[naga wgsl-in] Attempt automatic conversion for arguments to user def…

…ined function calls (#6577)

When lowering arguments for a user-defined function call, avoid
concretizing the argument types. Instead make use of the existing
`try_automatic_conversions()` machinery to attempt to convert each
argument to the type expected by the function. This is straightforward
as user-defined functions only have a single overload.

This additionally changes an argument type in the test
parse_pointers() from `ptr<private>` to `ptr<function>`. The former is
invalid code which is indeed caught by the validator, but the test
only asserts that parsing succeeds, not validation. With this patch,
this error is now caught during parsing which caused the test to fail.

CHANGELOG.md

@@ -115,6 +115,7 @@ By @ErichDonGubler in [#6456](https://github.com/gfx-rs/wgpu/pull/6456), [#6148]
 - Expose Ray Query flags as constants in WGSL. Implement candidate intersections. By @kvark in [#5429](https://github.com/gfx-rs/wgpu/pull/5429)
 - Allow for override-expressions in `workgroup_size`. By @KentSlaney in [#6635](https://github.com/gfx-rs/wgpu/pull/6635).
 - Add support for OpAtomicCompareExchange in SPIR-V frontend. By @schell in [#6590](https://github.com/gfx-rs/wgpu/pull/6590).
+- Implement type inference for abstract arguments to user-defined functions. By @jamienicol in [#6577](https://github.com/gfx-rs/wgpu/pull/6577).
 
 #### General
 

naga/src/front/wgsl/lower/mod.rs

@@ -2184,7 +2184,25 @@ impl<'source, 'temp> Lowerer<'source, 'temp> {
             Some(&LoweredGlobalDecl::Function(function)) => {
                 let arguments = arguments
                     .iter()
-                    .map(|&arg| self.expression(arg, ctx))
+                    .enumerate()
+                    .map(|(i, &arg)| {
+                        // Try to convert abstract values to the known argument types
+                        let Some(&crate::FunctionArgument {
+                            ty: parameter_ty, ..
+                        }) = ctx.module.functions[function].arguments.get(i)
+                        else {
+                            // Wrong number of arguments... just concretize the type here
+                            // and let the validator report the error.
+                            return self.expression(arg, ctx);
+                        };
+
+                        let expr = self.expression_for_abstract(arg, ctx)?;
+                        ctx.try_automatic_conversions(
+                            expr,
+                            &crate::proc::TypeResolution::Handle(parameter_ty),
+                            ctx.ast_expressions.get_span(arg),
+                        )
+                    })
                     .collect::<Result<Vec<_>, _>>()?;
 
                 let has_result = ctx.module.functions[function].result.is_some();

naga/src/front/wgsl/tests.rs

@@ -84,6 +84,28 @@ fn parse_type_cast() {
     .is_err());
 }
 
+#[test]
+fn parse_type_coercion() {
+    parse_str(
+        "
+        fn foo(bar: f32) {}
+        fn main() {
+            foo(0);
+        }
+    ",
+    )
+    .unwrap();
+    assert!(parse_str(
+        "
+        fn foo(bar: i32) {}
+        fn main() {
+            foo(0.0);
+        }
+    ",
+    )
+    .is_err());
+}
+
 #[test]
 fn parse_struct() {
     parse_str(
@@ -461,7 +483,7 @@ fn binary_expression_mixed_scalar_and_vector_operands() {
 #[test]
 fn parse_pointers() {
     parse_str(
-        "fn foo(a: ptr<private, f32>) -> f32 { return *a; }
+        "fn foo(a: ptr<function, f32>) -> f32 { return *a; }
     fn bar() {
         var x: f32 = 1.0;
         let px = &x;

