git: 9ba62b3b05d5 - stable/12 - Reapply r230021, r276851 and a few other commits to compiler-rt

From: Dimitry Andric <dim_at_FreeBSD.org>
Date: Wed, 22 Dec 2021 10:05:26 UTC
The branch stable/12 has been updated by dim:

URL: https://cgit.FreeBSD.org/src/commit/?id=9ba62b3b05d56268c1437995317a95af46689a07

commit 9ba62b3b05d56268c1437995317a95af46689a07
Author:     Dimitry Andric <dim@FreeBSD.org>
AuthorDate: 2020-08-02 18:07:16 +0000
Commit:     Dimitry Andric <dim@FreeBSD.org>
CommitDate: 2021-12-22 09:58:10 +0000

    Reapply r230021, r276851 and a few other commits to compiler-rt
    
    Reapply r230021 (by ed):
    
    Add a workaround to prevent endless recursion in compiler-rt.
    
    SPARC and MIPS CPUs don't have special instructions to count
    leading/trailing zeroes. The compiler-rt library provides fallback
    rountines for these. The 64-bit routines, __clzdi2 and __ctzdi2, are
    implemented as simple wrappers around the compiler built-in
    __builtin_clz(), assuming these will expand to either 32-bit
    CPU instructions or calls to __clzsi2 and __ctzsi2.
    
    Unfortunately, our GCC 4.2 probably thinks that because the operand is
    stored in a 64-bit register, it might just be a better idea to invoke
    its 64-bit equivalent, simply resulting into endless recursion. Fix this
    by defining __builtin_clz and __builtin_ctz to __clzsi2 and __ctzsi2
    explicitly.
    
    Reapply r276851:
    
    Update compiler-rt to trunk r224034.  This brings a number of new
    builtins, and also the various sanitizers.  Support for these will be
    added in a later commit.
    
    (cherry picked from commit 71daeec70a15871cfc0071321cd1e819399aa911)
---
 .../compiler-rt/lib/builtins/int_lib.h             | 27 +++++++++++++++++++++-
 1 file changed, 26 insertions(+), 1 deletion(-)

diff --git a/contrib/llvm-project/compiler-rt/lib/builtins/int_lib.h b/contrib/llvm-project/compiler-rt/lib/builtins/int_lib.h
index 991c4a99ea6e..713deddb5698 100644
--- a/contrib/llvm-project/compiler-rt/lib/builtins/int_lib.h
+++ b/contrib/llvm-project/compiler-rt/lib/builtins/int_lib.h
@@ -21,7 +21,9 @@
 // ABI macro definitions
 
 #if __ARM_EABI__
-#ifdef COMPILER_RT_ARMHF_TARGET
+#if defined(COMPILER_RT_ARMHF_TARGET) || (!defined(__clang__) && \
+    defined(__GNUC__) && (__GNUC__ < 4 || __GNUC__ == 4 && __GNUC_MINOR__ < 5))
+// The pcs attribute was introduced in GCC 4.5.0
 #define COMPILER_RT_ABI
 #else
 #define COMPILER_RT_ABI __attribute__((__pcs__("aapcs")))
@@ -92,6 +94,29 @@
 // Include internal utility function declarations.
 #include "int_util.h"
 
+/*
+ * Workaround for LLVM bug 11663.  Prevent endless recursion in
+ * __c?zdi2(), where calls to __builtin_c?z() are expanded to
+ * __c?zdi2() instead of __c?zsi2().
+ *
+ * Instead of placing this workaround in c?zdi2.c, put it in this
+ * global header to prevent other C files from making the detour
+ * through __c?zdi2() as well.
+ *
+ * This problem has been observed on FreeBSD for sparc64 and
+ * mips64 with GCC 4.2.1, and for riscv with GCC 5.2.0.
+ * Presumably it's any version of GCC, and targeting an arch that
+ * does not have dedicated bit counting instructions.
+ */
+#if defined(__FreeBSD__) && (defined(__sparc64__) || \
+    defined(__mips_n32) || defined(__mips_n64) || defined(__mips_o64) || \
+    defined(__riscv))
+si_int __clzsi2(si_int);
+si_int __ctzsi2(si_int);
+#define	__builtin_clz __clzsi2
+#define	__builtin_ctz __ctzsi2
+#endif /* FreeBSD && (sparc64 || mips_n32 || mips_n64 || mips_o64 || riscv) */
+
 COMPILER_RT_ABI int __paritysi2(si_int a);
 COMPILER_RT_ABI int __paritydi2(di_int a);