metal: fix max_storage_xx max_texture max_sampler and `max_vert…

…ex_xx` limits
gfx-rs · Sep 2, 2022 · 163035f · 163035f
1 parent c519901
commit 163035f
Show file tree

Hide file tree

Showing 5 changed files with 18 additions and 20 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -71,6 +71,7 @@ the same every time it is rendered, we now warn if it is missing.
 
 #### Metal
 - Add the missing `msg_send![view, retain]` call within `from_view` by @jinleili in [#2976](https://github.com/gfx-rs/wgpu/pull/2976)
+- Fix `max_buffer` `max_texture` and `max_vertex_buffers` limits by @jinleili in [#2978](https://github.com/gfx-rs/wgpu/pull/2978)
 
 #### Vulkan
 - Fix `astc_hdr` formats support by @jinleili in [#2971]](https://github.com/gfx-rs/wgpu/pull/2971)

diff --git a/wgpu-hal/src/metal/adapter.rs b/wgpu-hal/src/metal/adapter.rs
@@ -616,21 +616,17 @@ impl super::PrivateCapabilities {
             format_bgr10a2_all: Self::supports_any(device, BGR10A2_ALL),
             format_bgr10a2_no_write: !Self::supports_any(device, BGR10A2_ALL),
             max_buffers_per_stage: 31,
-            max_textures_per_stage: if os_is_mac {
-                128 // On macOS, minimun value is 128
-            } else if device.supports_feature_set(MTLFeatureSet::iOS_GPUFamily4_v1) {
+            max_vertex_buffers: 31,
+            max_textures_per_stage: if os_is_mac
+                || (family_check && device.supports_family(MTLGPUFamily::Apple6))
+            {
+                128
+            } else if family_check && device.supports_family(MTLGPUFamily::Apple4) {
                 96
             } else {
                 31
             },
-            max_samplers_per_stage: if (family_check
-                && device.supports_family(MTLGPUFamily::Apple6))
-                || (os_is_mac && rw_texture_tier == MTLReadWriteTextureTier::Tier2)
-            {
-                1024
-            } else {
-                16
-            },
+            max_samplers_per_stage: 16,
             buffer_alignment: if os_is_mac { 256 } else { 64 },
             max_buffer_size: if version.at_least((10, 14), (12, 0)) {
                 // maxBufferLength available on macOS 10.14+ and iOS 12.0+
@@ -833,15 +829,15 @@ impl super::PrivateCapabilities {
                     .max_dynamic_uniform_buffers_per_pipeline_layout,
                 max_dynamic_storage_buffers_per_pipeline_layout: base
                     .max_dynamic_storage_buffers_per_pipeline_layout,
-                max_sampled_textures_per_shader_stage: base.max_sampled_textures_per_shader_stage,
+                max_sampled_textures_per_shader_stage: self.max_textures_per_stage,
                 max_samplers_per_shader_stage: self.max_samplers_per_stage,
-                max_storage_buffers_per_shader_stage: base.max_storage_buffers_per_shader_stage,
-                max_storage_textures_per_shader_stage: base.max_storage_textures_per_shader_stage,
-                max_uniform_buffers_per_shader_stage: 12,
+                max_storage_buffers_per_shader_stage: self.max_buffers_per_stage,
+                max_storage_textures_per_shader_stage: self.max_textures_per_stage,
+                max_uniform_buffers_per_shader_stage: self.max_buffers_per_stage,
                 max_uniform_buffer_binding_size: self.max_buffer_size.min(!0u32 as u64) as u32,
                 max_storage_buffer_binding_size: self.max_buffer_size.min(!0u32 as u64) as u32,
-                max_vertex_buffers: base.max_vertex_buffers,
-                max_vertex_attributes: base.max_vertex_attributes,
+                max_vertex_buffers: self.max_vertex_buffers,
+                max_vertex_attributes: 31,
                 max_vertex_buffer_array_stride: base.max_vertex_buffer_array_stride,
                 max_push_constant_size: 0x1000,
                 min_uniform_buffer_offset_alignment: self.buffer_alignment as u32,

diff --git a/wgpu-hal/src/metal/command.rs b/wgpu-hal/src/metal/command.rs
@@ -724,7 +724,7 @@ impl crate::CommandEncoder<super::Api> for super::CommandEncoder {
         index: u32,
         binding: crate::BufferBinding<'a, super::Api>,
     ) {
-        let buffer_index = self.shared.private_caps.max_buffers_per_stage as u64 - 1 - index as u64;
+        let buffer_index = self.shared.private_caps.max_vertex_buffers as u64 - 1 - index as u64;
         let encoder = self.state.render.as_ref().unwrap();
         encoder.set_vertex_buffer(buffer_index, Some(&binding.buffer.raw), binding.offset);
     }

diff --git a/wgpu-hal/src/metal/device.rs b/wgpu-hal/src/metal/device.rs
@@ -890,7 +890,7 @@ impl crate::Device<super::Api> for super::Device {
         };
 
         if desc.layout.total_counters.vs.buffers + (desc.vertex_buffers.len() as u32)
-            > self.shared.private_caps.max_buffers_per_stage
+            > self.shared.private_caps.max_vertex_buffers
         {
             let msg = format!(
                 "pipeline needs too many buffers in the vertex stage: {} vertex and {} layout",
@@ -907,7 +907,7 @@ impl crate::Device<super::Api> for super::Device {
             let vertex_descriptor = mtl::VertexDescriptor::new();
             for (i, vb) in desc.vertex_buffers.iter().enumerate() {
                 let buffer_index =
-                    self.shared.private_caps.max_buffers_per_stage as u64 - 1 - i as u64;
+                    self.shared.private_caps.max_vertex_buffers as u64 - 1 - i as u64;
                 let buffer_desc = vertex_descriptor.layouts().object_at(buffer_index).unwrap();
 
                 buffer_desc.set_stride(vb.array_stride);

diff --git a/wgpu-hal/src/metal/mod.rs b/wgpu-hal/src/metal/mod.rs
@@ -204,6 +204,7 @@ struct PrivateCapabilities {
     format_bgr10a2_all: bool,
     format_bgr10a2_no_write: bool,
     max_buffers_per_stage: ResourceIndex,
+    max_vertex_buffers: ResourceIndex,
     max_textures_per_stage: ResourceIndex,
     max_samplers_per_stage: ResourceIndex,
     buffer_alignment: u64,