Unbreak *tf builtins for hexfloat (#82208)

This re-lands cc0065a7d082f0bd322a538cf62cfaef1c8f89f8 in a way that
keeps existing targets working.

---------

Original commit message:
#68132 ended up removing
__multc3 & __divtc3 from compiler-rt library builds that have
QUAD_PRECISION but not TF_MODE due to missing int128 support.
I added support for QUAD_PRECISION to use the native hex float long double representation.

---------

Co-authored-by: Sean Perry <perry@ca.ibm.com>
(cherry picked from commit 99c457dc2ef395872d7448c85609f6cb73a7f89b)

4 files changed, 34 insertions, 19 deletions

diff --git a/compiler-rt/lib/builtins/divtc3.c b/compiler-rt/lib/builtins/divtc3.c
index e970cef..099de58 100644
--- a/compiler-rt/lib/builtins/divtc3.c
+++ b/compiler-rt/lib/builtins/divtc3.c
@@ -13,7 +13,7 @@
 #define QUAD_PRECISION
 #include "fp_lib.h"
 
-#if defined(CRT_HAS_TF_MODE)
+#if defined(CRT_HAS_F128)
 
 // Returns: the quotient of (a + ib) / (c + id)
 
diff --git a/compiler-rt/lib/builtins/fp_lib.h b/compiler-rt/lib/builtins/fp_lib.h
index af406e7..c4f0a5b 100644
--- a/compiler-rt/lib/builtins/fp_lib.h
+++ b/compiler-rt/lib/builtins/fp_lib.h
@@ -22,6 +22,7 @@
 
 #include "int_lib.h"
 #include "int_math.h"
+#include "int_types.h"
 #include <limits.h>
 #include <stdbool.h>
 #include <stdint.h>
@@ -93,13 +94,14 @@ static __inline void wideMultiply(rep_t a, rep_t b, rep_t *hi, rep_t *lo) {
 COMPILER_RT_ABI fp_t __adddf3(fp_t a, fp_t b);
 
 #elif defined QUAD_PRECISION
-#if defined(CRT_HAS_TF_MODE)
+#if defined(CRT_HAS_F128) && defined(CRT_HAS_128BIT)
 typedef uint64_t half_rep_t;
 typedef __uint128_t rep_t;
 typedef __int128_t srep_t;
 typedef tf_float fp_t;
 #define HALF_REP_C UINT64_C
 #define REP_C (__uint128_t)
+#if defined(CRT_HAS_IEEE_TF)
 // Note: Since there is no explicit way to tell compiler the constant is a
 // 128-bit integer, we let the constant be casted to 128-bit integer
 #define significandBits 112
@@ -188,7 +190,10 @@ static __inline void wideMultiply(rep_t a, rep_t b, rep_t *hi, rep_t *lo) {
 #undef Word_HiMask
 #undef Word_LoMask
 #undef Word_FullMask
-#endif // defined(CRT_HAS_TF_MODE)
+#endif // defined(CRT_HAS_IEEE_TF)
+#else
+typedef long double fp_t;
+#endif // defined(CRT_HAS_F128) && defined(CRT_HAS_128BIT)
 #else
 #error SINGLE_PRECISION, DOUBLE_PRECISION or QUAD_PRECISION must be defined.
 #endif
@@ -196,19 +201,6 @@ static __inline void wideMultiply(rep_t a, rep_t b, rep_t *hi, rep_t *lo) {
 #if defined(SINGLE_PRECISION) || defined(DOUBLE_PRECISION) ||                  \
     (defined(QUAD_PRECISION) && defined(CRT_HAS_TF_MODE))
 #define typeWidth (sizeof(rep_t) * CHAR_BIT)
-#define exponentBits (typeWidth - significandBits - 1)
-#define maxExponent ((1 << exponentBits) - 1)
-#define exponentBias (maxExponent >> 1)
-
-#define implicitBit (REP_C(1) << significandBits)
-#define significandMask (implicitBit - 1U)
-#define signBit (REP_C(1) << (significandBits + exponentBits))
-#define absMask (signBit - 1U)
-#define exponentMask (absMask ^ significandMask)
-#define oneRep ((rep_t)exponentBias << significandBits)
-#define infRep exponentMask
-#define quietBit (implicitBit >> 1)
-#define qnanRep (exponentMask | quietBit)
 
 static __inline rep_t toRep(fp_t x) {
   const union {
@@ -226,6 +218,21 @@ static __inline fp_t fromRep(rep_t x) {
   return rep.f;
 }
 
