vulkan: repack 69-byte vertices without dynamic vertex input on MoltenVK

e38b738b84 repacks the 69-byte vertex stride to 72 only with dynamic
vertex input, unless EGLANTINE_EXPERIMENTAL_STATIC_VERTEX_REPACK=1.
MoltenVK has no VK_EXT_vertex_input_dynamic_state and Metal needs strides
that are a multiple of 4, so the stride was never fixed on Macs.

Enable the static path automatically when the device's minimum vertex
binding stride alignment does not divide 69. The device reports 4 for
MoltenVK, matching its portability subset limit. The environment variable
still overrides it: 1 forces the repack on, 0 forces it off.

Assisted-by: Claude Sonnet 5.5 (implementation and commit message)

8 files changed, 39 insertions, 11 deletions

diff --git a/src/tests/video_core/vertex_repack.cpp b/src/tests/video_core/vertex_repack.cpp
index 3705a97926..5e21e3714c 100644
--- a/src/tests/video_core/vertex_repack.cpp
+++ b/src/tests/video_core/vertex_repack.cpp
@@ -76,10 +76,15 @@ TEST_CASE("Vertex repacking requires the captured tail and stride", "[vertex_rep
     REQUIRE(Vulkan::VertexRepackMask(regs) == 0);
 }
 
-TEST_CASE("Static vertex repacking requires explicit experimental opt-in", "[vertex_repack]") {
+TEST_CASE("Static vertex repacking follows the device stride alignment", "[vertex_repack]") {
     REQUIRE(Vulkan::VertexRepackInputSupported(true, false));
     REQUIRE_FALSE(Vulkan::VertexRepackInputSupported(false, false));
     REQUIRE(Vulkan::VertexRepackInputSupported(false, true));
+    REQUIRE_FALSE(Vulkan::StaticVertexRepackSelected(nullptr, 1));
+    REQUIRE(Vulkan::StaticVertexRepackSelected(nullptr, 4));
+    REQUIRE(Vulkan::StaticVertexRepackSelected("1", 1));
+    REQUIRE_FALSE(Vulkan::StaticVertexRepackSelected("0", 4));
+    REQUIRE(Vulkan::StaticVertexRepackSelected("other", 4));
 }
 
 TEST_CASE("Vertex repack planning keeps descriptor offsets aligned", "[vertex_repack]") {
diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
index d1f78c0034..76a564aae9 100644
--- a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
+++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
@@ -269,7 +269,7 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d, DynamicFe
     if (xfb_enabled) {
         RefreshXfbState(xfb_state, regs);
     }
-    if (!dynamic_vertex_input && ExperimentalStaticVertexRepackEnabled()) {
+    if (!dynamic_vertex_input && features.has_static_vertex_repack) {
         const u32 repack_mask = VertexRepackMask(regs);
         // Refresh from guest values even when only the stride changed: offsets must not accumulate.
         for (size_t index = 0; index < attributes.size(); ++index) {
diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.h b/src/video_core/renderer_vulkan/fixed_pipeline_state.h
index 36a4b4b7a2..1fe5cd75f3 100644
--- a/src/video_core/renderer_vulkan/fixed_pipeline_state.h
+++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.h
@@ -33,6 +33,7 @@ struct DynamicFeatures {
     bool has_dynamic_state3_logic_op_enable;
     bool has_dynamic_state3_line_stipple_enable;
     bool has_dynamic_vertex_input;
+    bool has_static_vertex_repack;
     bool has_color_write_enable;
     bool has_provoking_vertex;
     bool has_provoking_vertex_first_mode;
diff --git a/src/video_core/renderer_vulkan/vertex_repack.h b/src/video_core/renderer_vulkan/vertex_repack.h
index 456d1c6f20..493eb41bd1 100644
--- a/src/video_core/renderer_vulkan/vertex_repack.h
+++ b/src/video_core/renderer_vulkan/vertex_repack.h
@@ -15,12 +15,21 @@ constexpr bool VertexRepackInputSupported(bool dynamic_input, bool experimental_
     return dynamic_input || experimental_static;
 }
 