naga/tests/in/abstract-types-function-calls.wgsl

@@ -0,0 +1,38 @@
+fn func_f(a: f32) {}
+fn func_i(a: i32) {}
+fn func_u(a: u32) {}
+
+fn func_vf(a: vec2<f32>) {}
+fn func_vi(a: vec2<i32>) {}
+fn func_vu(a: vec2<u32>) {}
+
+fn func_mf(a: mat2x2<f32>) {}
+
+fn func_af(a: array<f32, 2>) {}
+fn func_ai(a: array<i32, 2>) {}
+fn func_au(a: array<u32, 2>) {}
+
+fn func_f_i(a: f32, b: i32) {}
+
+fn main() {
+    func_f(0.0);
+    func_f(0);
+    func_i(0);
+    func_u(0);
+
+    func_vf(vec2(0.0));
+    func_vf(vec2(0));
+    func_vi(vec2(0));
+    func_vu(vec2(0));
+
+    func_mf(mat2x2(vec2(0.0), vec2(0.0)));
+    func_mf(mat2x2(vec2(0), vec2(0)));
+
+    func_af(array(0.0, 0.0));
+    func_af(array(0, 0));
+    func_ai(array(0, 0));
+    func_au(array(0, 0));
+
+    func_f_i(0.0, 0);
+    func_f_i(0, 0);
+}

naga/tests/out/hlsl/abstract-types-function-calls.ron

@@ -0,0 +1,8 @@
+(
+    vertex:[
+    ],
+    fragment:[
+    ],
+    compute:[
+    ],
+)

naga/tests/out/msl/abstract-types-function-calls.msl

@@ -0,0 +1,103 @@
+// language: metal1.0
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using metal::uint;
+
+struct type_7 {
+    float inner[2];
+};
+struct type_8 {
+    int inner[2];
+};
+struct type_9 {
+    uint inner[2];
+};
+
+void func_f(
+    float a
+) {
+    return;
+}
+
+void func_i(
+    int a_1
+) {
+    return;
+}
+
+void func_u(
+    uint a_2
+) {
+    return;
+}
+
+void func_vf(
+    metal::float2 a_3
+) {
+    return;
+}
+
+void func_vi(
+    metal::int2 a_4
+) {
+    return;
+}
+
+void func_vu(
+    metal::uint2 a_5
+) {
+    return;
+}
+
+void func_mf(
+    metal::float2x2 a_6
+) {
+    return;
+}
+
+void func_af(
+    type_7 a_7
+) {
+    return;
+}
+
+void func_ai(
+    type_8 a_8
+) {
+    return;
+}
+
+void func_au(
+    type_9 a_9
+) {
+    return;
+}
+
+void func_f_i(
+    float a_10,
+    int b
+) {
+    return;
+}
+
+void main_(
+) {
+    func_f(0.0);
+    func_f(0.0);
+    func_i(0);
+    func_u(0u);
+    func_vf(metal::float2(0.0));
+    func_vf(metal::float2(0.0));
+    func_vi(metal::int2(0));
+    func_vu(metal::uint2(0u));
+    func_mf(metal::float2x2(metal::float2(0.0), metal::float2(0.0)));
+    func_mf(metal::float2x2(metal::float2(0.0), metal::float2(0.0)));
+    func_af(type_7 {0.0, 0.0});
+    func_af(type_7 {0.0, 0.0});
+    func_ai(type_8 {0, 0});
+    func_au(type_9 {0u, 0u});
+    func_f_i(0.0, 0);
+    func_f_i(0.0, 0);
+    return;
+}

