Add is_helper_invocation, and cleanup arch & asm functions (#612)

crates/spirv-std/src/arch.rs

@@ -8,6 +8,7 @@ use crate::{scalar::Scalar, vector::Vector};
 mod arithmetic;
 #[cfg(feature = "const-generics")]
 mod barrier;
+mod demote_to_helper_invocation_ext;
 mod derivative;
 mod primitive;
 mod ray_tracing;
@@ -15,6 +16,7 @@ mod ray_tracing;
 pub use arithmetic::*;
 #[cfg(feature = "const-generics")]
 pub use barrier::*;
+pub use demote_to_helper_invocation_ext::*;
 pub use derivative::*;
 pub use primitive::*;
 pub use ray_tracing::*;
@@ -140,3 +142,18 @@ pub unsafe fn vector_insert_dynamic<T: Scalar, V: Vector<T, N>, const N: usize>(
 
     result
 }
+
+/// Fragment-shader discard. Equivalvent to `discard()` from GLSL
+///
+/// Ceases all further processing in any invocation that executes it: Only
+/// instructions these invocations executed before [kill] have observable side
+/// effects.
+#[spirv_std_macros::gpu_only]
+#[doc(alias = "OpKill", alias = "discard")]
+#[allow(clippy::empty_loop)]
+pub fn kill() -> ! {
+    unsafe {
+        asm!("OpKill", "%unused = OpLabel");
+    }
+    loop {}
+}

crates/spirv-std/src/arch/demote_to_helper_invocation_ext.rs

@@ -0,0 +1,48 @@
+/// Demote fragment shader invocation to a helper invocation. Equivalvent to
+/// `discard()` in HLSL. Any stores to memory after this instruction are
+/// suppressed and the fragment does not write outputs to the framebuffer.
+///
+/// Unlike [super::kill], this does not necessarily terminate the invocation. It
+/// is not considered a flow control instruction (flow control does not become
+/// non-uniform) and does not terminate the block.
+///
+/// - **Required Capabilities** `DemoteToHelperInvocationEXT`
+/// - **Required Extensions** `SPV_EXT_demote_to_helper_invocation`
+///
+/// # Safety
+/// After this instruction executes, the value of a `helper_invocation` builtin
+/// variable is undefined. Use `is_helper_invocation` to determine whether
+/// invocations are helper invocations in the presence
+/// of [demote_to_helper_invocation].
+#[spirv_std_macros::gpu_only]
+#[doc(alias = "OpDemoteToHelperInvocationEXT", alias = "discard")]
+pub unsafe fn demote_to_helper_invocation() {
+    asm!("OpDemoteToHelperInvocationEXT");
+}
+
+/// Returns `true` if the invocation is currently a helper invocation, otherwise
+/// result is `false`. An invocation is currently a helper invocation if it was
+/// originally invoked as a helper invocation or if it has been demoted to a
+/// helper invocation by [demote_to_helper_invocation].
+///
+/// - **Required Capabilities** `DemoteToHelperInvocationEXT`
+/// - **Required Extensions** `SPV_EXT_demote_to_helper_invocation`
+#[spirv_std_macros::gpu_only]
+#[doc(alias = "OpIsHelperInvocationEXT")]
+pub fn is_helper_invocation() -> bool {
+    let result: u32;
+
+    unsafe {
+        asm! {
+            "%bool = OpTypeBool",
+            "%u32 = OpTypeInt 32 0",
+            "%zero = OpConstant %u32 0",
+            "%one = OpConstant %u32 1",
+            "%result = OpIsHelperInvocationEXT %bool",
+            "{} = OpSelect %u32 %result %one %zero",
+            out(reg) result
+        };
+    }
+
+    result != 0
+}

crates/spirv-std/src/arch/derivative.rs

@@ -1,19 +1,10 @@
 use crate::float::Float;
 
