[naga msl-out] Fix truncation of pulled vertices (#7458)

Fixes #7410

Author: andyleiserson

naga/src/back/msl/writer.rs

@@ -5,6 +5,7 @@ use alloc::{
     vec::Vec,
 };
 use core::{
+    cmp::Ordering,
     fmt::{Display, Error as FmtError, Formatter, Write},
     iter,
 };
@@ -6384,14 +6385,12 @@ template <typename A>
                     // from the buffer elements.
                     for attribute in vbm.attributes {
                         let location = attribute.shader_location;
-                        let am_option = am_resolved.get(&location);
-                        if am_option.is_none() {
+                        let Some(am) = am_resolved.get(&location) else {
                             // This bound attribute isn't used in this entry point, so
                             // don't bother extracting the data. Too bad we emitted the
                             // unpacking function earlier -- it might not get used.
                             continue;
-                        }
-                        let am = am_option.unwrap();
+                        };
                         let attribute_name = &am.name;
                         let attribute_ty_name = &am.ty_name;
 
@@ -6401,36 +6400,62 @@ template <typename A>
                             .expect("Should have generated this unpacking function earlier.");
                         let func_name = &func.name;
 
-                        write!(self.out, "{}{attribute_name} = ", back::Level(2),)?;
-
                         // Check dimensionality of the attribute compared to the unpacking
-                        // function. If attribute dimension is < unpack dimension, then
-                        // we need to explicitly cast down the result. Otherwise, if attribute
-                        // dimension > unpack dimension, we have to pad out the unpack value
-                        // from a vec4(0, 0, 0, 1) of matching scalar type.
+                        // function. If attribute dimension > unpack dimension, we have to
+                        // pad out the unpack value from a vec4(0, 0, 0, 1) of matching
+                        // scalar type. Otherwise, if attribute dimension is < unpack
+                        // dimension, then we need to explicitly truncate the result.
+
+                        let needs_padding_or_truncation = am.dimension.cmp(&func.dimension);
+
+                        if needs_padding_or_truncation != Ordering::Equal {
+                            // Emit a comment flagging that a conversion is happening,
+                            // since the actual logic can be at the end of a long line.
+                            writeln!(
+                                self.out,
+                                "{}// {attribute_ty_name} <- {:?}",
+                                back::Level(2),
+                                attribute.format
+                            )?;
+                        }
+
+                        write!(self.out, "{}{attribute_name} = ", back::Level(2),)?;
 
-                        let needs_truncate_or_padding = am.dimension != func.dimension;
-                        if needs_truncate_or_padding {
+                        if needs_padding_or_truncation == Ordering::Greater {
+                            // Needs padding: emit constructor call for wider type
                             write!(self.out, "{attribute_ty_name}(")?;
                         }
 
+                        // Emit call to unpacking function
                         write!(self.out, "{func_name}({elem_name}.data[{offset}]",)?;
                         for i in (offset + 1)..(offset + func.byte_count) {
                             write!(self.out, ", {elem_name}.data[{i}]")?;
                         }
                         write!(self.out, ")")?;
 
-                        if needs_truncate_or_padding {
-                            let zero_value = if am.ty_is_int { "0" } else { "0.0" };
-                            let one_value = if am.ty_is_int { "1" } else { "1.0" };
-                            for i in func.dimension..am.dimension {
+                        match needs_padding_or_truncation {
+                            Ordering::Greater => {
+                                // Padding
+                                let zero_value = if am.ty_is_int { "0" } else { "0.0" };
+                                let one_value = if am.ty_is_int { "1" } else { "1.0" };
+                                for i in func.dimension..am.dimension {
+                                    write!(
+                                        self.out,
+                                        ", {}",
+                                        if i == 3 { one_value } else { zero_value }
+                                    )?;
+                                }
+                                write!(self.out, ")")?;
+                            }
+                            Ordering::Less => {
+                                // Truncate to the first `am.dimension` components
                                 write!(
                                     self.out,
-                                    ", {}",
-                                    if i == 3 { one_value } else { zero_value }
+                                    ".{}",
+                                    &"xyzw"[0..usize::try_from(am.dimension).unwrap()]
                                 )?;
                             }
-                            write!(self.out, ")")?;
+                            Ordering::Equal => {}
                         }
 
