about summary refs log tree commit diff
diff options
context:
space:
mode:
authorRalf Jung <post@ralfj.de>2024-02-18 09:36:36 +0100
committerRalf Jung <post@ralfj.de>2024-02-21 20:50:47 +0100
commit07b62409478d810b723bf9bf2a471e78c8421d0d (patch)
tree8a71f9f7d11681c12c087e6aa6e425a6354191c2
parent3dc631a61a7333123ece4f21b0d44e43cdff0980 (diff)
downloadrust-07b62409478d810b723bf9bf2a471e78c8421d0d.tar.gz
rust-07b62409478d810b723bf9bf2a471e78c8421d0d.zip
remove simd_reduce_{min,max}_nanless
-rw-r--r--compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs4
-rw-r--r--compiler/rustc_codegen_gcc/src/intrinsic/simd.rs3
-rw-r--r--compiler/rustc_codegen_llvm/src/builder.rs16
-rw-r--r--compiler/rustc_codegen_llvm/src/intrinsic.rs3
-rw-r--r--compiler/rustc_hir_analysis/src/check/intrinsic.rs4
-rw-r--r--compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp13
-rw-r--r--compiler/rustc_span/src/symbol.rs2
-rw-r--r--library/core/src/intrinsics/simd.rs26
-rw-r--r--tests/codegen/simd/issue-120720-reduce-nan.rs2
-rw-r--r--tests/ui/simd/intrinsic/generic-reduction-pass.rs6
10 files changed, 10 insertions, 69 deletions
diff --git a/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs b/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs
index ebdc744bcd8..8f662808522 100644
--- a/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs
+++ b/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs
@@ -743,7 +743,7 @@ pub(super) fn codegen_simd_intrinsic_call<'tcx>(
             simd_reduce(fx, v, None, ret, &|fx, _ty, a, b| fx.bcx.ins().bxor(a, b));
         }
 
-        sym::simd_reduce_min | sym::simd_reduce_min_nanless => {
+        sym::simd_reduce_min => {
             intrinsic_args!(fx, args => (v); intrinsic);
 
             if !v.layout().ty.is_simd() {
@@ -762,7 +762,7 @@ pub(super) fn codegen_simd_intrinsic_call<'tcx>(
             });
         }
 
-        sym::simd_reduce_max | sym::simd_reduce_max_nanless => {
+        sym::simd_reduce_max => {
             intrinsic_args!(fx, args => (v); intrinsic);
 
             if !v.layout().ty.is_simd() {
diff --git a/compiler/rustc_codegen_gcc/src/intrinsic/simd.rs b/compiler/rustc_codegen_gcc/src/intrinsic/simd.rs
index dedd4653858..d8091724d86 100644
--- a/compiler/rustc_codegen_gcc/src/intrinsic/simd.rs
+++ b/compiler/rustc_codegen_gcc/src/intrinsic/simd.rs
@@ -1041,9 +1041,6 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
 
     minmax_red!(simd_reduce_min: vector_reduce_min, vector_reduce_fmin);
     minmax_red!(simd_reduce_max: vector_reduce_max, vector_reduce_fmax);
-    // TODO(sadlerap): revisit these intrinsics to generate more optimal reductions
-    minmax_red!(simd_reduce_min_nanless: vector_reduce_min, vector_reduce_fmin);
-    minmax_red!(simd_reduce_max_nanless: vector_reduce_max, vector_reduce_fmax);
 
     macro_rules! bitwise_red {
         ($name:ident : $op:expr, $boolean:expr) => {
diff --git a/compiler/rustc_codegen_llvm/src/builder.rs b/compiler/rustc_codegen_llvm/src/builder.rs
index ed73c9f22e7..8cab2a3f27c 100644
--- a/compiler/rustc_codegen_llvm/src/builder.rs
+++ b/compiler/rustc_codegen_llvm/src/builder.rs
@@ -1406,22 +1406,6 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
             llvm::LLVMRustBuildVectorReduceFMax(self.llbuilder, src, /*NoNaNs:*/ false)
         }
     }
-    pub fn vector_reduce_fmin_fast(&mut self, src: &'ll Value) -> &'ll Value {
-        unsafe {
-            let instr =
-                llvm::LLVMRustBuildVectorReduceFMin(self.llbuilder, src, /*NoNaNs:*/ true);
-            llvm::LLVMRustSetFastMath(instr);
-            instr
-        }
-    }
-    pub fn vector_reduce_fmax_fast(&mut self, src: &'ll Value) -> &'ll Value {
-        unsafe {
-            let instr =
-                llvm::LLVMRustBuildVectorReduceFMax(self.llbuilder, src, /*NoNaNs:*/ true);
-            llvm::LLVMRustSetFastMath(instr);
-            instr
-        }
-    }
     pub fn vector_reduce_min(&mut self, src: &'ll Value, is_signed: bool) -> &'ll Value {
         unsafe { llvm::LLVMRustBuildVectorReduceMin(self.llbuilder, src, is_signed) }
     }
diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs
index b7782410012..23e6f054a7c 100644
--- a/compiler/rustc_codegen_llvm/src/intrinsic.rs
+++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs
@@ -1920,9 +1920,6 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
     minmax_red!(simd_reduce_min: vector_reduce_min, vector_reduce_fmin);
     minmax_red!(simd_reduce_max: vector_reduce_max, vector_reduce_fmax);
 