-#[cfg(target_arch = "spirv")]
-macro_rules! deriv_caps {
-    (true) => {
-        asm!("OpCapability DerivativeControl")
-    };
-    (false) => {};
-}
-
 #[cfg(target_arch = "spirv")]
 macro_rules! deriv_fn {
-    ($p:ident, $inst:ident, $needs_caps:tt) => {
+    ($p:ident, $inst:ident) => {
         unsafe {
             let mut o = Default::default();
-            deriv_caps!($needs_caps);
             asm!(
                 "%input = OpLoad _ {0}",
                 concat!("%result = ", stringify!($inst), " _ %input"),
@@ -32,7 +23,7 @@ macro_rules! deriv_fn {
 #[crate::macros::vectorized]
 #[crate::macros::gpu_only]
 pub fn ddx<F: Float>(component: F) -> F {
-    deriv_fn!(component, OpDPdx, false)
+    deriv_fn!(component, OpDPdx)
 }
 
 /// Returns the partial derivative of `component` with respect to the window's X
@@ -41,7 +32,7 @@ pub fn ddx<F: Float>(component: F) -> F {
 #[crate::macros::vectorized]
 #[crate::macros::gpu_only]
 pub fn ddx_fine<F: Float>(component: F) -> F {
-    deriv_fn!(component, OpDPdxFine, true)
+    deriv_fn!(component, OpDPdxFine)
 }
 
 /// Returns the partial derivative of `component` with respect to the window's X
@@ -53,7 +44,7 @@ pub fn ddx_fine<F: Float>(component: F) -> F {
 #[crate::macros::vectorized]
 #[crate::macros::gpu_only]
 pub fn ddx_coarse<F: Float>(component: F) -> F {
-    deriv_fn!(component, OpDPdxCoarse, true)
+    deriv_fn!(component, OpDPdxCoarse)
 }
 
 /// Returns the partial derivative of `component` with respect to the window's Y
@@ -62,7 +53,7 @@ pub fn ddx_coarse<F: Float>(component: F) -> F {
 #[crate::macros::vectorized]
 #[crate::macros::gpu_only]
 pub fn ddy<F: Float>(component: F) -> F {
-    deriv_fn!(component, OpDPdy, false)
+    deriv_fn!(component, OpDPdy)
 }
 
 /// Returns the partial derivative of `component` with respect to the window's Y
@@ -71,7 +62,7 @@ pub fn ddy<F: Float>(component: F) -> F {
 #[crate::macros::vectorized]
 #[crate::macros::gpu_only]
 pub fn ddy_fine<F: Float>(component: F) -> F {
-    deriv_fn!(component, OpDPdyFine, true)
+    deriv_fn!(component, OpDPdyFine)
 }
 
 /// Returns the partial derivative of `component` with respect to the window's Y
@@ -83,7 +74,7 @@ pub fn ddy_fine<F: Float>(component: F) -> F {
 #[crate::macros::vectorized]
 #[crate::macros::gpu_only]
 pub fn ddy_coarse<F: Float>(component: F) -> F {
-    deriv_fn!(component, OpDPdyCoarse, true)
+    deriv_fn!(component, OpDPdyCoarse)
 }
 
 /// Returns the sum of the absolute values of [`ddx`] and [`ddy`] as a single
@@ -91,7 +82,7 @@ pub fn ddy_coarse<F: Float>(component: F) -> F {
 #[crate::macros::vectorized]
 #[crate::macros::gpu_only]
 pub fn fwidth<F: Float>(component: F) -> F {
-    deriv_fn!(component, OpFwidth, false)
+    deriv_fn!(component, OpFwidth)
 }
 
 /// Returns the sum of the absolute values of [`ddx_fine`] and [`ddy_fine`] as a
@@ -99,7 +90,7 @@ pub fn fwidth<F: Float>(component: F) -> F {
 #[crate::macros::vectorized]
 #[crate::macros::gpu_only]
 pub fn fwidth_fine<F: Float>(component: F) -> F {
-    deriv_fn!(component, OpFwidthFine, true)
+    deriv_fn!(component, OpFwidthFine)
 }
 
