Rollup merge of #142176 - workingjubilee:dont-shuffle-bswaps-per-arch, r=nikic

tests: Split dont-shuffle-bswaps along opt-levels and arches

This duplicates dont-shuffle-bswaps in order to make each opt level its own test. Then -opt3.rs gets split into a revision per arch we want to test, with certain architectures gaining new target-cpu minimums.

2 files changed, 45 insertions, 16 deletions

diff --git a/tests/codegen/autovec/dont-shuffle-bswaps-opt2.rs b/tests/codegen/autovec/dont-shuffle-bswaps-opt2.rs
new file mode 100644
index 00000000000..c354228acc5
--- /dev/null
+++ b/tests/codegen/autovec/dont-shuffle-bswaps-opt2.rs
@@ -0,0 +1,31 @@
+//@ compile-flags: -Copt-level=2
+
+#![crate_type = "lib"]
+#![no_std]
+
+// This test is paired with the arch-specific -opt3.rs test.
+
+// The code is from https://github.com/rust-lang/rust/issues/122805.
+// Ensure we do not generate the shufflevector instruction
+// to avoid complicating the code.
+
+// CHECK-LABEL: define{{.*}}void @convert(
+// CHECK-NOT: shufflevector
+#[no_mangle]
+pub fn convert(value: [u16; 8]) -> [u8; 16] {
+    #[cfg(target_endian = "little")]
+    let bswap = u16::to_be;
+    #[cfg(target_endian = "big")]
+    let bswap = u16::to_le;
+    let addr16 = [
+        bswap(value[0]),
+        bswap(value[1]),
+        bswap(value[2]),
+        bswap(value[3]),
+        bswap(value[4]),
+        bswap(value[5]),
+        bswap(value[6]),
+        bswap(value[7]),
+    ];
+    unsafe { core::mem::transmute::<_, [u8; 16]>(addr16) }
+}
diff --git a/tests/codegen/dont-shuffle-bswaps.rs b/tests/codegen/autovec/dont-shuffle-bswaps-opt3.rs
index c1dab2bc295..203d12005de 100644
--- a/tests/codegen/dont-shuffle-bswaps.rs
+++ b/tests/codegen/autovec/dont-shuffle-bswaps-opt3.rs
@@ -1,29 +1,27 @@
-//@ revisions: OPT2 OPT3 OPT3_S390X
-//@[OPT2] compile-flags: -Copt-level=2
-//@[OPT3] compile-flags: -C opt-level=3
-// some targets don't do the opt we are looking for
-//@[OPT3] only-64bit
-//@[OPT3] ignore-s390x
-//@[OPT3_S390X] compile-flags: -C opt-level=3 -C target-cpu=z13
-//@[OPT3_S390X] only-s390x
+//@ revisions: AARCH64 X86_64 Z13
+//@ compile-flags: -Copt-level=3
+//@[AARCH64] only-aarch64
+//@[X86_64] only-x86_64
+//@[Z13] only-s390x
+//@[Z13] compile-flags: -Ctarget-cpu=z13
 
 #![crate_type = "lib"]
 #![no_std]
 
+// This test is paired with the arch-neutral -opt2.rs test
+
 // The code is from https://github.com/rust-lang/rust/issues/122805.
 // Ensure we do not generate the shufflevector instruction
 // to avoid complicating the code.
+
 // CHECK-LABEL: define{{.*}}void @convert(
 // CHECK-NOT: shufflevector
+
 // On higher opt levels, this should just be a bswap:
-// OPT3: load <8 x i16>
-// OPT3-NEXT: call <8 x i16> @llvm.bswap
-// OPT3-NEXT: store <8 x i16>
-// OPT3-NEXT: ret void
-// OPT3_S390X: load <8 x i16>
-// OPT3_S390X-NEXT: call <8 x i16> @llvm.bswap
-// OPT3_S390X-NEXT: store <8 x i16>
-// OPT3_S390X-NEXT: ret void
+// CHECK: load <8 x i16>
+// CHECK-NEXT: call <8 x i16> @llvm.bswap
+// CHECK-NEXT: store <8 x i16>
+// CHECK-NEXT: ret void
 #[no_mangle]
 pub fn convert(value: [u16; 8]) -> [u8; 16] {
     #[cfg(target_endian = "little")]