+#if !defined(QUAD_PRECISION) || defined(CRT_HAS_IEEE_TF)
+#define exponentBits (typeWidth - significandBits - 1)
+#define maxExponent ((1 << exponentBits) - 1)
+#define exponentBias (maxExponent >> 1)
+
+#define implicitBit (REP_C(1) << significandBits)
+#define significandMask (implicitBit - 1U)
+#define signBit (REP_C(1) << (significandBits + exponentBits))
+#define absMask (signBit - 1U)
+#define exponentMask (absMask ^ significandMask)
+#define oneRep ((rep_t)exponentBias << significandBits)
+#define infRep exponentMask
+#define quietBit (implicitBit >> 1)
+#define qnanRep (exponentMask | quietBit)
+
 static __inline int normalize(rep_t *significand) {
   const int shift = rep_clz(*significand) - rep_clz(implicitBit);
   *significand <<= shift;
@@ -328,6 +335,8 @@ static __inline fp_t __compiler_rt_scalbnX(fp_t x, int y) {
     return fromRep(sign | ((rep_t)exp << significandBits) | sig);
 }
 
+#endif // !defined(QUAD_PRECISION) || defined(CRT_HAS_IEEE_TF)
+
 // Avoid using fmax from libm.
 static __inline fp_t __compiler_rt_fmaxX(fp_t x, fp_t y) {
   // If either argument is NaN, return the other argument. If both are NaN,
@@ -405,6 +414,8 @@ static __inline tf_float __compiler_rt_fmaxtf(tf_float x, tf_float y) {
 #define __compiler_rt_logbl crt_logbl
 #define __compiler_rt_scalbnl crt_scalbnl
 #define __compiler_rt_fmaxl crt_fmaxl
+#define crt_fabstf crt_fabsl
+#define crt_copysigntf crt_copysignl
 #else
 #error Unsupported TF mode type
 #endif
diff --git a/compiler-rt/lib/builtins/int_types.h b/compiler-rt/lib/builtins/int_types.h
index 7624c728..ca97391 100644
--- a/compiler-rt/lib/builtins/int_types.h
+++ b/compiler-rt/lib/builtins/int_types.h
@@ -189,12 +189,16 @@ typedef long double tf_float;
 #define CRT_LDBL_IEEE_F128
 #endif
 #define TF_C(x) x##L
-#elif __LDBL_MANT_DIG__ == 113
-// Use long double instead of __float128 if it matches the IEEE 128-bit format.
+#elif __LDBL_MANT_DIG__ == 113 ||                                              \
+    (__FLT_RADIX__ == 16 && __LDBL_MANT_DIG__ == 28)
+// Use long double instead of __float128 if it matches the IEEE 128-bit format
+// or the IBM hexadecimal format.
 #define CRT_LDBL_128BIT
 #define CRT_HAS_F128
+#if __LDBL_MANT_DIG__ == 113
 #define CRT_HAS_IEEE_TF
 #define CRT_LDBL_IEEE_F128
+#endif
 typedef long double tf_float;
 #define TF_C(x) x##L
 #elif defined(__FLOAT128__) || defined(__SIZEOF_FLOAT128__)
diff --git a/compiler-rt/lib/builtins/multc3.c b/compiler-rt/lib/builtins/multc3.c
index f20e53c..61a3f45 100644
--- a/compiler-rt/lib/builtins/multc3.c
+++ b/compiler-rt/lib/builtins/multc3.c
@@ -15,7 +15,7 @@
 #include "int_lib.h"
 #include "int_math.h"
 
-#if defined(CRT_HAS_TF_MODE)
+#if defined(CRT_HAS_F128)
 
 // Returns: the product of a + ib and c + id