XOP Part 1: fma, hadd, and shifts-rotates

Added testing support for xop. Intel SDE doesn't support XOP, so we need qemu for that.

ci/docker/x86_64-unknown-linux-gnu-emulated/Dockerfile

@@ -6,10 +6,9 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
   make \
   ca-certificates \
   wget \
-  xz-utils
+  xz-utils \
+  qemu-user
 
 RUN wget https://downloadmirror.intel.com/813591/sde-external-9.33.0-2024-01-07-lin.tar.xz
 RUN tar -xJf sde-external-9.33.0-2024-01-07-lin.tar.xz
-ENV CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER="/sde-external-9.33.0-2024-01-07-lin/sde64 \
-            -cpuid-in /checkout/ci/docker/x86_64-unknown-linux-gnu-emulated/cpuid.def \
-            -rtm-mode full -tsx --"
+ENV PATH="$PATH:/sde-external-9.33.0-2024-01-07-lin"

ci/docker/x86_64-unknown-linux-gnu-emulated/cpuid.def

@@ -47,7 +47,7 @@
 0000001e ******** => 00000000 00004010 00000000 00000000 #AMX Tmul
 00000024 ******** => 00000000 00070001 00000000 00000000 #AVX10
 80000000 ******** => 80000008 00000000 00000000 00000000
-80000001 ******** => 00000000 00000000 00200961 2c100000
+80000001 ******** => 00000000 00000000 00200161 2c100000
 80000002 ******** => 00000000 00000000 00000000 00000000
 80000003 ******** => 00000000 00000000 00000000 00000000
 80000004 ******** => 00000000 00000000 00000000 00000000
@@ -57,5 +57,5 @@
 80000008 ******** => 00003028 00000200 00000200 00000000
 
 # This file was copied from intel-sde/misc/cpuid/future/cpuid.def, and modified to
-# use "AuthenticAMD" as the vendor and the support for `XOP`, `SSE4a`, `TBM` and
+# use "AuthenticAMD" as the vendor and the support for `SSE4a`, `TBM` and
 # `AVX512_VP2INTERSECT` was added in the CPUID.

ci/run.sh

@@ -101,6 +101,17 @@ STD_DETECT="--manifest-path=crates/std_detect/Cargo.toml"
 STDARCH_EXAMPLES="--manifest-path=examples/Cargo.toml"
 INTRINSIC_TEST="--manifest-path=crates/intrinsic-test/Cargo.toml"
 
+if [ "$STDARCH_TEST_EVERYTHING" = "1" ]; then
+  # Intel SDE doesn't support xop, so we need qemu (only testing xop to save time)
+  export CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER="qemu-x86_64 -cpu=max"
+  cargo_test "${CORE_ARCH} --release xop"
+
+  export CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER="sde64 \
+      -cpuid-in /checkout/ci/docker/x86_64-unknown-linux-gnu-emulated/cpuid.def \
+      -rtm-mode full -tsx --"
+  export SKIP_TESTS="xop"
+fi
+
 cargo_test "${CORE_ARCH} --release"
 
 if [ "$NOSTD" != "1" ]; then

crates/core_arch/src/lib.rs

@@ -35,7 +35,8 @@
     generic_arg_infer,
     asm_experimental_arch,
     sha512_sm_x86,
-    f16
+    f16,
+    xop_target_feature
 )]
 #![cfg_attr(test, feature(test, abi_vectorcall, stdarch_internal))]
 #![deny(clippy::missing_inline_in_public_items)]

crates/core_arch/src/x86/mod.rs

@@ -437,6 +437,11 @@ pub type _MM_MANTISSA_SIGN_ENUM = i32;
 #[unstable(feature = "stdarch_x86_avx512", issue = "111137")]
 pub type _MM_PERM_ENUM = i32;
 
+/// The `_MM_PCOMCTRL_ENUM` type used to specify comparison control operations in XOP intrinsics.
+#[allow(non_camel_case_types)]
+#[unstable(feature = "xop_target_feature", issue = "127208")]
+pub type _MM_PCOMCTRL_ENUM = i32;
+
 #[cfg(test)]
 mod test;
 #[cfg(test)]
@@ -1008,3 +1013,7 @@ pub use self::avxneconvert::*;
 mod avx512fp16;
 #[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")]
 pub use self::avx512fp16::*;
+
+mod xop;
+#[unstable(feature = "xop_target_feature", issue = "127208")]
+pub use self::xop::*;