                         writeln!(self.out, ";")?;

naga/tests/in/wgsl/msl-vpt-formats-x1.toml

@@ -0,0 +1,53 @@
+targets = "METAL"
+
+[msl_pipeline]
+allow_and_force_point_size = false
+vertex_pulling_transform = true
+
+[[msl_pipeline.vertex_buffer_mappings]]
+attributes = [
+    { offset = 0, shader_location = 0, format = "Uint8" },
+    { offset = 16, shader_location = 1, format = "Uint8x2" },
+    { offset = 32, shader_location = 2, format = "Uint8x4" },
+    { offset = 48, shader_location = 3, format = "Sint8" },
+    { offset = 64, shader_location = 4, format = "Sint8x2" },
+    { offset = 80, shader_location = 5, format = "Sint8x4" },
+    { offset = 96, shader_location = 6, format = "Unorm8" },
+    { offset = 112, shader_location = 7, format = "Unorm8x2" },
+    { offset = 128, shader_location = 8, format = "Unorm8x4" },
+    { offset = 144, shader_location = 9, format = "Snorm8" },
+    { offset = 160, shader_location = 10, format = "Snorm8x2" },
+    { offset = 176, shader_location = 11, format = "Snorm8x4" },
+    { offset = 192, shader_location = 12, format = "Uint16" },
+    { offset = 208, shader_location = 13, format = "Uint16x2" },
+    { offset = 224, shader_location = 14, format = "Uint16x4" },
+    { offset = 240, shader_location = 15, format = "Sint16" },
+    { offset = 256, shader_location = 16, format = "Sint16x2" },
+    { offset = 272, shader_location = 17, format = "Sint16x4" },
+    { offset = 288, shader_location = 18, format = "Unorm16" },
+    { offset = 304, shader_location = 19, format = "Unorm16x2" },
+    { offset = 320, shader_location = 20, format = "Unorm16x4" },
+    { offset = 336, shader_location = 21, format = "Snorm16" },
+    { offset = 352, shader_location = 22, format = "Snorm16x2" },
+    { offset = 368, shader_location = 23, format = "Snorm16x4" },
+    { offset = 384, shader_location = 24, format = "Float16" },
+    { offset = 400, shader_location = 25, format = "Float16x2" },
+    { offset = 416, shader_location = 26, format = "Float16x4" },
+    { offset = 432, shader_location = 27, format = "Float32" },
+    { offset = 448, shader_location = 28, format = "Float32x2" },
+    { offset = 464, shader_location = 29, format = "Float32x3" },
+    { offset = 480, shader_location = 30, format = "Float32x4" },
+    { offset = 496, shader_location = 31, format = "Uint32" },
+    { offset = 512, shader_location = 32, format = "Uint32x2" },
+    { offset = 528, shader_location = 33, format = "Uint32x3" },
+    { offset = 544, shader_location = 34, format = "Uint32x4" },
+    { offset = 560, shader_location = 35, format = "Sint32" },
+    { offset = 576, shader_location = 36, format = "Sint32x2" },
+    { offset = 592, shader_location = 37, format = "Sint32x3" },
+    { offset = 608, shader_location = 38, format = "Sint32x4" },
+    { offset = 624, shader_location = 39, format = "unorm10-10-10-2" },
+    { offset = 640, shader_location = 40, format = "unorm8x4-bgra" },
+]
+id = 1
+indexed_by_vertex = true
+stride = 644

