Auto merge of #49249 - gnzlbg:simd_minmax, r=alexcrichton

implement minmax intrinsics

This adds the `simd_{fmin,fmax}` intrinsics, which do a vertical (lane-wise) `min`/`max` for floating point vectors that's equivalent to Rust's `min`/`max` for `f32`/`f64`.

It might make sense to make `{f32,f64}::{min,max}` use the `minnum` and `minmax` intrinsics as well.

---

~~HELP: I need some help with these. Either I should go to sleep or there must be something that I must be missing. AFAICT I am calling the `maxnum` builder correctly, yet rustc/LLVM seem to insert a call to `llvm.minnum` there instead...~~ EDIT: Rust's LLVM version is too old :/

Author: bors

Diff for: src/librustc_llvm/ffi.rs

@@ -1248,6 +1248,9 @@ extern "C" {
                                          IsNaN: bool)
                                          -> ValueRef;
 
+    pub fn LLVMRustBuildMinNum(B: BuilderRef, LHS: ValueRef, LHS: ValueRef) -> ValueRef;
+    pub fn LLVMRustBuildMaxNum(B: BuilderRef, LHS: ValueRef, LHS: ValueRef) -> ValueRef;
+
     pub fn LLVMBuildIsNull(B: BuilderRef, Val: ValueRef, Name: *const c_char) -> ValueRef;
     pub fn LLVMBuildIsNotNull(B: BuilderRef, Val: ValueRef, Name: *const c_char) -> ValueRef;
     pub fn LLVMBuildPtrDiff(B: BuilderRef,

Diff for: src/librustc_trans/builder.rs

@@ -917,6 +917,27 @@ impl<'a, 'tcx> Builder<'a, 'tcx> {
         }
     }
 
+    pub fn minnum(&self, lhs: ValueRef, rhs: ValueRef) -> ValueRef {
+        self.count_insn("minnum");
+        unsafe {
+            let instr = llvm::LLVMRustBuildMinNum(self.llbuilder, lhs, rhs);
+            if instr.is_null() {
+                bug!("LLVMRustBuildMinNum is not available in LLVM version < 6.0");
+            }
+            instr
+        }
+    }
+    pub fn maxnum(&self, lhs: ValueRef, rhs: ValueRef) -> ValueRef {
+        self.count_insn("maxnum");
+        unsafe {
+            let instr = llvm::LLVMRustBuildMaxNum(self.llbuilder, lhs, rhs);
+            if instr.is_null() {
+                bug!("LLVMRustBuildMaxNum is not available in LLVM version < 6.0");
+            }
+            instr
+        }
+    }
+
     pub fn select(&self, cond: ValueRef, then_val: ValueRef, else_val: ValueRef) -> ValueRef {
         self.count_insn("select");
         unsafe {

Diff for: src/librustc_trans/intrinsic.rs

@@ -1432,6 +1432,8 @@ unsupported {} from `{}` with element `{}` of size `{}` to `{}`"#,
         simd_and: TyUint, TyInt => and;
         simd_or: TyUint, TyInt => or;
         simd_xor: TyUint, TyInt => xor;
+        simd_fmax: TyFloat => maxnum;
+        simd_fmin: TyFloat => minnum;
     }
     span_bug!(span, "unknown SIMD intrinsic");
 }

Diff for: src/librustc_typeck/check/intrinsic.rs

@@ -355,7 +355,8 @@ pub fn check_platform_intrinsic_type<'a, 'tcx>(tcx: TyCtxt<'a, 'tcx, 'tcx>,
         }
         "simd_add" | "simd_sub" | "simd_mul" | "simd_rem" |
         "simd_div" | "simd_shl" | "simd_shr" |
-        "simd_and" | "simd_or" | "simd_xor" => {
+        "simd_and" | "simd_or" | "simd_xor" |
+        "simd_fmin" | "simd_fmax" => {
             (1, vec![param(0), param(0)], param(0))
         }
         "simd_insert" => (2, vec![param(0), tcx.types.u32, param(1)], param(0)),