-    minmax_red!(simd_reduce_min_nanless: vector_reduce_min, vector_reduce_fmin_fast);
-    minmax_red!(simd_reduce_max_nanless: vector_reduce_max, vector_reduce_fmax_fast);
-
     macro_rules! bitwise_red {
         ($name:ident : $red:ident, $boolean:expr) => {
             if name == sym::$name {
diff --git a/compiler/rustc_hir_analysis/src/check/intrinsic.rs b/compiler/rustc_hir_analysis/src/check/intrinsic.rs
index 05fab60fd8d..d7277b22c84 100644
--- a/compiler/rustc_hir_analysis/src/check/intrinsic.rs
+++ b/compiler/rustc_hir_analysis/src/check/intrinsic.rs
@@ -606,9 +606,7 @@ pub fn check_platform_intrinsic_type(
         | sym::simd_reduce_or
         | sym::simd_reduce_xor
         | sym::simd_reduce_min
-        | sym::simd_reduce_max
-        | sym::simd_reduce_min_nanless
-        | sym::simd_reduce_max_nanless => (2, 0, vec![param(0)], param(1)),
+        | sym::simd_reduce_max => (2, 0, vec![param(0)], param(1)),
         sym::simd_shuffle => (3, 0, vec![param(0), param(0), param(1)], param(2)),
         sym::simd_shuffle_generic => (2, 1, vec![param(0), param(0)], param(1)),
         _ => {
diff --git a/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp b/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp
index 664d2adbdc1..af2353fbb19 100644
--- a/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp
+++ b/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp
@@ -418,17 +418,11 @@ extern "C" LLVMAttributeRef LLVMRustCreateMemoryEffectsAttr(LLVMContextRef C,
   }
 }
 
-<<<<<<< HEAD
 // Enable all fast-math flags, including those which will cause floating-point operations
 // to return poison for some well-defined inputs. This function can only be used to build
 // unsafe Rust intrinsics. That unsafety does permit additional optimizations, but at the
 // time of writing, their value is not well-understood relative to those enabled by
 // LLVMRustSetAlgebraicMath.
-||||||| parent of 019019d83e2 (make simd_reduce_{mul,add}_unordered use only the 'reassoc' flag, not all fast-math flags)
-// Enable a fast-math flag
-=======
-// Enable all fast-math flags
->>>>>>> 019019d83e2 (make simd_reduce_{mul,add}_unordered use only the 'reassoc' flag, not all fast-math flags)
 //
 // https://llvm.org/docs/LangRef.html#fast-math-flags
 extern "C" void LLVMRustSetFastMath(LLVMValueRef V) {
@@ -456,7 +450,12 @@ extern "C" void LLVMRustSetAlgebraicMath(LLVMValueRef V) {
   }
 }
 
-// Enable the reassoc fast-math flag
+// Enable the reassoc fast-math flag, allowing transformations that pretend
+// floating-point addition and multiplication are associative.
+//
+// Note that this does NOT enable any flags which can cause a floating-point operation on
+// well-defined inputs to return poison, and therefore this function can be used to build
+// safe Rust intrinsics (such as fadd_algebraic).
 //
 // https://llvm.org/docs/LangRef.html#fast-math-flags
 extern "C" void LLVMRustSetAllowReassoc(LLVMValueRef V) {
diff --git a/compiler/rustc_span/src/symbol.rs b/compiler/rustc_span/src/symbol.rs
index 181ab0d4d56..46472a131ff 100644
--- a/compiler/rustc_span/src/symbol.rs
+++ b/compiler/rustc_span/src/symbol.rs
@@ -1553,9 +1553,7 @@ symbols! {
         simd_reduce_and,
         simd_reduce_any,
         simd_reduce_max,
-        simd_reduce_max_nanless,
         simd_reduce_min,
-        simd_reduce_min_nanless,
         simd_reduce_mul_ordered,
         simd_reduce_mul_unordered,
         simd_reduce_or,
diff --git a/library/core/src/intrinsics/simd.rs b/library/core/src/intrinsics/simd.rs
index 2be0c542f1c..d8a1ad778d7 100644
--- a/library/core/src/intrinsics/simd.rs
+++ b/library/core/src/intrinsics/simd.rs
@@ -385,19 +385,6 @@ extern "platform-intrinsic" {
     /// For floating-point values, uses IEEE-754 `maxNum`.
     pub fn simd_reduce_max<T, U>(x: T) -> U;
 