naga/tests/in/wgsl/msl-vpt-formats-x1.wgsl

@@ -0,0 +1,62 @@
+// Tests for the vertex pulling transform in the MSL backend.
+//
+// Test loading a scalar from each vertex format.
+//
+// (Note: only the wgsl files are different for each of the -formats-xN tests.
+// The toml files are all the same.)
+
+struct VertexOutput {
+  @builtin(position) position: vec4<f32>,
+}
+
+struct VertexInput {
+  @location( 0) v_uint8     : u32,
+  @location( 1) v_uint8x2   : u32,
+  @location( 2) v_uint8x4   : u32,
+  @location( 3) v_sint8     : i32,
+  @location( 4) v_sint8x2   : i32,
+  @location( 5) v_sint8x4   : i32,
+  @location( 6) v_unorm8    : f32,
+  @location( 7) v_unorm8x2  : f32,
+  @location( 8) v_unorm8x4  : f32,
+  @location( 9) v_snorm8    : f32,
+  @location(10) v_snorm8x2  : f32,
+  @location(11) v_snorm8x4  : f32,
+  @location(12) v_uint16    : u32,
+  @location(13) v_uint16x2  : u32,
+  @location(14) v_uint16x4  : u32,
+  @location(15) v_sint16    : i32,
+  @location(16) v_sint16x2  : i32,
+  @location(17) v_sint16x4  : i32,
+  @location(18) v_unorm16   : f32,
+  @location(19) v_unorm16x2 : f32,
+  @location(20) v_unorm16x4 : f32,
+  @location(21) v_snorm16   : f32,
+  @location(22) v_snorm16x2 : f32,
+  @location(23) v_snorm16x4 : f32,
+  @location(24) v_float16   : f32,
+  @location(25) v_float16x2 : f32,
+  @location(26) v_float16x4 : f32,
+  @location(27) v_float32   : f32,
+  @location(28) v_float32x2 : f32,
+  @location(29) v_float32x3 : f32,
+  @location(30) v_float32x4 : f32,
+  @location(31) v_uint32    : u32,
+  @location(32) v_uint32x2  : u32,
+  @location(33) v_uint32x3  : u32,
+  @location(34) v_uint32x4  : u32,
+  @location(35) v_sint32    : i32,
+  @location(36) v_sint32x2  : i32,
+  @location(37) v_sint32x3  : i32,
+  @location(38) v_sint32x4  : i32,
+  @location(39) v_unorm10_10_10_2: f32,
+  @location(40) v_unorm8x4_bgra: f32,
+}
+
+@vertex
+fn render_vertex(
+  v_in: VertexInput,
+) -> VertexOutput
+{
+  return VertexOutput(vec4f(v_in.v_uint8));
+}

naga/tests/in/wgsl/msl-vpt-formats-x2.toml

@@ -0,0 +1,53 @@
+targets = "METAL"
+
+[msl_pipeline]
+allow_and_force_point_size = false
+vertex_pulling_transform = true
+
+[[msl_pipeline.vertex_buffer_mappings]]
+attributes = [
+    { offset = 0, shader_location = 0, format = "Uint8" },
+    { offset = 16, shader_location = 1, format = "Uint8x2" },
+    { offset = 32, shader_location = 2, format = "Uint8x4" },
+    { offset = 48, shader_location = 3, format = "Sint8" },
+    { offset = 64, shader_location = 4, format = "Sint8x2" },
+    { offset = 80, shader_location = 5, format = "Sint8x4" },
+    { offset = 96, shader_location = 6, format = "Unorm8" },
+    { offset = 112, shader_location = 7, format = "Unorm8x2" },
+    { offset = 128, shader_location = 8, format = "Unorm8x4" },
+    { offset = 144, shader_location = 9, format = "Snorm8" },
+    { offset = 160, shader_location = 10, format = "Snorm8x2" },
+    { offset = 176, shader_location = 11, format = "Snorm8x4" },
+    { offset = 192, shader_location = 12, format = "Uint16" },
+    { offset = 208, shader_location = 13, format = "Uint16x2" },
+    { offset = 224, shader_location = 14, format = "Uint16x4" },
+    { offset = 240, shader_location = 15, format = "Sint16" },
+    { offset = 256, shader_location = 16, format = "Sint16x2" },
+    { offset = 272, shader_location = 17, format = "Sint16x4" },
+    { offset = 288, shader_location = 18, format = "Unorm16" },
+    { offset = 304, shader_location = 19, format = "Unorm16x2" },
+    { offset = 320, shader_location = 20, format = "Unorm16x4" },
+    { offset = 336, shader_location = 21, format = "Snorm16" },
+    { offset = 352, shader_location = 22, format = "Snorm16x2" },
+    { offset = 368, shader_location = 23, format = "Snorm16x4" },
+    { offset = 384, shader_location = 24, format = "Float16" },
+    { offset = 400, shader_location = 25, format = "Float16x2" },
+    { offset = 416, shader_location = 26, format = "Float16x4" },
+    { offset = 432, shader_location = 27, format = "Float32" },
+    { offset = 448, shader_location = 28, format = "Float32x2" },
+    { offset = 464, shader_location = 29, format = "Float32x3" },
+    { offset = 480, shader_location = 30, format = "Float32x4" },
+    { offset = 496, shader_location = 31, format = "Uint32" },
+    { offset = 512, shader_location = 32, format = "Uint32x2" },
+    { offset = 528, shader_location = 33, format = "Uint32x3" },
+    { offset = 544, shader_location = 34, format = "Uint32x4" },
+    { offset = 560, shader_location = 35, format = "Sint32" },
+    { offset = 576, shader_location = 36, format = "Sint32x2" },
+    { offset = 592, shader_location = 37, format = "Sint32x3" },
+    { offset = 608, shader_location = 38, format = "Sint32x4" },
+    { offset = 624, shader_location = 39, format = "unorm10-10-10-2" },
+    { offset = 640, shader_location = 40, format = "unorm8x4-bgra" },
+]
+id = 1
+indexed_by_vertex = true
+stride = 644

