* [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures
@ 2017-08-07 23:52 Babu Moger
2017-08-07 23:52 ` [PATCH v2 1/4] arch/sparc: Separate the exception handlers from NG4memcpy Babu Moger
` (4 more replies)
0 siblings, 5 replies; 7+ messages in thread
From: Babu Moger @ 2017-08-07 23:52 UTC (permalink / raw)
To: davem; +Cc: sparclinux, linux-kernel, babu.moger
This series of patches updates the memcpy, memset, copy_to_user, copy_from_user
etc for SPARC M7/M8 architecture.
New algorithm here takes advantage of the M7/M8 block init store ASIs, with much
more optimized way to improve the performance. More detail are in code comments.
Tested and compared the latency measured in ticks(NG4memcpy vs new M7memcpy).
1. Memset numbers(Aligned memset)
No.of bytes NG4memset M7memset Delta ((B-A)/A)*100
(Avg.Ticks A) (Avg.Ticks B) (latency reduction)
3 77 25 -67.53
7 43 33 -23.25
32 72 68 -5.55
128 164 44 -73.17
256 335 68 -79.70
512 511 220 -56.94
1024 1552 627 -59.60
2048 3515 1322 -62.38
4096 6303 2472 -60.78
8192 13118 4867 -62.89
16384 26206 10371 -60.42
32768 52501 18569 -64.63
65536 100219 35899 -64.17
2. Memcpy numbers(Aligned memcpy)
No.of bytes NG4memcpy M7memcpy Delta ((B-A)/A)*100
(Avg.Ticks A) (Avg.Ticks B) (latency reduction)
3 20 19 -5
7 29 27 -6.89
32 30 28 -6.66
128 89 69 -22.47
256 142 143 0.70
512 341 283 -17.00
1024 1588 655 -58.75
2048 3553 1357 -61.80
4096 7218 2590 -64.11
8192 13701 5231 -61.82
16384 28304 10716 -62.13
32768 56516 22995 -59.31
65536 115443 50840 -55.96
3. Memset numbers(un-aligned memset)
No.of bytes NG4memset M7memset Delta ((B-A)/A)*100
(Avg.Ticks A) (Avg.Ticks B) (latency reduction)
3 40 31 -22.5
7 52 29 -44.2307692308
32 89 86 -3.3707865169
128 201 74 -63.184079602
256 340 154 -54.7058823529
512 961 335 -65.1404786681
1024 1799 686 -61.8677042802
2048 3575 1260 -64.7552447552
4096 6560 2627 -59.9542682927
8192 13161 6018 -54.273991338
16384 26465 10439 -60.5554505951
32768 52119 18649 -64.2184232238
65536 101593 35724 -64.8361599717
4. Memcpy numbers(un-aligned memcpy)
No.of bytes NG4memcpy M7memcpy Delta ((B-A)/A)*100
(Avg.Ticks A) (Avg.Ticks B) (latency reduction)
3 26 19 -26.9230769231
7 48 45 -6.25
32 52 49 -5.7692307692
128 284 334 17.6056338028
256 430 482 12.0930232558
512 646 690 6.8111455108
1024 1051 1016 -3.3301617507
2048 1787 1818 1.7347509793
4096 3309 3376 2.0247809006
8192 8151 7444 -8.673782358
16384 34222 34556 0.9759803635
32768 87851 95044 8.1877269468
65536 158331 159572 0.7838010244
There is not much difference in numbers with Un-aligned copies
between NG4memcpy and M7memcpy because they both mostly use the
same algorithems.
v2:
1. Fixed indentation issues found by David Miller
2. Used ENTRY and ENDPROC for the labels in M7patch.S as suggested by David Miller
3. Now M8 also will use M7memcpy. Also tested on M8 config.
4. These patches are created on top of below M8 patches
https://patchwork.ozlabs.org/patch/792661/
https://patchwork.ozlabs.org/patch/792662/
However, I did not see these patches in sparc-next tree. It may be in queue now.
