cmov: add riscv64 optimised mask generation (#1397)

Same as #1396 but for `riscv64`.

Add tests in CI with target `riscv64gc-unknown-linux-gnu`.

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

Author: NicsTr

.github/workflows/cmov.yml

@@ -116,6 +116,9 @@ jobs:
             rust: 1.85.0 # MSRV
           - target: powerpc-unknown-linux-gnu
             rust: stable
+          # RISCV64
+          - target: riscv64gc-unknown-linux-gnu
+            rust: stable
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v6

cmov/README.md

@@ -37,6 +37,7 @@ on the following CPU architectures:
 - [x] `arm` (mask generation only)
 - [x] `aarch64` (`CSEL`)
 - [x] `riscv32` (mask generation only)
+- [x] `riscv64` (mask generation only)
 
 On other target architectures, a "best effort" portable fallback implementation
 based on bitwise arithmetic is used instead, augmented with tactical usage of

cmov/src/backends/soft.rs

@@ -122,13 +122,13 @@ fn maskne64(x: u64, y: u64) -> u64 {
 }
 
 /// Return a `u32::MAX` mask if `condition` is non-zero, otherwise return zero for a zero input.
-#[cfg(not(any(target_arch = "arm", target_arch = "riscv32")))]
+#[cfg(not(any(target_arch = "arm", target_arch = "riscv32", target_arch = "riscv64")))]
 fn masknz32(condition: u32) -> u32 {
     masknz!(condition: u32)
 }
 
 /// Return a `u64::MAX` mask if `condition` is non-zero, otherwise return zero for a zero input.
-#[cfg(not(any(target_arch = "arm", target_arch = "riscv32")))]
+#[cfg(not(any(target_arch = "arm", target_arch = "riscv32", target_arch = "riscv64")))]
 fn masknz64(condition: u64) -> u64 {
     masknz!(condition: u64)
 }
@@ -169,6 +169,28 @@ fn masknz32(condition: u32) -> u32 {
     mask
 }
 
+/// Optimized mask generation for riscv32 targets.
+#[cfg(target_arch = "riscv64")]
+fn masknz32(condition: u32) -> u32 {
+    (masknz64(condition.into()) & 0xFFFF_FFFF) as u32
+}
+
+/// Optimized mask generation for riscv32 targets.
+#[cfg(target_arch = "riscv64")]
+fn masknz64(condition: u64) -> u64 {
+    let mut mask: u64;
+    unsafe {
+        core::arch::asm!(
+            "seqz {0}, {1}",  // Set-if-not-zero pseudo-instruction
+            "addi {0}, {0}, -1", // Subtract 1, to have either full ones or full zeroes mask
+            lateout(reg) mask,
+            in(reg) condition,
+            options(nostack, nomem),
+        );
+    }
+    mask
+}
+
 /// 64-bit wrapper for targets that implement 32-bit mask generation in assembly.
 #[cfg(any(target_arch = "arm", target_arch = "riscv32"))]
 fn masknz64(condition: u64) -> u64 {