naga/tests/in/wgsl/msl-vpt-formats-x2.wgsl

@@ -0,0 +1,62 @@
+// Tests for the vertex pulling transform in the MSL backend.
+//
+// Test loading `vec2` from each vertex format.
+//
+// (Note: only the wgsl files are different for each of the -formats-xN tests.
+// The toml files are all the same.)
+
+struct VertexOutput {
+  @builtin(position) position: vec4<f32>,
+}
+
+struct VertexInput {
+  @location( 0) v_uint8     : vec2<u32>,
+  @location( 1) v_uint8x2   : vec2<u32>,
+  @location( 2) v_uint8x4   : vec2<u32>,
+  @location( 3) v_sint8     : vec2<i32>,
+  @location( 4) v_sint8x2   : vec2<i32>,
+  @location( 5) v_sint8x4   : vec2<i32>,
+  @location( 6) v_unorm8    : vec2<f32>,
+  @location( 7) v_unorm8x2  : vec2<f32>,
+  @location( 8) v_unorm8x4  : vec2<f32>,
+  @location( 9) v_snorm8    : vec2<f32>,
+  @location(10) v_snorm8x2  : vec2<f32>,
+  @location(11) v_snorm8x4  : vec2<f32>,
+  @location(12) v_uint16    : vec2<u32>,
+  @location(13) v_uint16x2  : vec2<u32>,
+  @location(14) v_uint16x4  : vec2<u32>,
+  @location(15) v_sint16    : vec2<i32>,
+  @location(16) v_sint16x2  : vec2<i32>,
+  @location(17) v_sint16x4  : vec2<i32>,
+  @location(18) v_unorm16   : vec2<f32>,
+  @location(19) v_unorm16x2 : vec2<f32>,
+  @location(20) v_unorm16x4 : vec2<f32>,
+  @location(21) v_snorm16   : vec2<f32>,
+  @location(22) v_snorm16x2 : vec2<f32>,
+  @location(23) v_snorm16x4 : vec2<f32>,
+  @location(24) v_float16   : vec2<f32>,
+  @location(25) v_float16x2 : vec2<f32>,
+  @location(26) v_float16x4 : vec2<f32>,
+  @location(27) v_float32   : vec2<f32>,
+  @location(28) v_float32x2 : vec2<f32>,
+  @location(29) v_float32x3 : vec2<f32>,
+  @location(30) v_float32x4 : vec2<f32>,
+  @location(31) v_uint32    : vec2<u32>,
+  @location(32) v_uint32x2  : vec2<u32>,
+  @location(33) v_uint32x3  : vec2<u32>,
+  @location(34) v_uint32x4  : vec2<u32>,
+  @location(35) v_sint32    : vec2<i32>,
+  @location(36) v_sint32x2  : vec2<i32>,
+  @location(37) v_sint32x3  : vec2<i32>,
+  @location(38) v_sint32x4  : vec2<i32>,
+  @location(39) v_unorm10_10_10_2: vec2<f32>,
+  @location(40) v_unorm8x4_bgra: vec2<f32>,
+}
+
+@vertex
+fn render_vertex(
+  v_in: VertexInput,
+) -> VertexOutput
+{
+  return VertexOutput(vec4f(v_in.v_uint8.x));
+}