It is possible these patches might cause some build problems. It will resolve
once all M8 patches are in sparc-next tree.
v0: Initial version
Babu Moger (4):
arch/sparc: Separate the exception handlers from NG4memcpy
arch/sparc: Rename exception handlers
arch/sparc: Optimized memcpy, memset, copy_to_user, copy_from_user
for M7/M8
arch/sparc: Add accurate exception reporting in M7memcpy
arch/sparc/kernel/head_64.S | 16 +-
arch/sparc/lib/M7copy_from_user.S | 40 ++
arch/sparc/lib/M7copy_to_user.S | 51 ++
arch/sparc/lib/M7memcpy.S | 923 +++++++++++++++++++++++++++++++++++++
arch/sparc/lib/M7memset.S | 352 ++++++++++++++
arch/sparc/lib/M7patch.S | 51 ++
arch/sparc/lib/Makefile | 5 +
arch/sparc/lib/Memcpy_utils.S | 345 ++++++++++++++
arch/sparc/lib/NG4memcpy.S | 277 +++---------
9 files changed, 1845 insertions(+), 215 deletions(-)
create mode 100644 arch/sparc/lib/M7copy_from_user.S
create mode 100644 arch/sparc/lib/M7copy_to_user.S
create mode 100644 arch/sparc/lib/M7memcpy.S
create mode 100644 arch/sparc/lib/M7memset.S
create mode 100644 arch/sparc/lib/M7patch.S
create mode 100644 arch/sparc/lib/Memcpy_utils.S
^ permalink raw reply [flat|nested] 7+ messages in thread
* [PATCH v2 1/4] arch/sparc: Separate the exception handlers from NG4memcpy
2017-08-07 23:52 [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures Babu Moger
@ 2017-08-07 23:52 ` Babu Moger
2017-08-07 23:52 ` [PATCH v2 2/4] arch/sparc: Rename exception handlers Babu Moger
` (3 subsequent siblings)
4 siblings, 0 replies; 7+ messages in thread
From: Babu Moger @ 2017-08-07 23:52 UTC (permalink / raw)
To: davem; +Cc: sparclinux, linux-kernel, babu.moger
Separate the exception handlers from NG4memcpy so that it can be
used with new memcpy routines. Make a separate file for all these handlers.
Signed-off-by: Babu Moger <babu.moger@oracle.com>
---
arch/sparc/lib/Makefile | 2 +
arch/sparc/lib/Memcpy_utils.S | 163 +++++++++++++++++++++++++++++++++++++++++
arch/sparc/lib/NG4memcpy.S | 149 -------------------------------------
3 files changed, 165 insertions(+), 149 deletions(-)
create mode 100644 arch/sparc/lib/Memcpy_utils.S
diff --git a/arch/sparc/lib/Makefile b/arch/sparc/lib/Makefile
index 07c03e7..37930c0 100644
--- a/arch/sparc/lib/Makefile
+++ b/arch/sparc/lib/Makefile
@@ -36,6 +36,8 @@ lib-$(CONFIG_SPARC64) += NG2patch.o
lib-$(CONFIG_SPARC64) += NG4memcpy.o NG4copy_from_user.o NG4copy_to_user.o
lib-$(CONFIG_SPARC64) += NG4patch.o NG4copy_page.o NG4clear_page.o NG4memset.o
+lib-$(CONFIG_SPARC64) += Memcpy_utils.o
+
lib-$(CONFIG_SPARC64) += GENmemcpy.o GENcopy_from_user.o GENcopy_to_user.o
lib-$(CONFIG_SPARC64) += GENpatch.o GENpage.o GENbzero.o
diff --git a/arch/sparc/lib/Memcpy_utils.S b/arch/sparc/lib/Memcpy_utils.S
new file mode 100644
index 0000000..f7a26e0
--- /dev/null
+++ b/arch/sparc/lib/Memcpy_utils.S
@@ -0,0 +1,163 @@
+#ifndef __ASM_MEMCPY_UTILS
+#define __ASM_MEMCPY_UTILS
+
+#include <linux/linkage.h>
+#include <asm/asi.h>
+#include <asm/visasm.h>
+
+ENTRY(__restore_asi_fp)
+ VISExitHalf
+ retl
+ wr %g0, ASI_AIUS, %asi
+ENDPROC(__restore_asi_fp)
