LDS access proposal

rfcs/000-safe-lds-access.md

@@ -0,0 +1,136 @@
+
+# Summary
+
+Provide safe access to a (limited) set of LDS memory without race conditions and without violating SPIR-V rules.
+
+# Explanation
+
+
+```rust
+struct LdsWriter;
+
+impl LdsWriter {
+    fn write_thread_idx(&mut self, value: T);
+    fn barrier(self) -> LdsReader;
+}
+
+struct LdsReader;
+
+impl LdsReader {
+    fn read(&self, idx: usize) -> T;
+    fn barrier(self) -> LdsWriter;
+}
+```
+## Barriers in non-uniform control flow
+
+```rust
+// sample 1:
+let lds = LdsWriter::<u32>::new();
+
+lds.write_thread_idx(0);
+
+// barrier should be in uniform control flow
+let value = if something {
+    lds.barrier().read(12)
+} else {
+    lds.barrier().read(23)
+};
+```
+
+## Multiple barriers with write in one branch
+```rust
+// sample 2:
+
+let lds = LdsWriter::<u32>::new();
+
+lds.write_thread_idx(0);
+
+let value = if something {
+    lds.barrier().read(12)
+} else {
+    let rdr = lds.barrier();
+    let value = rdr.read(23);
+
+    rdr.barrier().write_thread_idx(12);
+    value
+};
+```
+
+## Multiple consecutive writes to same location
+```rust
+// sample 3:
+let lds = LdsWriter::<u32>::new();
+
+lds.write_thread_idx(0);
+lds.write_thread_idx(666); // race?
+```
+
+## Multiple writes to same memory location after a read
+```rust
+// sample 4:
+let lds = LdsWriter::<u32>::new();
+
+lds.write_thread_idx(0);
+
+let (value, rdr) = if something {
+    let rdr = lds.barrier();
+    let value = rdr.read(12)
+    (value, rdr)
+} else {
+    let rdr = lds.barrier();
+    let value = rdr.read(12)
+    (value, rdr)
+};
+
+let wrt = rdr.barrier();
+wrt.write_thread_idx(12234);
+```
+
+# Drawbacks
+
+ * Limited amount of indexing for write operations
+
+# Alternatives
+
+@Tobski proposed making join operations explict and potentially passing in ranges in, to partition the data
+```rust
+struct BufferWriter;
+
+impl BufferWriter {
+    fn write_thread_idx(&mut self, value: T);
+}
+
+struct BufferReader;
+
+
+impl BufferReader {
+    fn read(&self, idx: usize) -> T;
+    fn partition_readers(self, ranges: [(usize, usize)]) -> [BufferReader];
+    fn partition_writers(self, ranges: [(usize, usize)]) -> [BufferWriter];
+}
+
+fn join_writers(a: BufferWriter, b: BufferWriter) -> BufferReader;
+fn join_readers(a: BufferReader, b: BufferReader) -> BufferReader;
+```
+
+@Jasper-Bekkers proposed making thread index a type with limited safe operations to extend the writer types with some more flexibilty. One concern here is also that this should be done in uniform controlflow most likely.
+
+```rust
+struct ThreadIdx;
+
+impl ThreadIdx {
+    /// XOR shuffle with a constant
+    fn butterfly_shuffle(&self, i: u32) -> ThreadIdx;
+
+    /// Shuffle all lanes up / down by a constant
+    fn up(&self, i: u32) -> ThreadIdx;
+    fn down(&self, i: u32) -> ThreadIdx;
+
+    /// Selecting arbitrary threads is unsafe
+    unsafe fn new(idx: u32) -> ThreadIdx;
+}
+```
+
+# Prior art
+
+ * Regular shading languages only provide unsafe/unsound access