-inline bool ExperimentalStaticVertexRepackEnabled() {
-    static const bool enabled = [] {
-        const char* value = std::getenv("EGLANTINE_EXPERIMENTAL_STATIC_VERTEX_REPACK");
-        return value && std::strcmp(value, "1") == 0;
-    }();
-    return enabled;
+// Pipelines without dynamic vertex input repack when the device cannot fetch the 69-byte stride.
+// EGLANTINE_EXPERIMENTAL_STATIC_VERTEX_REPACK=1 forces the repack on, =0 forces it off.
+inline bool StaticVertexRepackSelected(const char* override_value, u32 stride_alignment) {
+    if (override_value && std::strcmp(override_value, "1") == 0) {
+        return true;
+    }
+    if (override_value && std::strcmp(override_value, "0") == 0) {
+        return false;
+    }
+    return stride_alignment > 1 && 69 % stride_alignment != 0;
+}
+
+inline bool StaticVertexRepackEnabled(u32 stride_alignment) {
+    return StaticVertexRepackSelected(std::getenv("EGLANTINE_EXPERIMENTAL_STATIC_VERTEX_REPACK"),
+                                      stride_alignment);
 }
 
 struct VertexRepackParameters {
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
index 4d5ccb6520..43cb7d9414 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
@@ -365,6 +365,10 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& m
     LOG_INFO(Render_Vulkan, "69-byte vertex layout repacking {} ({})",
              enable_vertex_repack ? "enabled" : "disabled",
              "portable Vulkan layout fallback");
+    static_vertex_repack =
+        StaticVertexRepackEnabled(device.GetMinVertexInputBindingStrideAlignment());
+    LOG_INFO(Render_Vulkan, "Vertex repacking without dynamic vertex input {}",
+             static_vertex_repack ? "enabled" : "disabled");
     if (enable_vertex_repack) {
         vertex_repack_pass = std::make_unique<VertexRepackPass>(
             device, scheduler, descriptor_pool, staging_pool, compute_pass_descriptor_queue);
@@ -387,8 +391,7 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& m
 u32 BufferCacheRuntime::SetVertexRepackLayout(const Tegra::Engines::Maxwell3D::Regs& regs,
                                             bool dynamic_input) {
     const u32 old_mask = vertex_repack_mask;
-    const bool input_supported = VertexRepackInputSupported(
-        dynamic_input, ExperimentalStaticVertexRepackEnabled());
+    const bool input_supported = VertexRepackInputSupported(dynamic_input, static_vertex_repack);
     vertex_repack_mask = vertex_repack_pass && input_supported ? VertexRepackMask(regs) : 0;
     if (vertex_repack_mask != old_mask) {
         LOG_DEBUG(Render_Vulkan, "Vertex repack selection: mask={:#x}, dynamic_input={}",
@@ -714,7 +717,7 @@ void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bi
             bindings.strides[i] = 72;
             static bool reported = false;
             if (!reported) {
-                LOG_WARNING(Render_Vulkan, "Experimental GPU vertex repacking active: 69 -> 72 bytes");
+                LOG_INFO(Render_Vulkan, "GPU vertex repacking active: 69 -> 72 bytes");
                 reported = true;
             }
         }
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h
index a8d0274016..f65b6daf2c 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.h
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h
@@ -271,6 +271,7 @@ private:
     std::unique_ptr<Uint8Pass> uint8_pass;
     std::unique_ptr<VertexRepackPass> vertex_repack_pass;
     u32 vertex_repack_mask{};
+    bool static_vertex_repack{};
     QuadIndexedPass quad_index_pass;
 
     MultiRangeBufferCache multi_range_buffers;
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index 4df940bf71..a3cd93484e 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -34,6 +34,7 @@
 #include "video_core/renderer_vulkan/maxwell_to_vk.h"
 #include "video_core/renderer_vulkan/pipeline_helper.h"
 #include "video_core/renderer_vulkan/pipeline_statistics.h"
+#include "video_core/renderer_vulkan/vertex_repack.h"
 #include "video_core/renderer_vulkan/vk_compute_pipeline.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
 #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
@@ -583,6 +584,8 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
     dynamic_features.has_dynamic_vertex_input =
         device.IsExtVertexInputDynamicStateSupported() &&
         Settings::values.vertex_input_dynamic_state.GetValue();
+    dynamic_features.has_static_vertex_repack =
+        StaticVertexRepackEnabled(device.GetMinVertexInputBindingStrideAlignment());
 
     dynamic_features.has_provoking_vertex = device.IsExtProvokingVertexSupported();
     dynamic_features.has_provoking_vertex_first_mode =
diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h
index 622d792210..5f75bcc18e 100644
--- a/src/video_core/vulkan_common/vulkan_device.h
+++ b/src/video_core/vulkan_common/vulkan_device.h
@@ -1161,6 +1161,12 @@ FN_MAX_LIMIT_LIST
         return properties.driver.driverID == VK_DRIVER_ID_MOLTENVK;
     }
 
+    /// The multiple vertex input binding strides must be. Metal needs 4 bytes, which MoltenVK
+    /// reports as minVertexInputBindingStrideAlignment through VK_KHR_portability_subset.
+    u32 GetMinVertexInputBindingStrideAlignment() const noexcept {
+        return IsMoltenVK() ? 4 : 1;
+    }
+
     NvidiaArchitecture GetNvidiaArch() const noexcept {
         return nvidia_arch;
     }