naga/tests/out/spv/abstract-types-function-calls.spvasm

@@ -0,0 +1,145 @@
+; SPIR-V
+; Version: 1.1
+; Generator: rspirv
+; Bound: 100
+OpCapability Shader
+OpCapability Linkage
+%1 = OpExtInstImport "GLSL.std.450"
+OpMemoryModel Logical GLSL450
+OpDecorate %10 ArrayStride 4
+OpDecorate %12 ArrayStride 4
+OpDecorate %13 ArrayStride 4
+%2 = OpTypeVoid
+%3 = OpTypeFloat 32
+%4 = OpTypeInt 32 1
+%5 = OpTypeInt 32 0
+%6 = OpTypeVector %3 2
+%7 = OpTypeVector %4 2
+%8 = OpTypeVector %5 2
+%9 = OpTypeMatrix %6 2
+%11 = OpConstant  %5  2
+%10 = OpTypeArray %3 %11
+%12 = OpTypeArray %4 %11
+%13 = OpTypeArray %5 %11
+%17 = OpTypeFunction %2 %3
+%22 = OpTypeFunction %2 %4
+%27 = OpTypeFunction %2 %5
+%32 = OpTypeFunction %2 %6
+%37 = OpTypeFunction %2 %7
+%42 = OpTypeFunction %2 %8
+%47 = OpTypeFunction %2 %9
+%52 = OpTypeFunction %2 %10
+%57 = OpTypeFunction %2 %12
+%62 = OpTypeFunction %2 %13
+%68 = OpTypeFunction %2 %3 %4
+%72 = OpTypeFunction %2
+%73 = OpConstant  %3  0.0
+%74 = OpConstant  %4  0
+%75 = OpConstant  %5  0
+%76 = OpConstantComposite  %6  %73 %73
+%77 = OpConstantComposite  %7  %74 %74
+%78 = OpConstantComposite  %8  %75 %75
+%79 = OpConstantComposite  %9  %76 %76
+%80 = OpConstantComposite  %10  %73 %73
+%81 = OpConstantComposite  %12  %74 %74
+%82 = OpConstantComposite  %13  %75 %75
+%16 = OpFunction  %2  None %17
+%15 = OpFunctionParameter  %3
+%14 = OpLabel
+OpBranch %18
+%18 = OpLabel
+OpReturn
+OpFunctionEnd
+%21 = OpFunction  %2  None %22
+%20 = OpFunctionParameter  %4
+%19 = OpLabel
+OpBranch %23
+%23 = OpLabel
+OpReturn
+OpFunctionEnd
+%26 = OpFunction  %2  None %27
+%25 = OpFunctionParameter  %5
+%24 = OpLabel
+OpBranch %28
+%28 = OpLabel
+OpReturn
+OpFunctionEnd
+%31 = OpFunction  %2  None %32
+%30 = OpFunctionParameter  %6
+%29 = OpLabel
+OpBranch %33
+%33 = OpLabel
+OpReturn
+OpFunctionEnd
+%36 = OpFunction  %2  None %37
+%35 = OpFunctionParameter  %7
+%34 = OpLabel
+OpBranch %38
+%38 = OpLabel
+OpReturn
+OpFunctionEnd
+%41 = OpFunction  %2  None %42
+%40 = OpFunctionParameter  %8
+%39 = OpLabel
+OpBranch %43
+%43 = OpLabel
+OpReturn
+OpFunctionEnd
+%46 = OpFunction  %2  None %47
+%45 = OpFunctionParameter  %9
+%44 = OpLabel
+OpBranch %48
+%48 = OpLabel
+OpReturn
+OpFunctionEnd
+%51 = OpFunction  %2  None %52
+%50 = OpFunctionParameter  %10
+%49 = OpLabel
+OpBranch %53
+%53 = OpLabel
+OpReturn
+OpFunctionEnd
+%56 = OpFunction  %2  None %57
+%55 = OpFunctionParameter  %12
+%54 = OpLabel
+OpBranch %58
+%58 = OpLabel
+OpReturn
+OpFunctionEnd
+%61 = OpFunction  %2  None %62
+%60 = OpFunctionParameter  %13
+%59 = OpLabel
+OpBranch %63
+%63 = OpLabel
+OpReturn
+OpFunctionEnd
+%67 = OpFunction  %2  None %68
+%65 = OpFunctionParameter  %3
+%66 = OpFunctionParameter  %4
+%64 = OpLabel
+OpBranch %69
+%69 = OpLabel
+OpReturn
+OpFunctionEnd
+%71 = OpFunction  %2  None %72
+%70 = OpLabel
+OpBranch %83
+%83 = OpLabel
+%84 = OpFunctionCall  %2  %16 %73
+%85 = OpFunctionCall  %2  %16 %73
+%86 = OpFunctionCall  %2  %21 %74
+%87 = OpFunctionCall  %2  %26 %75
+%88 = OpFunctionCall  %2  %31 %76
+%89 = OpFunctionCall  %2  %31 %76
+%90 = OpFunctionCall  %2  %36 %77
+%91 = OpFunctionCall  %2  %41 %78
+%92 = OpFunctionCall  %2  %46 %79
+%93 = OpFunctionCall  %2  %46 %79
+%94 = OpFunctionCall  %2  %51 %80
+%95 = OpFunctionCall  %2  %51 %80
+%96 = OpFunctionCall  %2  %56 %81
+%97 = OpFunctionCall  %2  %61 %82
+%98 = OpFunctionCall  %2  %67 %73 %74
+%99 = OpFunctionCall  %2  %67 %73 %74
+OpReturn
+OpFunctionEnd