naga/tests/in/wgsl/msl-vpt-formats-x3.toml

@@ -0,0 +1,53 @@
+targets = "METAL"
+
+[msl_pipeline]
+allow_and_force_point_size = false
+vertex_pulling_transform = true
+
+[[msl_pipeline.vertex_buffer_mappings]]
+attributes = [
+    { offset = 0, shader_location = 0, format = "Uint8" },
+    { offset = 16, shader_location = 1, format = "Uint8x2" },
+    { offset = 32, shader_location = 2, format = "Uint8x4" },
+    { offset = 48, shader_location = 3, format = "Sint8" },
+    { offset = 64, shader_location = 4, format = "Sint8x2" },
+    { offset = 80, shader_location = 5, format = "Sint8x4" },
+    { offset = 96, shader_location = 6, format = "Unorm8" },
+    { offset = 112, shader_location = 7, format = "Unorm8x2" },
+    { offset = 128, shader_location = 8, format = "Unorm8x4" },
+    { offset = 144, shader_location = 9, format = "Snorm8" },
+    { offset = 160, shader_location = 10, format = "Snorm8x2" },
+    { offset = 176, shader_location = 11, format = "Snorm8x4" },
+    { offset = 192, shader_location = 12, format = "Uint16" },
+    { offset = 208, shader_location = 13, format = "Uint16x2" },
+    { offset = 224, shader_location = 14, format = "Uint16x4" },
+    { offset = 240, shader_location = 15, format = "Sint16" },
+    { offset = 256, shader_location = 16, format = "Sint16x2" },
+    { offset = 272, shader_location = 17, format = "Sint16x4" },
+    { offset = 288, shader_location = 18, format = "Unorm16" },
+    { offset = 304, shader_location = 19, format = "Unorm16x2" },
+    { offset = 320, shader_location = 20, format = "Unorm16x4" },
+    { offset = 336, shader_location = 21, format = "Snorm16" },
+    { offset = 352, shader_location = 22, format = "Snorm16x2" },
+    { offset = 368, shader_location = 23, format = "Snorm16x4" },
+    { offset = 384, shader_location = 24, format = "Float16" },
+    { offset = 400, shader_location = 25, format = "Float16x2" },
+    { offset = 416, shader_location = 26, format = "Float16x4" },
+    { offset = 432, shader_location = 27, format = "Float32" },
+    { offset = 448, shader_location = 28, format = "Float32x2" },
+    { offset = 464, shader_location = 29, format = "Float32x3" },
+    { offset = 480, shader_location = 30, format = "Float32x4" },
+    { offset = 496, shader_location = 31, format = "Uint32" },
+    { offset = 512, shader_location = 32, format = "Uint32x2" },
+    { offset = 528, shader_location = 33, format = "Uint32x3" },
+    { offset = 544, shader_location = 34, format = "Uint32x4" },
+    { offset = 560, shader_location = 35, format = "Sint32" },
+    { offset = 576, shader_location = 36, format = "Sint32x2" },
+    { offset = 592, shader_location = 37, format = "Sint32x3" },
+    { offset = 608, shader_location = 38, format = "Sint32x4" },
+    { offset = 624, shader_location = 39, format = "unorm10-10-10-2" },
+    { offset = 640, shader_location = 40, format = "unorm8x4-bgra" },
+]
+id = 1
+indexed_by_vertex = true
+stride = 644