openwrtv3/target/linux/generic/hack-4.14/300-MIPS-r4k_cache-use-more-efficient-cache-blast.patch
Felix Fietkau 4e8f1e9f4c kernel: unroll MIPS r4k cache blast function
Optimize the compiler output for larger cache blast cases that are
common for DMA-based networking.

On ar71xx, I measured a routing throughput increase of ~8%

Signed-off-by: Ben Menchaca <ben.menchaca@qca.qualcomm.com>
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
2018-03-10 11:58:19 +01:00

66 lines
2.2 KiB
Diff

From: Ben Menchaca <ben.menchaca@qca.qualcomm.com>
Date: Fri, 7 Jun 2013 18:35:22 -0500
Subject: MIPS: r4k_cache: use more efficient cache blast
Optimize the compiler output for larger cache blast cases that are
common for DMA-based networking.
Signed-off-by: Ben Menchaca <ben.menchaca@qca.qualcomm.com>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
---
--- a/arch/mips/include/asm/r4kcache.h
+++ b/arch/mips/include/asm/r4kcache.h
@@ -682,16 +682,48 @@ static inline void prot##extra##blast_##
unsigned long end) \
{ \
unsigned long lsize = cpu_##desc##_line_size(); \
+ unsigned long lsize_2 = lsize * 2; \
+ unsigned long lsize_3 = lsize * 3; \
+ unsigned long lsize_4 = lsize * 4; \
+ unsigned long lsize_5 = lsize * 5; \
+ unsigned long lsize_6 = lsize * 6; \
+ unsigned long lsize_7 = lsize * 7; \
+ unsigned long lsize_8 = lsize * 8; \
unsigned long addr = start & ~(lsize - 1); \
- unsigned long aend = (end - 1) & ~(lsize - 1); \
+ unsigned long aend = (end + lsize - 1) & ~(lsize - 1); \
+ int lines = (aend - addr) / lsize; \
\
__##pfx##flush_prologue \
\
- while (1) { \
+ while (lines >= 8) { \
+ prot##cache_op(hitop, addr); \
+ prot##cache_op(hitop, addr + lsize); \
+ prot##cache_op(hitop, addr + lsize_2); \
+ prot##cache_op(hitop, addr + lsize_3); \
+ prot##cache_op(hitop, addr + lsize_4); \
+ prot##cache_op(hitop, addr + lsize_5); \
+ prot##cache_op(hitop, addr + lsize_6); \
+ prot##cache_op(hitop, addr + lsize_7); \
+ addr += lsize_8; \
+ lines -= 8; \
+ } \
+ \
+ if (lines & 0x4) { \
+ prot##cache_op(hitop, addr); \
+ prot##cache_op(hitop, addr + lsize); \
+ prot##cache_op(hitop, addr + lsize_2); \
+ prot##cache_op(hitop, addr + lsize_3); \
+ addr += lsize_4; \
+ } \
+ \
+ if (lines & 0x2) { \
+ prot##cache_op(hitop, addr); \
+ prot##cache_op(hitop, addr + lsize); \
+ addr += lsize_2; \
+ } \
+ \
+ if (lines & 0x1) { \
prot##cache_op(hitop, addr); \
- if (addr == aend) \
- break; \
- addr += lsize; \
} \
\
__##pfx##flush_epilogue \