mirror of
https://github.com/torvalds/linux.git
synced 2024-11-22 04:08:54 +00:00
5125d033c8
This allows compiling a full 128-bit product of two 64-bit integers as a
mul/mulh pair, instead of a nasty long sequence of 20+ instructions.
However, after selecting ARCH_SUPPORTS_INT128, when optimizing for size
the compiler generates calls to __ashlti3, __ashrti3, and __lshrti3 for
shifting __int128 values, causing a link failure:
loongarch64-unknown-linux-gnu-ld: kernel/sched/fair.o: in
function `mul_u64_u32_shr':
<PATH>/include/linux/math64.h:161:(.text+0x5e4): undefined
reference to `__lshrti3'
So provide the implementation of these functions if ARCH_SUPPORTS_INT128.
Closes: https://lore.kernel.org/loongarch/CAAhV-H5EZ=7OF7CSiYyZ8_+wWuenpo=K2WT8-6mAT4CvzUC_4g@mail.gmail.com/
Signed-off-by: Xi Ruoyao <xry111@xry111.site>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
57 lines
1018 B
ArmAsm
57 lines
1018 B
ArmAsm
/* SPDX-License-Identifier: GPL-2.0-only */
|
|
|
|
#include <asm/asmmacro.h>
|
|
#include <linux/export.h>
|
|
#include <linux/linkage.h>
|
|
|
|
SYM_FUNC_START(__ashlti3)
|
|
srli.d t2, a0, 1
|
|
nor t3, zero, a2
|
|
sll.d t1, a1, a2
|
|
srl.d t2, t2, t3
|
|
andi t0, a2, 64
|
|
sll.d a0, a0, a2
|
|
or t1, t2, t1
|
|
maskeqz a1, a0, t0
|
|
masknez a0, a0, t0
|
|
masknez t0, t1, t0
|
|
or a1, t0, a1
|
|
jr ra
|
|
SYM_FUNC_END(__ashlti3)
|
|
EXPORT_SYMBOL(__ashlti3)
|
|
|
|
SYM_FUNC_START(__ashrti3)
|
|
nor t3, zero, a2
|
|
slli.d t2, a1, 1
|
|
srl.d t1, a0, a2
|
|
sll.d t2, t2, t3
|
|
andi t0, a2, 64
|
|
or t1, t2, t1
|
|
sra.d a2, a1, a2
|
|
srai.d a1, a1, 63
|
|
maskeqz a0, a2, t0
|
|
maskeqz a1, a1, t0
|
|
masknez a2, a2, t0
|
|
masknez t0, t1, t0
|
|
or a1, a1, a2
|
|
or a0, t0, a0
|
|
jr ra
|
|
SYM_FUNC_END(__ashrti3)
|
|
EXPORT_SYMBOL(__ashrti3)
|
|
|
|
SYM_FUNC_START(__lshrti3)
|
|
slli.d t2, a1, 1
|
|
nor t3, zero, a2
|
|
srl.d t1, a0, a2
|
|
sll.d t2, t2, t3
|
|
andi t0, a2, 64
|
|
srl.d a1, a1, a2
|
|
or t1, t2, t1
|
|
maskeqz a0, a1, t0
|
|
masknez a1, a1, t0
|
|
masknez t0, t1, t0
|
|
or a0, t0, a0
|
|
jr ra
|
|
SYM_FUNC_END(__lshrti3)
|
|
EXPORT_SYMBOL(__lshrti3)
|