+
+ENTRY(__restore_asi)
+ retl
+ wr %g0, ASI_AIUS, %asi
+ENDPROC(__restore_asi)
+
+ENTRY(NG4_retl_o2)
+ ba,pt %xcc, __restore_asi
+ mov %o2, %o0
+ENDPROC(NG4_retl_o2)
+ENTRY(NG4_retl_o2_plus_1)
+ ba,pt %xcc, __restore_asi
+ add %o2, 1, %o0
+ENDPROC(NG4_retl_o2_plus_1)
+ENTRY(NG4_retl_o2_plus_4)
+ ba,pt %xcc, __restore_asi
+ add %o2, 4, %o0
+ENDPROC(NG4_retl_o2_plus_4)
+ENTRY(NG4_retl_o2_plus_o5)
+ ba,pt %xcc, __restore_asi
+ add %o2, %o5, %o0
+ENDPROC(NG4_retl_o2_plus_o5)
+ENTRY(NG4_retl_o2_plus_o5_plus_4)
+ add %o5, 4, %o5
+ ba,pt %xcc, __restore_asi
+ add %o2, %o5, %o0
+ENDPROC(NG4_retl_o2_plus_o5_plus_4)
+ENTRY(NG4_retl_o2_plus_o5_plus_8)
+ add %o5, 8, %o5
+ ba,pt %xcc, __restore_asi
+ add %o2, %o5, %o0
+ENDPROC(NG4_retl_o2_plus_o5_plus_8)
+ENTRY(NG4_retl_o2_plus_o5_plus_16)
+ add %o5, 16, %o5
+ ba,pt %xcc, __restore_asi
+ add %o2, %o5, %o0
+ENDPROC(NG4_retl_o2_plus_o5_plus_16)
+ENTRY(NG4_retl_o2_plus_o5_plus_24)
+ add %o5, 24, %o5
+ ba,pt %xcc, __restore_asi
+ add %o2, %o5, %o0
+ENDPROC(NG4_retl_o2_plus_o5_plus_24)
+ENTRY(NG4_retl_o2_plus_o5_plus_32)
+ add %o5, 32, %o5
+ ba,pt %xcc, __restore_asi
+ add %o2, %o5, %o0
+ENDPROC(NG4_retl_o2_plus_o5_plus_32)
+ENTRY(NG4_retl_o2_plus_g1)
+ ba,pt %xcc, __restore_asi
+ add %o2, %g1, %o0
+ENDPROC(NG4_retl_o2_plus_g1)
+ENTRY(NG4_retl_o2_plus_g1_plus_1)
+ add %g1, 1, %g1
+ ba,pt %xcc, __restore_asi
+ add %o2, %g1, %o0
+ENDPROC(NG4_retl_o2_plus_g1_plus_1)
+ENTRY(NG4_retl_o2_plus_g1_plus_8)
+ add %g1, 8, %g1
+ ba,pt %xcc, __restore_asi
+ add %o2, %g1, %o0
+ENDPROC(NG4_retl_o2_plus_g1_plus_8)
+ENTRY(NG4_retl_o2_plus_o4)
+ ba,pt %xcc, __restore_asi
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4)
+ENTRY(NG4_retl_o2_plus_o4_plus_8)
+ add %o4, 8, %o4
+ ba,pt %xcc, __restore_asi
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_8)
+ENTRY(NG4_retl_o2_plus_o4_plus_16)
+ add %o4, 16, %o4
+ ba,pt %xcc, __restore_asi
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_16)
+ENTRY(NG4_retl_o2_plus_o4_plus_24)
+ add %o4, 24, %o4
+ ba,pt %xcc, __restore_asi
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_24)
+ENTRY(NG4_retl_o2_plus_o4_plus_32)
+ add %o4, 32, %o4
+ ba,pt %xcc, __restore_asi
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_32)
+ENTRY(NG4_retl_o2_plus_o4_plus_40)
+ add %o4, 40, %o4
+ ba,pt %xcc, __restore_asi
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_40)
+ENTRY(NG4_retl_o2_plus_o4_plus_48)
+ add %o4, 48, %o4
+ ba,pt %xcc, __restore_asi
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_48)
+ENTRY(NG4_retl_o2_plus_o4_plus_56)
+ add %o4, 56, %o4
+ ba,pt %xcc, __restore_asi
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_56)
+ENTRY(NG4_retl_o2_plus_o4_plus_64)
+ add %o4, 64, %o4
+ ba,pt %xcc, __restore_asi
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_64)
+ENTRY(NG4_retl_o2_plus_o4_fp)
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_fp)
+ENTRY(NG4_retl_o2_plus_o4_plus_8_fp)
+ add %o4, 8, %o4
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_8_fp)
+ENTRY(NG4_retl_o2_plus_o4_plus_16_fp)
+ add %o4, 16, %o4
