]> git.proxmox.com Git - rustc.git/blobdiff - vendor/packed_simd/src/codegen/reductions/mask/x86.rs
New upstream version 1.52.1+dfsg1
[rustc.git] / vendor / packed_simd / src / codegen / reductions / mask / x86.rs
diff --git a/vendor/packed_simd/src/codegen/reductions/mask/x86.rs b/vendor/packed_simd/src/codegen/reductions/mask/x86.rs
new file mode 100644 (file)
index 0000000..2ae4ed8
--- /dev/null
@@ -0,0 +1,194 @@
+//! Mask reductions implementation for `x86` and `x86_64` targets
+
+#[cfg(target_feature = "sse")]
+#[macro_use]
+mod sse;
+
+#[cfg(target_feature = "sse2")]
+#[macro_use]
+mod sse2;
+
+#[cfg(target_feature = "avx")]
+#[macro_use]
+mod avx;
+
+#[cfg(target_feature = "avx2")]
+#[macro_use]
+mod avx2;
+
+/// x86 64-bit m8x8 implementation
+macro_rules! x86_m8x8_impl {
+    ($id:ident) => {
+        cfg_if! {
+            if #[cfg(all(target_arch = "x86_64", target_feature = "sse"))] {
+                x86_m8x8_sse_impl!($id);
+            } else {
+                fallback_impl!($id);
+            }
+        }
+    };
+}
+
+/// x86 128-bit m8x16 implementation
+macro_rules! x86_m8x16_impl {
+    ($id:ident) => {
+        cfg_if! {
+            if #[cfg(target_feature = "sse2")] {
+                x86_m8x16_sse2_impl!($id);
+            } else {
+                fallback_impl!($id);
+            }
+        }
+    };
+}
+
+/// x86 128-bit m32x4 implementation
+macro_rules! x86_m32x4_impl {
+    ($id:ident) => {
+        cfg_if! {
+            if #[cfg(target_feature = "sse")] {
+                x86_m32x4_sse_impl!($id);
+            } else {
+                fallback_impl!($id);
+            }
+        }
+    };
+}
+
+/// x86 128-bit m64x2 implementation
+macro_rules! x86_m64x2_impl {
+    ($id:ident) => {
+        cfg_if! {
+            if #[cfg(target_feature = "sse2")] {
+                x86_m64x2_sse2_impl!($id);
+            } else if #[cfg(target_feature = "sse")] {
+                x86_m32x4_sse_impl!($id);
+            } else {
+                fallback_impl!($id);
+            }
+        }
+    };
+}
+
+/// x86 256-bit m8x32 implementation
+macro_rules! x86_m8x32_impl {
+    ($id:ident, $half_id:ident) => {
+        cfg_if! {
+            if #[cfg(target_feature = "avx2")] {
+                x86_m8x32_avx2_impl!($id);
+            } else if #[cfg(target_feature = "avx")] {
+                x86_m8x32_avx_impl!($id);
+            } else if #[cfg(target_feature = "sse2")] {
+                recurse_half!($id, $half_id);
+            } else {
+                fallback_impl!($id);
+            }
+        }
+    };
+}
+
+/// x86 256-bit m32x8 implementation
+macro_rules! x86_m32x8_impl {
+    ($id:ident, $half_id:ident) => {
+        cfg_if! {
+            if #[cfg(target_feature = "avx")] {
+                x86_m32x8_avx_impl!($id);
+            } else if #[cfg(target_feature = "sse")] {
+                recurse_half!($id, $half_id);
+            } else {
+                fallback_impl!($id);
+            }
+        }
+    };
+}
+
+/// x86 256-bit m64x4 implementation
+macro_rules! x86_m64x4_impl {
+    ($id:ident, $half_id:ident) => {
+        cfg_if! {
+            if #[cfg(target_feature = "avx")] {
+                x86_m64x4_avx_impl!($id);
+            } else if #[cfg(target_feature = "sse")] {
+                recurse_half!($id, $half_id);
+            } else {
+                fallback_impl!($id);
+            }
+        }
+    };
+}
+
+/// Fallback implementation.
+macro_rules! x86_intr_impl {
+    ($id:ident) => {
+    impl All for $id {
+        #[inline]
+        unsafe fn all(self) -> bool {
+        use crate::llvm::simd_reduce_all;
+            simd_reduce_all(self.0)
+        }
+    }
+        impl Any for $id {
+            #[inline]
+            unsafe fn any(self) -> bool {
+            use crate::llvm::simd_reduce_any;
+                simd_reduce_any(self.0)
+            }
+        }
+    };
+}
+
+/// Mask reduction implementation for `x86` and `x86_64` targets
+macro_rules! impl_mask_reductions {
+    // 64-bit wide masks
+    (m8x8) => { x86_m8x8_impl!(m8x8); };
+    (m16x4) => { x86_m8x8_impl!(m16x4); };
+    (m32x2) => { x86_m8x8_impl!(m32x2); };
+    // 128-bit wide masks
+    (m8x16) => { x86_m8x16_impl!(m8x16); };
+    (m16x8) => { x86_m8x16_impl!(m16x8); };
+    (m32x4) => { x86_m32x4_impl!(m32x4); };
+    (m64x2) => { x86_m64x2_impl!(m64x2); };
+    (m128x1) => { x86_intr_impl!(m128x1); };
+    // 256-bit wide masks:
+    (m8x32) => { x86_m8x32_impl!(m8x32, m8x16); };
+    (m16x16) => { x86_m8x32_impl!(m16x16, m16x8); };
+    (m32x8) => { x86_m32x8_impl!(m32x8, m32x4); };
+    (m64x4) => { x86_m64x4_impl!(m64x4, m64x2); };
+    (m128x2) => { x86_intr_impl!(m128x2); };
+    (msizex2) => {
+        cfg_if! {
+            if #[cfg(target_pointer_width = "64")] {
+                fallback_to_other_impl!(msizex2, m64x2);
+            } else if #[cfg(target_pointer_width = "32")] {
+                fallback_to_other_impl!(msizex2, m32x2);
+            } else {
+                compile_error!("unsupported target_pointer_width");
+            }
+        }
+    };
+    (msizex4) => {
+        cfg_if! {
+            if #[cfg(target_pointer_width = "64")] {
+                fallback_to_other_impl!(msizex4, m64x4);
+            } else if #[cfg(target_pointer_width = "32")] {
+                fallback_to_other_impl!(msizex4, m32x4);
+            } else {
+                compile_error!("unsupported target_pointer_width");
+            }
+        }
+    };
+    (msizex8) => {
+        cfg_if! {
+            if #[cfg(target_pointer_width = "64")] {
+                fallback_to_other_impl!(msizex8, m64x8);
+            } else if #[cfg(target_pointer_width = "32")] {
+                fallback_to_other_impl!(msizex8, m32x8);
+            } else {
+                compile_error!("unsupported target_pointer_width");
+            }
+        }
+    };
+
+    // Fallback to LLVM's default code-generation:
+    ($id:ident) => { fallback_impl!($id); };
+}