertex_streams[0].stride.Assign(69);
+    auto& position = regs.vertex_attrib_format[0];
+    position.constant.Assign(0);
+    position.size.Assign(Size::Size_R32_G32_B32);
+    auto& packed = regs.vertex_attrib_format[1];
+    packed.constant.Assign(0);
+    packed.offset.Assign(65);
+    packed.size.Assign(Size::Size_A2_B10_G10_R10);
+    return regs;
+}
+}
+
+TEST_CASE("Vertex repacking preserves aligned fields and relocates packed tail", "[vertex_repack]") {
+    const auto regs = MakeLayout();
+    const auto mask = Vulkan::VertexRepackMask(regs);
+    REQUIRE(mask == 1);
+    REQUIRE(Vulkan::RepackedVertexOffset(mask, 0, 0) == 0);
+    REQUIRE(Vulkan::RepackedVertexOffset(mask, 0, 64) == 64);
+    REQUIRE(Vulkan::RepackedVertexOffset(mask, 0, 65) == 68);
+    REQUIRE(Vulkan::RepackedVertexStride(mask, 0, 69) == 72);
+    REQUIRE(Vulkan::RepackedVertexStride(mask, 1, 16) == 16);
+}
+
+TEST_CASE("Vertex repacking excludes overlapping and oversized attributes", "[vertex_repack]") {
+    auto regs = MakeLayout();
+    auto& attr = regs.vertex_attrib_format[2];
+    attr.constant.Assign(0);
+    attr.offset.Assign(64);
+    attr.size.Assign(Size::Size_R32);
+    REQUIRE(Vulkan::VertexRepackMask(regs) == 0);
+    attr.offset.Assign(68);
+    REQUIRE(Vulkan::VertexRepackMask(regs) == 0);
+    attr.constant.Assign(1);
+    REQUIRE(Vulkan::VertexRepackMask(regs) == 1);
+}
+
+TEST_CASE("Vertex repacking ignores constant invalid formats", "[vertex_repack]") {
+    auto regs = MakeLayout();
+    REQUIRE(Vulkan::VertexRepackMask(regs) == 1);
+    regs.vertex_attrib_format[2].constant.Assign(0);
+    REQUIRE(Vulkan::VertexRepackMask(regs) == 0);
+}
+
+TEST_CASE("Vertex repacking selects bindings independently", "[vertex_repack]") {
+    auto regs = MakeLayout();
+    regs.vertex_streams[31].stride.Assign(69);
+    auto& attr = regs.vertex_attrib_format[2];
+    attr = regs.vertex_attrib_format[1];
+    attr.buffer.Assign(31);
+    REQUIRE(Vulkan::VertexRepackMask(regs) == 0x80000001U);
+    regs.vertex_streams[0].stride.Assign(72);
+    REQUIRE(Vulkan::VertexRepackMask(regs) == 0x80000000U);
+    REQUIRE(Vulkan::RepackedVertexOffset(0x80000000U, 0, 65) == 65);
+}
+
+TEST_CASE("Vertex repacking requires the captured tail and stride", "[vertex_repack]") {
+    auto regs = MakeLayout();
+    regs.vertex_streams[0].stride.Assign(68);
+    REQUIRE(Vulkan::VertexRepackMask(regs) == 0);
+    regs.vertex_streams[0].stride.Assign(69);
+    regs.vertex_attrib_format[1].offset.Assign(64);
+    REQUIRE(Vulkan::VertexRepackMask(regs) == 0);
+}
+
+TEST_CASE("Static vertex repacking requires explicit experimental opt-in", "[vertex_repack]") {
+    REQUIRE(Vulkan::VertexRepackInputSupported(true, false));
+    REQUIRE_FALSE(Vulkan::VertexRepackInputSupported(false, false));
+    REQUIRE(Vulkan::VertexRepackInputSupported(false, true));
+}
+
+TEST_CASE("Vertex repack planning keeps descriptor offsets aligned", "[vertex_repack]") {
+    for (const u32 alignment : {4U, 16U, 64U, 256U, 4096U}) {
+        for (const u32 offset : {0U, 1U, 3U, 255U, 4099U}) {
+            const auto plan = Vulkan::MakeVertexRepackParameters(offset, 70, alignment, 1U << 20, 65535);
+            REQUIRE(plan.has_value());
+            REQUIRE(plan->source_base % alignment == 0);
+            REQUIRE(plan->source_base + plan->source_bias == offset);
+            REQUIRE(plan->vertices == 2);
+            REQUIRE(plan->output_size == 144);
+            REQUIRE(plan->source_range >= plan->source_bias + 70);
+            REQUIRE(plan->source_range % 4 == 0);
+            REQUIRE(plan->dispatch_groups == 1);
+        }
+    }
+}
+
+TEST_CASE("Vertex repack planning rejects overflow and device-limit violations", "[vertex_repack]") {
+    REQUIRE_FALSE(Vulkan::MakeVertexRepackParameters(0, 0, 256, 1U << 20, 65535));
+    REQUIRE_FALSE(Vulkan::MakeVertexRepackParameters(0, 69, 0, 1U << 20, 65535));
+    REQUIRE_FALSE(Vulkan::MakeVertexRepackParameters(0, 69, 256, 71, 65535));
+    REQUIRE_FALSE(Vulkan::MakeVertexRepackParameters(255, 69, 256, 323, 65535));
+    REQUIRE_FALSE(Vulkan::MakeVertexRepackParameters(0, 69, 256, 1U << 20, 0));
+    REQUIRE_FALSE(Vulkan::MakeVertexRepackParameters(0, 0xffffffffU, 256, ~u64{0}, 0xffffffffU));
+    const auto one = Vulkan::MakeVertexRepackParameters(0, 69, 256, 72, 1);
+    REQUIRE(one.has_value());
+    REQUIRE(one->source_range == 72);
+}