-    /// Return the maximum element of a vector.
-    ///
-    /// `T` must be a vector of integer or floating-point primitive types.
-    ///
-    /// `U` must be the element type of `T`.
-    ///
-    /// For floating-point values, uses IEEE-754 `maxNum`.
-    ///
-    /// # Safety
-    ///
-    /// All input elements must be finite (i.e., not NAN and not +/- INF).
-    pub fn simd_reduce_max_nanless<T, U>(x: T) -> U;
-
     /// Return the minimum element of a vector.
     ///
     /// `T` must be a vector of integer or floating-point primitive types.
@@ -407,19 +394,6 @@ extern "platform-intrinsic" {
     /// For floating-point values, uses IEEE-754 `minNum`.
     pub fn simd_reduce_min<T, U>(x: T) -> U;
 
-    /// Return the minimum element of a vector.
-    ///
-    /// `T` must be a vector of integer or floating-point primitive types.
-    ///
-    /// `U` must be the element type of `T`.
-    ///
-    /// For floating-point values, uses IEEE-754 `minNum`.
-    ///
-    /// # Safety
-    ///
-    /// All input elements must be finite (i.e., not NAN and not +/- INF).
-    pub fn simd_reduce_min_nanless<T, U>(x: T) -> U;
-
     /// Logical "and" all elements together.
     ///
     /// `T` must be a vector of integer or floating-point primitive types.
diff --git a/tests/codegen/simd/issue-120720-reduce-nan.rs b/tests/codegen/simd/issue-120720-reduce-nan.rs
index 233131aa01c..c2ffca4bd87 100644
--- a/tests/codegen/simd/issue-120720-reduce-nan.rs
+++ b/tests/codegen/simd/issue-120720-reduce-nan.rs
@@ -12,7 +12,7 @@ use std::arch::x86_64::*;
 #[no_mangle]
 #[target_feature(enable = "avx512f")] // Function-level target feature mismatches inhibit inlining
 pub unsafe fn demo() -> bool {
-    // CHECK: %0 = tail call reassoc nsz arcp contract double @llvm.vector.reduce.fadd.v8f64(
+    // CHECK: %0 = tail call reassoc double @llvm.vector.reduce.fadd.v8f64(
     // CHECK: %_0.i = fcmp uno double %0, 0.000000e+00
     // CHECK: ret i1 %_0.i
     let res = unsafe {
diff --git a/tests/ui/simd/intrinsic/generic-reduction-pass.rs b/tests/ui/simd/intrinsic/generic-reduction-pass.rs
index 4928ea7bac7..cf4669cd61f 100644
--- a/tests/ui/simd/intrinsic/generic-reduction-pass.rs
+++ b/tests/ui/simd/intrinsic/generic-reduction-pass.rs
@@ -31,8 +31,6 @@ extern "platform-intrinsic" {
     fn simd_reduce_mul_ordered<T, U>(x: T, acc: U) -> U;
     fn simd_reduce_min<T, U>(x: T) -> U;
     fn simd_reduce_max<T, U>(x: T) -> U;
-    fn simd_reduce_min_nanless<T, U>(x: T) -> U;
-    fn simd_reduce_max_nanless<T, U>(x: T) -> U;
     fn simd_reduce_and<T, U>(x: T) -> U;
     fn simd_reduce_or<T, U>(x: T) -> U;
     fn simd_reduce_xor<T, U>(x: T) -> U;
@@ -127,10 +125,6 @@ fn main() {
         assert_eq!(r, -2_f32);
         let r: f32 = simd_reduce_max(x);
         assert_eq!(r, 4_f32);
-        let r: f32 = simd_reduce_min_nanless(x);
-        assert_eq!(r, -2_f32);
-        let r: f32 = simd_reduce_max_nanless(x);
-        assert_eq!(r, 4_f32);
     }
 
     unsafe {