[clang][x86] Add constexpr support for all remaining TBM intrinsics

BEXTR is already handled, so we just need to tag the rest of the intrinsics, which are all expanded to generic patterns. As these are expanded I felt it better to test against the equivalent pattern instead of the final result but I can change this if reviewers prefer (I intend to take the same approach for BMI/BMI2 cases).
llvm · Sep 30, 2024 · dd3eb48 · dd3eb48
1 parent f256222
commit dd3eb48
Show file tree

Hide file tree

Showing 3 changed files with 26 additions and 2 deletions.
diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
@@ -516,7 +516,7 @@ X86 Support
   all bextr intrinsics in bmiintrin.h
   all tzcnt intrinsics in bmiintrin.h
   all bzhi intrinsics in bmi2intrin.h
-  all bextr intrinsics in tbmintrin.h
+  all intrinsics in tbmintrin.h
 
 Arm and AArch64 Support
 ^^^^^^^^^^^^^^^^^^^^^^^

diff --git a/clang/lib/Headers/tbmintrin.h b/clang/lib/Headers/tbmintrin.h
@@ -15,7 +15,13 @@
 #define __TBMINTRIN_H
 
 /* Define the default attributes for the functions in this file. */
-#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("tbm")))
+#if defined(__cplusplus) && (__cplusplus >= 201103L)
+#define __DEFAULT_FN_ATTRS                                                     \
+  __attribute__((__always_inline__, __nodebug__, __target__("tbm"))) constexpr
+#else
+#define __DEFAULT_FN_ATTRS                                                     \
+  __attribute__((__always_inline__, __nodebug__, __target__("tbm")))
+#endif
 
 #define __bextri_u32(a, b) \
   ((unsigned int)__builtin_ia32_bextri_u32((unsigned int)(a), \

diff --git a/clang/test/CodeGen/X86/tbm-builtins.c b/clang/test/CodeGen/X86/tbm-builtins.c
@@ -185,9 +185,27 @@ char bextri32_0[__bextri_u32(0x00000000, 0x00000000) == 0x00000000 ? 1 : -1];
 char bextri32_1[__bextri_u32(0x000003F0, 0xFFFF1004) == 0x0000003F ? 1 : -1];
 char bextri32_2[__bextri_u32(0x000003F0, 0xFFFF3008) == 0x00000003 ? 1 : -1];
 
+char blcfill32[__blcfill_u32(0x89ABCDEF) == (0x89ABCDEF & (0x89ABCDEF + 1)) ? 1 : -1];
+char blci32[__blci_u32(0x89ABCDEF) == (0x89ABCDEF | ~(0x89ABCDEF + 1)) ? 1 : -1];
+char blcmsk32[__blcmsk_u32(0x89ABCDEF) == (0x89ABCDEF ^ (0x89ABCDEF + 1)) ? 1 : -1];
+char blcs32[__blcs_u32(0x89ABCDEF) == (0x89ABCDEF | (0x89ABCDEF + 1)) ? 1 : -1];
+char blsfill32[__blsfill_u32(0x89ABCDEF) == (0x89ABCDEF | (0x89ABCDEF - 1)) ? 1 : -1];
+char blsic32[__blsic_u32(0x89ABCDEF) == (~0x89ABCDEF | (0x89ABCDEF - 1)) ? 1 : -1];
+char t1mskc32[__t1mskc_u32(0x89ABCDEF) == (~0x89ABCDEF | (0x89ABCDEF + 1)) ? 1 : -1];
+char tzmsk32[__tzmsk_u32(0x89ABCDEF) == (~0x89ABCDEF & (0x89ABCDEF - 1)) ? 1 : -1];
+
 #ifdef __x86_64__
 char bextri64_0[__bextri_u64(0x0000000000000000ULL, 0x0000000000000000ULL) == 0x0000000000000000ULL ? 1 : -1];
 char bextri64_1[__bextri_u64(0xF000000000000001ULL, 0x0000000000004001ULL) == 0x7800000000000000ULL ? 1 : -1];
 char bextri64_2[__bextri_u64(0xF000000000000001ULL, 0xFFFFFFFFFFFF1001ULL) == 0x0000000000000000ULL ? 1 : -1];
+
+char blcfill64[__blcfill_u64(0xFEDCBA9876543210) == (0xFEDCBA9876543210 & (0xFEDCBA9876543210 + 1)) ? 1 : -1];
+char blci64[__blci_u64(0xFEDCBA9876543210) == (0xFEDCBA9876543210 | ~(0xFEDCBA9876543210 + 1)) ? 1 : -1];
+char blcmsk64[__blcmsk_u64(0xFEDCBA9876543210) == (0xFEDCBA9876543210 ^ (0xFEDCBA9876543210 + 1)) ? 1 : -1];
+char blcs64[__blcs_u64(0xFEDCBA9876543210) == (0xFEDCBA9876543210 | (0xFEDCBA9876543210 + 1)) ? 1 : -1];
+char blsfill64[__blsfill_u64(0xFEDCBA9876543210) == (0xFEDCBA9876543210 | (0xFEDCBA9876543210 - 1)) ? 1 : -1];
+char blsic64[__blsic_u64(0xFEDCBA9876543210) == (~0xFEDCBA9876543210 | (0xFEDCBA9876543210 - 1)) ? 1 : -1];
+char t1mskc64[__t1mskc_u64(0xFEDCBA9876543210) == (~0xFEDCBA9876543210 | (0xFEDCBA9876543210 + 1)) ? 1 : -1];
+char tzmsk64[__tzmsk_u64(0xFEDCBA9876543210) == (~0xFEDCBA9876543210 & (0xFEDCBA9876543210 - 1)) ? 1 : -1];
 #endif
 #endif