 /// Returns the sum of the absolute values of [`ddx_coarse`] and [`ddy_coarse`]
@@ -107,5 +98,5 @@ pub fn fwidth_fine<F: Float>(component: F) -> F {
 #[crate::macros::vectorized]
 #[crate::macros::gpu_only]
 pub fn fwidth_coarse<F: Float>(component: F) -> F {
-    deriv_fn!(component, OpFwidthCoarse, true)
+    deriv_fn!(component, OpFwidthCoarse)
 }

crates/spirv-std/src/arch/ray_tracing.rs

@@ -38,24 +38,27 @@ pub unsafe fn report_intersection(hit: f32, hit_kind: u32) -> bool {
 
 /// Ignores the current potential intersection, terminating the invocation that
 /// executes it, and continues the ray traversal.  This instruction is allowed
-/// only in `any_hit` execution model.  This instruction must be the last
-/// instruction in a block.
+/// only in `any_hit` execution model.
 #[spirv_std_macros::gpu_only]
 #[doc(alias = "OpIgnoreIntersectionKHR")]
 #[inline]
-pub unsafe fn ignore_intersection() {
-    asm!("OpIgnoreIntersectionKHR", "%unused = OpLabel")
+#[allow(clippy::empty_loop)]
+pub unsafe fn ignore_intersection() -> ! {
+    asm!("OpIgnoreIntersectionKHR", "%unused = OpLabel");
+    loop {}
 }
 
 /// Terminates the invocation that executes it, stops the ray traversal, accepts
 /// the current hit, and invokes the `closest_hit` execution model
-/// (if active).  This instruction is allowed only in the `any_hit`
-/// execution model.  This instruction must be the last instruction in a block.
+/// (if active). This instruction is allowed only in the `any_hit`
+/// execution model.
 #[spirv_std_macros::gpu_only]
 #[doc(alias = "OpTerminateRayKHR")]
 #[inline]
-pub unsafe fn terminate_ray() {
-    asm!("OpTerminateRayKHR", "%unused = OpLabel")
+#[allow(clippy::empty_loop)]
+pub unsafe fn terminate_ray() -> ! {
+    asm!("OpTerminateRayKHR", "%unused = OpLabel");
+    loop {}
 }
 
 /// Invoke a callable shader.

crates/spirv-std/src/lib.rs

@@ -93,26 +93,6 @@ pub use crate::macros::Image;
 pub use num_traits;
 pub use textures::*;
 
-/// Calls the `OpDemoteToHelperInvocationEXT` instruction, which corresponds to discard() in HLSL
-#[spirv_std_macros::gpu_only]
-pub fn demote_to_helper_invocation() {
-    unsafe {
-        asm!(
-            "OpExtension \"SPV_EXT_demote_to_helper_invocation\"",
-            "OpCapability DemoteToHelperInvocationEXT",
-            "OpDemoteToHelperInvocationEXT"
-        );
-    }
-}
-
-/// Calls the `OpKill` instruction, which corresponds to discard() in GLSL
-#[spirv_std_macros::gpu_only]
-pub fn discard() {
-    unsafe {
-        asm!("OpKill", "%unused = OpLabel");
-    }
-}
-
 #[cfg(all(not(test), target_arch = "spirv"))]
 #[panic_handler]
 fn panic(_: &core::panic::PanicInfo<'_>) -> ! {

crates/spirv-std/src/ray_tracing.rs

@@ -491,17 +491,15 @@ impl RayQuery {
     #[doc(alias = "OpRayQueryGetIntersectionFrontFaceKHR")]
     #[inline]
     pub unsafe fn get_intersection_front_face<const INTERSECTION: u32>(&self) -> bool {
-        let mut result = 0u8;
+        let mut result: u32;
 