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_16_fp)
+ENTRY(NG4_retl_o2_plus_o4_plus_24_fp)
+ add %o4, 24, %o4
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_24_fp)
+ENTRY(NG4_retl_o2_plus_o4_plus_32_fp)
+ add %o4, 32, %o4
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_32_fp)
+ENTRY(NG4_retl_o2_plus_o4_plus_40_fp)
+ add %o4, 40, %o4
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_40_fp)
+ENTRY(NG4_retl_o2_plus_o4_plus_48_fp)
+ add %o4, 48, %o4
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_48_fp)
+ENTRY(NG4_retl_o2_plus_o4_plus_56_fp)
+ add %o4, 56, %o4
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_56_fp)
+ENTRY(NG4_retl_o2_plus_o4_plus_64_fp)
+ add %o4, 64, %o4
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o4, %o0
+ENDPROC(NG4_retl_o2_plus_o4_plus_64_fp)
+
+#endif
diff --git a/arch/sparc/lib/NG4memcpy.S b/arch/sparc/lib/NG4memcpy.S
index 78ea962..e20d654 100644
--- a/arch/sparc/lib/NG4memcpy.S
+++ b/arch/sparc/lib/NG4memcpy.S
@@ -94,155 +94,6 @@
.text
#ifndef EX_RETVAL
#define EX_RETVAL(x) x
-__restore_asi_fp:
- VISExitHalf
-__restore_asi:
- retl
- wr %g0, ASI_AIUS, %asi
-
-ENTRY(NG4_retl_o2)
- ba,pt %xcc, __restore_asi
- mov %o2, %o0
-ENDPROC(NG4_retl_o2)
-ENTRY(NG4_retl_o2_plus_1)
- ba,pt %xcc, __restore_asi
- add %o2, 1, %o0
-ENDPROC(NG4_retl_o2_plus_1)
-ENTRY(NG4_retl_o2_plus_4)
- ba,pt %xcc, __restore_asi
- add %o2, 4, %o0
-ENDPROC(NG4_retl_o2_plus_4)
-ENTRY(NG4_retl_o2_plus_o5)
- ba,pt %xcc, __restore_asi
- add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5)
-ENTRY(NG4_retl_o2_plus_o5_plus_4)
- add %o5, 4, %o5
- ba,pt %xcc, __restore_asi
- add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5_plus_4)
-ENTRY(NG4_retl_o2_plus_o5_plus_8)
- add %o5, 8, %o5
- ba,pt %xcc, __restore_asi
- add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5_plus_8)
-ENTRY(NG4_retl_o2_plus_o5_plus_16)
- add %o5, 16, %o5
- ba,pt %xcc, __restore_asi
- add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5_plus_16)
-ENTRY(NG4_retl_o2_plus_o5_plus_24)
- add %o5, 24, %o5
- ba,pt %xcc, __restore_asi
- add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5_plus_24)
-ENTRY(NG4_retl_o2_plus_o5_plus_32)
- add %o5, 32, %o5
- ba,pt %xcc, __restore_asi
- add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5_plus_32)
-ENTRY(NG4_retl_o2_plus_g1)
- ba,pt %xcc, __restore_asi
- add %o2, %g1, %o0
-ENDPROC(NG4_retl_o2_plus_g1)
-ENTRY(NG4_retl_o2_plus_g1_plus_1)
- add %g1, 1, %g1
- ba,pt %xcc, __restore_asi
- add %o2, %g1, %o0
-ENDPROC(NG4_retl_o2_plus_g1_plus_1)
-ENTRY(NG4_retl_o2_plus_g1_plus_8)
- add %g1, 8, %g1
- ba,pt %xcc, __restore_asi
- add %o2, %g1, %o0
-ENDPROC(NG4_retl_o2_plus_g1_plus_8)
-ENTRY(NG4_retl_o2_plus_o4)
- ba,pt %xcc, __restore_asi
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4)
-ENTRY(NG4_retl_o2_plus_o4_plus_8)
- add %o4, 8, %o4
- ba,pt %xcc, __restore_asi
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_8)
-ENTRY(NG4_retl_o2_plus_o4_plus_16)
- add %o4, 16, %o4
- ba,pt %xcc, __restore_asi
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_16)
