===================================================================
@@ -1274,6 +1274,84 @@ LSYM(Lover12):
#endif
#endif /* L_dvmd_lnx */
+
+#ifdef L_muldi3
+
+/* ------------------------------------------------------------------------ */
+/* Dword multiplication operation.
+
+ The THUMB ISA lacks an instruction to compute the higher half of the
+ 64-bit result from a 32-bit by 32-bit multiplication. This makes 64-bit
+ multiplication difficult to implement efficiently. The ARM ISAs after V3M
+ have UMULL and MLA which can be used to implement 64-bit muliplication
+ efficiently. On a target that support both ARM V3M+ and THUMB ISA's (but
+ not THUMB2), we want to use the ARM version of _muldi3 in the THUMB libgcc.
+
+ We do not need to use the ARM version for THUMB2 targets as the THUMB2
+ targets also support MLA and UMULL. */
+
+/* We cannot use the faster version for following situations:
+
+ -ARM architetures older V3M lack the UMULL instruction.
+ -Target is ARMV6M, which does not run ARM code. */
+
+#undef USE_FAST_MULDI3
+#if (__ARM_ARCH__ > 3 || defined(__ARM_ARCH_3M__)) && !defined(__ARM_ARCH_6M__)
+#define USE_FAST_MULDI3
+#endif
+
+/* Force using ARM code if it is possible except for THUMB2 target. */
+#if defined(USE_FAST_MULDI3) && !defined(__thumb2__)
+ ARM_FUNC_START muldi3
+ ARM_FUNC_ALIAS aeabi_lmul muldi3
+#else
+ FUNC_START muldi3
+ FUNC_ALIAS aeabi_lmul muldi3
+#endif
+
+#if defined(USE_FAST_MULDI3)
+ /* Fast version for ARM with umull and THUMB2. */
+ mul xxh, yyl
+ mla yyh, xxl, yyh, xxh
+ umull xxl, xxh, yyl, xxl
+ add xxh, yyh
+ RET
+#else
+ /* Slow version for both THUMB and older ARMs lacking umull. */
+ mul xxh, yyl /* xxh := AH*BL */
+ push {r4, r5, r6, r7}
+ mul yyh, xxl /* yyh := AL*BH */
+ ldr r4, .L_mask
+ lsr r5, xxl, #16 /* r5 := (AL>>16) */
+ lsr r6, yyl, #16 /* r6 := (BL>>16) */
+ lsr r7, xxl, #16 /* r7 := (AL>>16) */
+ mul r5, r6 /* r5 = (AL>>16) * (BL>>16) */
+ and xxl, r4 /* xxl = AL & 0xffff */
+ and yyl, r4 /* yyl = BL & 0xffff */
+ add xxh, yyh /* xxh = AH*BL+AL*BH */
+ mul r6, xxl /* r6 = (AL&0xffff) * (BL>>16) */
+ mul r7, yyl /* r7 = (AL>>16) * (BL&0xffff) */
+ add xxh, r5
+ mul xxl, yyl /* xxl = (AL&0xffff) * (BL&0xffff) */
+ mov r4, #0
+ adds r6, r7 /* partial sum to result[47:16]. */
+ adc r4, r4 /* carry to result[48]. */
+ lsr yyh, r6, #16
+ lsl r4, r4, #16
+ lsl yyl, r6, #16
+ add xxh, r4
+ adds xxl, yyl
+ adc xxh, yyh
+ pop {r4, r5, r6, r7}
+ RET
+ .align 2
+.L_mask:
+ .word 65535
+#endif
+
+ FUNC_END muldi3
+#endif
+
#ifdef L_clear_cache
#if defined __ARM_EABI__ && defined __linux__
@ EABI GNU/Linux call to cacheflush syscall.