         asm! {
-            "%u8 = OpTypeInt 8 0",
             "%u32 = OpTypeInt 32 0",
             "%intersection = OpConstant %u32 {intersection}",
-            "%result = OpRayQueryGetIntersectionFrontFaceKHR %u8 {ray_query} %intersection",
-            "OpStore {result} %result",
+            "{result} = OpRayQueryGetIntersectionFrontFaceKHR %u32 {ray_query} %intersection",
             ray_query = in(reg) self,
             intersection = const INTERSECTION,
-            result = in(reg) &mut result,
+            result = out(reg) result,
         }
 
         result != 0
@@ -513,14 +511,13 @@ impl RayQuery {
     #[doc(alias = "OpRayQueryGetIntersectionCandidateAABBOpaqueKHR")]
     #[inline]
     pub unsafe fn get_intersection_candidate_aabb_opaque(&self) -> bool {
-        let mut result = 0u8;
+        let result: u32;
 
         asm! {
-            "%u8 = OpTypeInt 8 0",
-            "%result = OpRayQueryGetIntersectionCandidateAABBOpaqueKHR %u8 {ray_query}",
-            "OpStore {result} %result",
+            "%u32 = OpTypeInt 8 0",
+            "{result} = OpRayQueryGetIntersectionCandidateAABBOpaqueKHR %u32 {ray_query}",
             ray_query = in(reg) self,
-            result = in(reg) &mut result,
+            result = out(reg) result,
         }
 
         result != 0

tests/ui/arch/demote_to_helper_invocation.rs

@@ -0,0 +1,9 @@
+// build-pass
+//
+// compile-flags: -C target-feature=+DemoteToHelperInvocationEXT,+ext:SPV_EXT_demote_to_helper_invocation
+
+#[spirv(fragment)]
+pub fn main() {
+    unsafe { spirv_std::arch::demote_to_helper_invocation() };
+    assert!(spirv_std::arch::is_helper_invocation());
+}

tests/ui/arch/emit_stream_vertex.rs

@@ -1,9 +1,9 @@
 // build-pass
+// compile-flags: -C target-feature=+GeometryStreams
 
 #[spirv(geometry(input_lines = 2, output_points = 2))]
 pub fn main() {
     unsafe {
-        asm!("OpCapability GeometryStreams");
         spirv_std::arch::emit_stream_vertex::<2>();
     };
 }

tests/ui/arch/emit_vertex.rs

@@ -1,9 +1,9 @@
 // build-pass
+// compile-flags: -Ctarget-feature=+Geometry
 
 #[spirv(geometry(input_lines = 2, output_points = 2))]
 pub fn main() {
     unsafe {
-        asm!("OpCapability Geometry");
         spirv_std::arch::emit_vertex();
     };
 }

tests/ui/arch/end_primitive.rs

@@ -1,9 +1,9 @@
 // build-pass
+// compile-flags: -Ctarget-feature=+Geometry
 
 #[spirv(geometry(input_lines = 2, output_points = 2))]
 pub fn main() {
     unsafe {
-        asm!("OpCapability Geometry");
         spirv_std::arch::end_primitive();
     };
 }

tests/ui/arch/end_stream_primitive.rs

@@ -1,9 +1,9 @@
 // build-pass
+// compile-flags: -C target-feature=+GeometryStreams
 
 #[spirv(geometry(input_lines = 2, output_points = 2))]
 pub fn main() {
     unsafe {
-        asm!("OpCapability GeometryStreams");
         spirv_std::arch::end_stream_primitive::<2>();
     };
 }

tests/ui/arch/execute_callable.rs

@@ -1,4 +1,5 @@
 // build-pass
+// compile-flags: -Ctarget-feature=+RayTracingKHR,+ext:SPV_KHR_ray_tracing
 