Diff for: src/rustllvm/RustWrapper.cpp

@@ -1503,3 +1503,23 @@ LLVMBuildExactUDiv(LLVMBuilderRef B, LLVMValueRef LHS,
   return wrap(unwrap(B)->CreateExactUDiv(unwrap(LHS), unwrap(RHS), Name));
 }
 #endif
+
+#if LLVM_VERSION_GE(6, 0)
+extern "C" LLVMValueRef
+LLVMRustBuildMinNum(LLVMBuilderRef B, LLVMValueRef LHS, LLVMValueRef RHS) {
+    return wrap(unwrap(B)->CreateMinNum(unwrap(LHS),unwrap(RHS)));
+}
+extern "C" LLVMValueRef
+LLVMRustBuildMaxNum(LLVMBuilderRef B, LLVMValueRef LHS, LLVMValueRef RHS) {
+    return wrap(unwrap(B)->CreateMaxNum(unwrap(LHS),unwrap(RHS)));
+}
+#else
+extern "C" LLVMValueRef
+LLVMRustBuildMinNum(LLVMBuilderRef, LLVMValueRef, LLVMValueRef) {
+   return nullptr;
+}
+extern "C" LLVMValueRef
+LLVMRustBuildMaxNum(LLVMBuilderRef, LLVMValueRef, LLVMValueRef) {
+   return nullptr;
+}
+#endif

Diff for: src/test/codegen/simd-intrinsic-float-minmax.rs

@@ -0,0 +1,43 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+// min-llvm-version 6.0
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#[allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fmin<T>(x: T, y: T) -> T;
+    fn simd_fmax<T>(x: T, y: T) -> T;
+}
+
+// CHECK-LABEL: @fmin
+#[no_mangle]
+pub unsafe fn fmin(a: f32x4, b: f32x4) -> f32x4 {
+    // CHECK: call <4 x float> @llvm.minnum.v4f32
+    simd_fmin(a, b)
+}
+
+// FIXME(49261)
+// // C_HECK-LABEL: @fmax
+// #[no_mangle]
+// pub unsafe fn fmax(a: f32x4, b: f32x4) -> f32x4 {
+// // C_HECK: call <4 x float> @llvm.maxnum.v4f32
+//     simd_fmax(a, b)
+// }

Diff for: src/test/run-fail/simd-intrinsic-float-minmax.rs

@@ -0,0 +1,57 @@
+// Copyright 2015 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+// min-llvm-version 6.0
+// error-pattern: panicked
+
+// Test that the simd_f{min,max} intrinsics produce the correct results.
+
+#![feature(repr_simd, platform_intrinsics)]
+#[allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fmin<T>(x: T, y: T) -> T;
+    fn simd_fmax<T>(x: T, y: T) -> T;
+}
+
+fn main() {
+    let x = f32x4(1.0, 2.0, 3.0, 4.0);
+    let y = f32x4(2.0, 1.0, 4.0, 3.0);
+    let nan = ::std::f32::NAN;
+    let n = f32x4(nan, nan, nan, nan);
+
+    unsafe {
+        let min0 = simd_fmin(x, y);
+        let min1 = simd_fmin(y, x);
+        assert_eq!(min0, min1);
+        let e = f32x4(1.0, 1.0, 3.0, 3.0);
+        assert_eq!(min0, e);
+        let minn = simd_fmin(x, n);
+        assert_eq!(minn, x);
+        let minn = simd_fmin(y, n);
+        assert_eq!(minn, y);
+
+        // FIXME(49261)
+        let max0 = simd_fmax(x, y);
+        let max1 = simd_fmax(y, x);
+        assert_eq!(max0, max1);
+        let e = f32x4(2.0, 2.0, 4.0, 4.0);
+        assert_eq!(max0, e);
+        let maxn = simd_fmax(x, n);
+        assert_eq!(maxn, x);
+        let maxn = simd_fmax(y, n);
+        assert_eq!(maxn, y);
+    }
+}