Fix i8mm feature with bootstrap compiler. (#1252)

Author: ehuss

crates/core_arch/src/arm_shared/neon/mod.rs

@@ -4808,7 +4808,8 @@ pub unsafe fn vpadalq_u32(a: uint64x2_t, b: uint32x4_t) -> uint64x2_t {
 
 /// 8-bit integer matrix multiply-accumulate
 #[inline]
-#[target_feature(enable = "neon,i8mm")]
+#[cfg_attr(not(bootstrap), target_feature(enable = "i8mm"))]
+#[target_feature(enable = "neon")]
 #[cfg_attr(target_arch = "arm", target_feature(enable = "v8"))]
 #[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
 #[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(smmla))]
@@ -4827,7 +4828,8 @@ pub unsafe fn vmmlaq_s32(a: int32x4_t, b: int8x16_t, c: int8x16_t) -> int32x4_t
 
 /// 8-bit integer matrix multiply-accumulate
 #[inline]
-#[target_feature(enable = "neon,i8mm")]
+#[cfg_attr(not(bootstrap), target_feature(enable = "i8mm"))]
+#[target_feature(enable = "neon")]
 #[cfg_attr(target_arch = "arm", target_feature(enable = "v8"))]
 #[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
 #[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ummla))]
@@ -4846,7 +4848,8 @@ pub unsafe fn vmmlaq_u32(a: uint32x4_t, b: uint8x16_t, c: uint8x16_t) -> uint32x
 
 /// Unsigned and signed 8-bit integer matrix multiply-accumulate
 #[inline]
-#[target_feature(enable = "neon,i8mm")]
+#[cfg_attr(not(bootstrap), target_feature(enable = "i8mm"))]
+#[target_feature(enable = "neon")]
 #[cfg_attr(target_arch = "arm", target_feature(enable = "v8"))]
 #[cfg_attr(all(test, target_arch = "arm"), assert_instr(nop))]
 #[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(usmmla))]