 #[spirv(ray_generation)]
 // Rustfmt will eat long attributes (https://github.com/rust-lang/rustfmt/issues/4579)
@@ -9,8 +10,6 @@ pub fn main(
     #[spirv(incoming_callable_data)] payload: &glam::Vec3,
 ) {
     unsafe {
-        asm!(r#"OpExtension "SPV_KHR_ray_tracing""#);
-        asm!("OpCapability RayTracingKHR");
         spirv_std::arch::execute_callable::<_, 5>(payload);
     }
 }

tests/ui/arch/kill.rs

@@ -0,0 +1,6 @@
+// build-pass
+
+#[spirv(fragment)]
+pub fn main() {
+    spirv_std::arch::kill();
+}

tests/ui/image/issue_527.rs

@@ -1,3 +1,7 @@
+// build-pass
+//
+// compile-flags: -C target-feature=+StorageImageWriteWithoutFormat
+
 use glam::*;
 
 #[spirv(compute(threads(1)))]
@@ -6,7 +10,6 @@ pub fn main_cs(
     #[spirv(storage_buffer, descriptor_set = 0, binding = 0)] points_buffer: &mut [UVec2; 100],
     #[spirv(descriptor_set = 1, binding = 1)] image: &spirv_std::Image!(2D, type=f32, sampled=false),
 ) {
-    unsafe { asm!("OpCapability StorageImageWriteWithoutFormat") };
     let position = id.xy();
     for i in 0..100usize {
         let p0 = &points_buffer[i];

tests/ui/image/query/query_levels.rs

@@ -1,4 +1,5 @@
 // build-pass
+// compile-flags: -C target-feature=+ImageQuery
 
 use spirv_std::{arch, Image};
 
@@ -7,6 +8,5 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 0)] image: &Image!(2D, type=f32, sampled),
     output: &mut u32,
 ) {
-    unsafe { asm!("OpCapability ImageQuery") };
     *output = image.query_levels();
 }

tests/ui/image/query/query_levels_err.rs

@@ -1,5 +1,6 @@
 // build-fail
 // normalize-stderr-test "\S*/crates/spirv-std/src/" -> "$$SPIRV_STD_SRC/"
+// compile-flags: -C target-feature=+ImageQuery
 
 use spirv_std::{arch, Image};
 
@@ -8,6 +9,5 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 0)] image: &Image!(rect, type=f32, sampled),
     output: &mut u32,
 ) {
-    unsafe { asm!("OpCapability ImageQuery") };
     *output = image.query_levels();
 }

tests/ui/image/query/query_lod.rs

@@ -1,4 +1,5 @@
 // build-pass
+// compile-flags: -C target-feature=+ImageQuery
 
 use spirv_std::{arch, Image, Sampler};
 
@@ -8,6 +9,5 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 1)] sampler: &Sampler,
     output: &mut glam::Vec2,
 ) {
-    unsafe { asm!("OpCapability ImageQuery") };
     *output = image.query_lod(*sampler, glam::Vec2::new(0.0, 1.0));
 }

tests/ui/image/query/query_lod_err.rs

@@ -1,5 +1,6 @@
 // build-fail
 // normalize-stderr-test "\S*/crates/spirv-std/src/" -> "$$SPIRV_STD_SRC/"
+// compile-flags: -C target-feature=+ImageQuery
 
 use spirv_std::{arch, Image, Sampler};
 
@@ -9,6 +10,5 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 1)] sampler: &Sampler,
     output: &mut glam::Vec2,
 ) {
-    unsafe { asm!("OpCapability ImageQuery") };
     *output = image.query_lod(*sampler, glam::Vec2::new(0.0, 1.0));
 }

tests/ui/image/query/query_samples.rs

@@ -1,4 +1,5 @@
 // build-pass
+// compile-flags: -C target-feature=+ImageQuery
 
 use spirv_std::{arch, Image};
 
@@ -7,6 +8,5 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 0)] image: &Image!(2D, type=f32, sampled, multisampled),
     output: &mut u32,
 ) {
-    unsafe { asm!("OpCapability ImageQuery") };
     *output = image.query_samples();
 }

tests/ui/image/query/query_size.rs

@@ -1,4 +1,5 @@
 // build-pass
+// compile-flags: -C target-feature=+ImageQuery
 
 use spirv_std::{arch, Image};
 