-ENTRY(NG4_retl_o2_plus_o4_plus_24)
- add %o4, 24, %o4
- ba,pt %xcc, __restore_asi
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_24)
-ENTRY(NG4_retl_o2_plus_o4_plus_32)
- add %o4, 32, %o4
- ba,pt %xcc, __restore_asi
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_32)
-ENTRY(NG4_retl_o2_plus_o4_plus_40)
- add %o4, 40, %o4
- ba,pt %xcc, __restore_asi
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_40)
-ENTRY(NG4_retl_o2_plus_o4_plus_48)
- add %o4, 48, %o4
- ba,pt %xcc, __restore_asi
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_48)
-ENTRY(NG4_retl_o2_plus_o4_plus_56)
- add %o4, 56, %o4
- ba,pt %xcc, __restore_asi
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_56)
-ENTRY(NG4_retl_o2_plus_o4_plus_64)
- add %o4, 64, %o4
- ba,pt %xcc, __restore_asi
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_64)
-ENTRY(NG4_retl_o2_plus_o4_fp)
- ba,pt %xcc, __restore_asi_fp
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_8_fp)
- add %o4, 8, %o4
- ba,pt %xcc, __restore_asi_fp
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_8_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_16_fp)
- add %o4, 16, %o4
- ba,pt %xcc, __restore_asi_fp
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_16_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_24_fp)
- add %o4, 24, %o4
- ba,pt %xcc, __restore_asi_fp
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_24_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_32_fp)
- add %o4, 32, %o4
- ba,pt %xcc, __restore_asi_fp
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_32_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_40_fp)
- add %o4, 40, %o4
- ba,pt %xcc, __restore_asi_fp
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_40_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_48_fp)
- add %o4, 48, %o4
- ba,pt %xcc, __restore_asi_fp
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_48_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_56_fp)
- add %o4, 56, %o4
- ba,pt %xcc, __restore_asi_fp
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_56_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_64_fp)
- add %o4, 64, %o4
- ba,pt %xcc, __restore_asi_fp
- add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_64_fp)
#endif
.align 64
--
1.7.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH v2 2/4] arch/sparc: Rename exception handlers
2017-08-07 23:52 [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures Babu Moger
2017-08-07 23:52 ` [PATCH v2 1/4] arch/sparc: Separate the exception handlers from NG4memcpy Babu Moger
@ 2017-08-07 23:52 ` Babu Moger
2017-08-07 23:52 ` [PATCH v2 3/4] arch/sparc: Optimized memcpy, memset, copy_to_user, copy_from_user for M7/M8 Babu Moger
` (2 subsequent siblings)
4 siblings, 0 replies; 7+ messages in thread
From: Babu Moger @ 2017-08-07 23:52 UTC (permalink / raw)
To: davem; +Cc: sparclinux, linux-kernel, babu.moger
Rename exception handlers to memcpy_xxx as these
are going to be used by new memcpy routines and these
handlers are not exclusive to NG4memcpy anymore.