===================================================================
@@ -16,7 +16,8 @@
# along with GCC; see the file COPYING3. If not see
# <http://www.gnu.org/licenses/>.
-LIB1ASMFUNCS += _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_tls _bb_init_func _clzsi2 _clzdi2
+LIB1ASMFUNCS += _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_tls _bb_init_func \
+ _clzsi2 _clzdi2 _muldi3
# We want fine grained libraries, so use the new code to build the
# floating point emulation libraries.
===================================================================
@@ -16,7 +16,8 @@
# along with GCC; see the file COPYING3. If not see
# <http://www.gnu.org/licenses/>.
-LIB1ASMFUNCS += _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_tls _bb_init_func _call_via_rX _interwork_call_via_rX _clzsi2 _clzdi2
+LIB1ASMFUNCS += _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_tls _bb_init_func \
+ _call_via_rX _interwork_call_via_rX _clzsi2 _clzdi2 _muldi3
# We want fine grained libraries, so use the new code to build the
# floating point emulation libraries.
===================================================================
@@ -17,7 +17,7 @@
# along with GCC; see the file COPYING3. If not see
# <http://www.gnu.org/licenses/>.
-LIB1ASMFUNCS += _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_tls _call_via_rX _interwork_call_via_rX _clzsi2 _clzdi2
+LIB1ASMFUNCS += _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_tls _call_via_rX _interwork_call_via_rX _clzsi2 _clzdi2 _muldi3
# We want fine grained libraries, so use the new code to build the
# floating point emulation libraries.
===================================================================
@@ -29,7 +29,7 @@ LIB1ASMFUNCS += _udivsi3 _divsi3 _umodsi3 _modsi3
_arm_truncdfsf2 _arm_negsf2 _arm_addsubsf3 _arm_muldivsf3 \
_arm_cmpsf2 _arm_unordsf2 _arm_fixsfsi _arm_fixunssfsi \
_arm_floatdidf _arm_floatdisf _arm_floatundidf _arm_floatundisf \
- _clzsi2 _clzdi2
+ _clzsi2 _clzdi2 _muldi3
MULTILIB_OPTIONS = marm/mthumb
MULTILIB_DIRNAMES = arm thumb
===================================================================
@@ -23,7 +23,7 @@ TARGET_LIBGCC2_CFLAGS = -fomit-frame-pointer -fPIC
LIB1ASMSRC = arm/lib1funcs.asm
LIB1ASMFUNCS = _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_lnx _clzsi2 _clzdi2 \
- _arm_addsubdf3 _arm_addsubsf3
+ _arm_addsubdf3 _arm_addsubsf3 _muldi3
# MULTILIB_OPTIONS = mhard-float/msoft-float
# MULTILIB_DIRNAMES = hard-float soft-float
===================================================================
@@ -16,7 +16,8 @@
# along with GCC; see the file COPYING3. If not see
# <http://www.gnu.org/licenses/>.
-LIB1ASMFUNCS += _bb_init_func _call_via_rX _interwork_call_via_rX _clzsi2 _clzdi2
+LIB1ASMFUNCS += _bb_init_func _call_via_rX _interwork_call_via_rX _clzsi2 \
+ _clzdi2 _muldi3
# These functions have __aeabi equivalents and will never be called by GCC.
# By putting them in LIB1ASMFUNCS, we avoid the standard libgcc2.c code being
===================================================================
@@ -16,7 +16,8 @@
# along with GCC; see the file COPYING3. If not see
# <http://www.gnu.org/licenses/>.
-LIB1ASMFUNCS += _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_tls _call_via_rX _interwork_call_via_rX _clzsi2 _clzdi2
+LIB1ASMFUNCS += _udivsi3 _divsi3 _umodsi3 _modsi3 _dvmd_tls _call_via_rX \
+ _interwork_call_via_rX _clzsi2 _clzdi2 _muldi3
# We want fine grained libraries, so use the new code to build the
# floating point emulation libraries.
Hi, This patch had been submitted for a previous gcc release quite some time ago but I missed the deadline last time. The patch implemented hand-tuned 64-bit multiplication run-time __aeabi_lmul. On ARM and THUMB-2, this is done using a very short, 5-instruction routine. On THUMB-1, this is done using the ARM version via interworking or a slower, hand-optimized version, depending on whether the interworking is supported or not. This is tested on arm-unknown-linux-gnueabi in qemu with not new regression. -Doug 2010-10-22 Doug Kwan <dougkwan@google.com> * config/arm/lib1funcs.asm (muldi3): Implement muldi3 in hand-written assembly. * config/arm/t-strongarm-elf (LIB1ASMFUNCS): Add _muldi3. * config/arm/t-vxworks: Ditto. * config/arm/t-pe: Ditto. * config/arm/t-arm-elf: Ditto. * config/arm/t-linux: Ditto. * config/arm/t-symbian: Ditto. * config/arm/t-wince-pe: Ditto.