@@ -7,6 +8,5 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 0)] image: &Image!(2D, type=f32, sampled=false),
     output: &mut glam::UVec2,
 ) {
-    unsafe { asm!("OpCapability ImageQuery") };
     *output = image.query_size();
 }

tests/ui/image/query/query_size_err.rs

@@ -1,5 +1,6 @@
 // build-fail
 // normalize-stderr-test "\S*/crates/spirv-std/src/" -> "$$SPIRV_STD_SRC/"
+// compile-flags: -C target-feature=+ImageQuery
 
 use spirv_std::{arch, Image};
 
@@ -8,6 +9,5 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 0)] image: &Image!(2D, type=f32, sampled),
     output: &mut glam::UVec2,
 ) {
-    unsafe { asm!("OpCapability ImageQuery") };
     *output = image.query_size();
 }

tests/ui/image/query/query_size_lod.rs

@@ -1,4 +1,5 @@
 // build-pass
+// compile-flags: -C target-feature=+ImageQuery
 
 use spirv_std::{arch, Image};
 
@@ -7,6 +8,5 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 0)] image: &Image!(2D, type=f32, sampled),
     output: &mut glam::UVec2,
 ) {
-    unsafe { asm!("OpCapability ImageQuery") };
     *output = image.query_size_lod(0);
 }

tests/ui/image/query/query_size_lod_err.rs

@@ -1,5 +1,6 @@
 // build-fail
 // normalize-stderr-test "\S*/crates/spirv-std/src/" -> "$$SPIRV_STD_SRC/"
+// compile-flags: -C target-feature=+ImageQuery
 
 use spirv_std::{arch, Image};
 
@@ -8,6 +9,5 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 0)] image: &Image!(rect, type=f32, sampled),
     output: &mut glam::UVec2,
 ) {
-    unsafe { asm!("OpCapability ImageQuery") };
     *output = image.query_size_lod(0);
 }

tests/ui/image/read.rs

@@ -1,5 +1,6 @@
 // Test `OpImageRead`
 // build-pass
+// compile-flags: -C target-feature=+StorageImageReadWithoutFormat
 
 use spirv_std::{arch, Image};
 
@@ -8,7 +9,6 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 0)] image: &Image!(2D, type=f32, sampled=false),
     output: &mut glam::Vec4,
 ) {
-    unsafe { asm!("OpCapability StorageImageReadWithoutFormat") };
     let coords = image.read(glam::IVec2::new(0, 1));
     *output = coords;
 }

tests/ui/image/write.rs

@@ -1,5 +1,6 @@
 // Test `OpImageWrite`
 // build-pass
+// compile-flags: -C target-feature=+StorageImageWriteWithoutFormat
 
 use spirv_std::{arch, Image};
 
@@ -9,7 +10,6 @@ pub fn main(
     #[spirv(descriptor_set = 0, binding = 0)] image: &Image!(2D, type=f32, sampled=false),
 ) {
     unsafe {
-        asm!("OpCapability StorageImageWriteWithoutFormat");
         image.write(glam::UVec2::new(0, 1), texels);
     }
 }

tests/ui/lang/consts/nested-ref.rs

@@ -2,6 +2,7 @@
 // contain references, and where the `T` values aren't immediatelly loaded from.
 
 // build-pass
+// compile-flags: -C target-feature=+VariablePointers
 
 use spirv_std as _;
 
@@ -26,7 +27,6 @@ pub fn main(
     bool_out: &mut bool,
     vec_out: &mut Vec2,
 ) {
-    unsafe { asm!("OpCapability VariablePointers") };
     *scalar_out = deep_load(&&123);
     *bool_out = vec_in == &Vec2::ZERO;
     *vec_out = deep_transpose(&ROT90) * *vec_in;

tests/ui/target_features_err.rs

@@ -7,4 +7,3 @@ use spirv_std as _;
 pub fn main() {
     unsafe { spirv_std::arch::terminate_ray() }
 }
-