Subgroup intrinsics (#14)

* subgroup: add trait VectorOrScalar, representing either a vector or a scalar type

* subgroup: added all non-uniform subgroup operations

* subgroup: remove all target_feature cfgs, replaced with docs

* subgroup: added all subgroupBarrier*() functions from glsl

* subgroup: added non group-op tests

* subgroup: fixed asm for instructions taking GROUP_OP generic

* subgroup: added tests for group-op instructions

* gitignore: added rustc-ice* error reports

* subgroup: added test for subgroup buildins

* subgroup: make SubgroupMask a struct to prevent implicit casts to and from UVec4

* subgroup: fixed clippy lints

* subgroup: drop the `non_uniform` from all subgroup functions, matching glsl

* changelog: add subgroup intrinsics PR

* subgroup: make VectorOrScalar trait match discussions in EmbarkStudios/rust-gpu#1030

* cleanup: remove internal type F32x2 for glam::Vec2

---------

Co-authored-by: Firestar99 <[email protected]>

Author: Firestar99

.gitignore

@@ -3,3 +3,4 @@ target/
 .vim/
 tests/Cargo.lock
 .github/install-spirv-tools/Cargo.lock
+rustc-ice-*.txt

CHANGELOG.md

@@ -33,6 +33,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Signed for loops like `for _ in 0..4i32 {}` no longer compile. We recommend switching to unsigned for loops and casting back to signed integers in the meanwhile.
 
 ### Changed 🛠
+- [PR#14](https://github.com/Rust-GPU/rust-gpu/pull/14) add subgroup intrinsics matching glsl's [`GL_KHR_shader_subgroup`](https://github.com/KhronosGroup/GLSL/blob/main/extensions/khr/GL_KHR_shader_subgroup.txt)
 - [PR#13](https://github.com/Rust-GPU/rust-gpu/pull/13) allow cargo features to be passed to the shader crate
 - [PR#12](https://github.com/rust-gpu/rust-gpu/pull/12) updated toolchain to `nightly-2024-04-24`
 - [PR#9](https://github.com/Rust-GPU/rust-gpu/pull/9) relaxed `glam` version requirements (`>=0.22, <=0.29`)

crates/rustc_codegen_spirv/src/builder/spirv_asm.rs

@@ -2,11 +2,13 @@ use super::Builder;
 use crate::builder_spirv::{BuilderCursor, SpirvValue};
 use crate::codegen_cx::CodegenCx;
 use crate::spirv_type::SpirvType;
+use num_traits::FromPrimitive;
 use rspirv::dr;
 use rspirv::grammar::{reflect, LogicalOperand, OperandKind, OperandQuantifier};
 use rspirv::spirv::{
-    FPFastMathMode, FragmentShadingRate, FunctionControl, ImageOperands, KernelProfilingInfo,
-    LoopControl, MemoryAccess, MemorySemantics, Op, RayFlags, SelectionControl, StorageClass, Word,
+    FPFastMathMode, FragmentShadingRate, FunctionControl, GroupOperation, ImageOperands,
+    KernelProfilingInfo, LoopControl, MemoryAccess, MemorySemantics, Op, RayFlags,
+    SelectionControl, StorageClass, Word,
 };
 use rustc_ast::ast::{InlineAsmOptions, InlineAsmTemplatePiece};
 use rustc_codegen_ssa::mir::place::PlaceRef;
@@ -1347,10 +1349,15 @@ impl<'cx, 'tcx> Builder<'cx, 'tcx> {
                 Ok(x) => inst.operands.push(dr::Operand::Scope(x)),
                 Err(()) => self.err(format!("unknown Scope {word}")),
             },
-            (OperandKind::GroupOperation, Some(word)) => match word.parse() {
-                Ok(x) => inst.operands.push(dr::Operand::GroupOperation(x)),
-                Err(()) => self.err(format!("unknown GroupOperation {word}")),
-            },
+            (OperandKind::GroupOperation, Some(word)) => {
+                match word.parse::<u32>().ok().and_then(GroupOperation::from_u32) {
+                    Some(id) => inst.operands.push(dr::Operand::GroupOperation(id)),
+                    None => match word.parse() {
+                        Ok(x) => inst.operands.push(dr::Operand::GroupOperation(x)),
+                        Err(()) => self.err(format!("unknown GroupOperation {word}")),
+                    },
+                }
+            }
             (OperandKind::KernelEnqueueFlags, Some(word)) => match word.parse() {
                 Ok(x) => inst.operands.push(dr::Operand::KernelEnqueueFlags(x)),
                 Err(()) => self.err(format!("unknown KernelEnqueueFlags {word}")),

crates/spirv-std/src/arch.rs

@@ -19,13 +19,15 @@ mod demote_to_helper_invocation_ext;
 mod derivative;
 mod primitive;
 mod ray_tracing;
+mod subgroup;
 
 pub use atomics::*;
 pub use barrier::*;
 pub use demote_to_helper_invocation_ext::*;
 pub use derivative::*;
 pub use primitive::*;
 pub use ray_tracing::*;
+pub use subgroup::*;
 
 /// Result is true if any component of `vector` is true, otherwise result is
 /// false.