about summary refs log tree commit diff
path: root/compiler/rustc_codegen_llvm/src
diff options
context:
space:
mode:
author许杰友 Jieyou Xu (Joe) <39484203+jieyouxu@users.noreply.github.com>2024-11-30 12:57:32 +0800
committerGitHub <noreply@github.com>2024-11-30 12:57:32 +0800
commit1aa01927d3b8eeec06db1cd72e23b08983aa3104 (patch)
treea0bfe91315b9e8d7d0ce9cb8ce2613b76b18aad2 /compiler/rustc_codegen_llvm/src
parent76f3ff605962d7046bc1537597ceed5e12325f54 (diff)
parentdf8feb5067f99e20059c7ee8021d9ba5273bfe68 (diff)
downloadrust-1aa01927d3b8eeec06db1cd72e23b08983aa3104.tar.gz
rust-1aa01927d3b8eeec06db1cd72e23b08983aa3104.zip
Rollup merge of #131551 - taiki-e:ppc-asm-vreg-inout, r=Amanieu
Support input/output in vector registers of PowerPC inline assembly

This extends currently clobber-only vector registers (`vreg`) support to allow passing `#[repr(simd)]` types as input/output.

| Architecture | Register class | Target feature | Allowed types |
| ------------ | -------------- | -------------- | -------------- |
| PowerPC      | `vreg` | `altivec` | `i8x16`, `i16x8`, `i32x4`, `f32x4` |
| PowerPC      | `vreg` | `vsx` | `f32`, `f64`, `i64x2`, `f64x2` |

In addition to floats and `core::simd` types listed above, `core::arch` types and custom `#[repr(simd)]` types of the same size and type are also allowed. All allowed types and relevant target features are currently unstable.

r? `@Amanieu`

`@rustbot` label +O-PowerPC +A-inline-assembly
Diffstat (limited to 'compiler/rustc_codegen_llvm/src')
-rw-r--r--compiler/rustc_codegen_llvm/src/asm.rs52
1 files changed, 46 insertions, 6 deletions
diff --git a/compiler/rustc_codegen_llvm/src/asm.rs b/compiler/rustc_codegen_llvm/src/asm.rs
index 9aa01bd1b95..d1804cb49ad 100644
--- a/compiler/rustc_codegen_llvm/src/asm.rs
+++ b/compiler/rustc_codegen_llvm/src/asm.rs
@@ -656,9 +656,8 @@ fn reg_to_llvm(reg: InlineAsmRegOrRegClass, layout: Option<&TyAndLayout<'_>>) ->
             PowerPC(PowerPCInlineAsmRegClass::reg) => "r",
             PowerPC(PowerPCInlineAsmRegClass::reg_nonzero) => "b",
             PowerPC(PowerPCInlineAsmRegClass::freg) => "f",
-            PowerPC(PowerPCInlineAsmRegClass::cr)
-            | PowerPC(PowerPCInlineAsmRegClass::xer)
-            | PowerPC(PowerPCInlineAsmRegClass::vreg) => {
+            PowerPC(PowerPCInlineAsmRegClass::vreg) => "v",
+            PowerPC(PowerPCInlineAsmRegClass::cr) | PowerPC(PowerPCInlineAsmRegClass::xer) => {
                 unreachable!("clobber-only")
             }
             RiscV(RiscVInlineAsmRegClass::reg) => "r",
@@ -825,9 +824,8 @@ fn dummy_output_type<'ll>(cx: &CodegenCx<'ll, '_>, reg: InlineAsmRegClass) -> &'
         PowerPC(PowerPCInlineAsmRegClass::reg) => cx.type_i32(),
         PowerPC(PowerPCInlineAsmRegClass::reg_nonzero) => cx.type_i32(),
         PowerPC(PowerPCInlineAsmRegClass::freg) => cx.type_f64(),
-        PowerPC(PowerPCInlineAsmRegClass::cr)
-        | PowerPC(PowerPCInlineAsmRegClass::xer)
-        | PowerPC(PowerPCInlineAsmRegClass::vreg) => {
+        PowerPC(PowerPCInlineAsmRegClass::vreg) => cx.type_vector(cx.type_i32(), 4),
+        PowerPC(PowerPCInlineAsmRegClass::cr) | PowerPC(PowerPCInlineAsmRegClass::xer) => {
             unreachable!("clobber-only")
         }
         RiscV(RiscVInlineAsmRegClass::reg) => cx.type_i32(),
@@ -1042,6 +1040,26 @@ fn llvm_fixup_input<'ll, 'tcx>(
             let value = bx.or(value, bx.const_u32(0xFFFF_0000));
             bx.bitcast(value, bx.type_f32())
         }
+        (PowerPC(PowerPCInlineAsmRegClass::vreg), BackendRepr::Scalar(s))
+            if s.primitive() == Primitive::Float(Float::F32) =>
+        {
+            let value = bx.insert_element(
+                bx.const_undef(bx.type_vector(bx.type_f32(), 4)),
+                value,
+                bx.const_usize(0),
+            );
+            bx.bitcast(value, bx.type_vector(bx.type_f32(), 4))
+        }
+        (PowerPC(PowerPCInlineAsmRegClass::vreg), BackendRepr::Scalar(s))
+            if s.primitive() == Primitive::Float(Float::F64) =>
+        {
+            let value = bx.insert_element(
+                bx.const_undef(bx.type_vector(bx.type_f64(), 2)),
+                value,
+                bx.const_usize(0),
+            );
+            bx.bitcast(value, bx.type_vector(bx.type_f64(), 2))
+        }
         _ => value,
     }
 }
@@ -1177,6 +1195,18 @@ fn llvm_fixup_output<'ll, 'tcx>(
             let value = bx.trunc(value, bx.type_i16());
             bx.bitcast(value, bx.type_f16())
         }
+        (PowerPC(PowerPCInlineAsmRegClass::vreg), BackendRepr::Scalar(s))
+            if s.primitive() == Primitive::Float(Float::F32) =>
+        {
+            let value = bx.bitcast(value, bx.type_vector(bx.type_f32(), 4));
+            bx.extract_element(value, bx.const_usize(0))
+        }
+        (PowerPC(PowerPCInlineAsmRegClass::vreg), BackendRepr::Scalar(s))
+            if s.primitive() == Primitive::Float(Float::F64) =>
+        {
+            let value = bx.bitcast(value, bx.type_vector(bx.type_f64(), 2));
+            bx.extract_element(value, bx.const_usize(0))
+        }
         _ => value,
     }
 }
@@ -1301,6 +1331,16 @@ fn llvm_fixup_output_type<'ll, 'tcx>(
         {
             cx.type_f32()
         }
+        (PowerPC(PowerPCInlineAsmRegClass::vreg), BackendRepr::Scalar(s))
+            if s.primitive() == Primitive::Float(Float::F32) =>
+        {
+            cx.type_vector(cx.type_f32(), 4)
+        }
+        (PowerPC(PowerPCInlineAsmRegClass::vreg), BackendRepr::Scalar(s))
+            if s.primitive() == Primitive::Float(Float::F64) =>
+        {
+            cx.type_vector(cx.type_f64(), 2)
+        }
         _ => layout.llvm_type(cx),
     }
 }