Signed-off-by: Babu Moger <babu.moger@oracle.com>
---
arch/sparc/lib/Memcpy_utils.S | 120 +++++++++++++++++++-------------------
arch/sparc/lib/NG4memcpy.S | 128 ++++++++++++++++++++--------------------
2 files changed, 124 insertions(+), 124 deletions(-)
diff --git a/arch/sparc/lib/Memcpy_utils.S b/arch/sparc/lib/Memcpy_utils.S
index f7a26e0..bcc5d77 100644
--- a/arch/sparc/lib/Memcpy_utils.S
+++ b/arch/sparc/lib/Memcpy_utils.S
@@ -16,148 +16,148 @@ ENTRY(__restore_asi)
wr %g0, ASI_AIUS, %asi
ENDPROC(__restore_asi)
-ENTRY(NG4_retl_o2)
+ENTRY(memcpy_retl_o2)
ba,pt %xcc, __restore_asi
mov %o2, %o0
-ENDPROC(NG4_retl_o2)
-ENTRY(NG4_retl_o2_plus_1)
+ENDPROC(memcpy_retl_o2)
+ENTRY(memcpy_retl_o2_plus_1)
ba,pt %xcc, __restore_asi
add %o2, 1, %o0
-ENDPROC(NG4_retl_o2_plus_1)
-ENTRY(NG4_retl_o2_plus_4)
+ENDPROC(memcpy_retl_o2_plus_1)
+ENTRY(memcpy_retl_o2_plus_4)
ba,pt %xcc, __restore_asi
add %o2, 4, %o0
-ENDPROC(NG4_retl_o2_plus_4)
-ENTRY(NG4_retl_o2_plus_o5)
+ENDPROC(memcpy_retl_o2_plus_4)
+ENTRY(memcpy_retl_o2_plus_o5)
ba,pt %xcc, __restore_asi
add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5)
-ENTRY(NG4_retl_o2_plus_o5_plus_4)
+ENDPROC(memcpy_retl_o2_plus_o5)
+ENTRY(memcpy_retl_o2_plus_o5_plus_4)
add %o5, 4, %o5
ba,pt %xcc, __restore_asi
add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5_plus_4)
-ENTRY(NG4_retl_o2_plus_o5_plus_8)
+ENDPROC(memcpy_retl_o2_plus_o5_plus_4)
+ENTRY(memcpy_retl_o2_plus_o5_plus_8)
add %o5, 8, %o5
ba,pt %xcc, __restore_asi
add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5_plus_8)
-ENTRY(NG4_retl_o2_plus_o5_plus_16)
+ENDPROC(memcpy_retl_o2_plus_o5_plus_8)
+ENTRY(memcpy_retl_o2_plus_o5_plus_16)
add %o5, 16, %o5
ba,pt %xcc, __restore_asi
add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5_plus_16)
-ENTRY(NG4_retl_o2_plus_o5_plus_24)
+ENDPROC(memcpy_retl_o2_plus_o5_plus_16)
+ENTRY(memcpy_retl_o2_plus_o5_plus_24)
add %o5, 24, %o5
ba,pt %xcc, __restore_asi
add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5_plus_24)
-ENTRY(NG4_retl_o2_plus_o5_plus_32)
+ENDPROC(memcpy_retl_o2_plus_o5_plus_24)
+ENTRY(memcpy_retl_o2_plus_o5_plus_32)
add %o5, 32, %o5
ba,pt %xcc, __restore_asi
add %o2, %o5, %o0
-ENDPROC(NG4_retl_o2_plus_o5_plus_32)
-ENTRY(NG4_retl_o2_plus_g1)
+ENDPROC(memcpy_retl_o2_plus_o5_plus_32)
+ENTRY(memcpy_retl_o2_plus_g1)
ba,pt %xcc, __restore_asi
add %o2, %g1, %o0
-ENDPROC(NG4_retl_o2_plus_g1)
-ENTRY(NG4_retl_o2_plus_g1_plus_1)
+ENDPROC(memcpy_retl_o2_plus_g1)
+ENTRY(memcpy_retl_o2_plus_g1_plus_1)
add %g1, 1, %g1
ba,pt %xcc, __restore_asi
add %o2, %g1, %o0
-ENDPROC(NG4_retl_o2_plus_g1_plus_1)
-ENTRY(NG4_retl_o2_plus_g1_plus_8)
+ENDPROC(memcpy_retl_o2_plus_g1_plus_1)
+ENTRY(memcpy_retl_o2_plus_g1_plus_8)
add %g1, 8, %g1
ba,pt %xcc, __restore_asi
add %o2, %g1, %o0
-ENDPROC(NG4_retl_o2_plus_g1_plus_8)
-ENTRY(NG4_retl_o2_plus_o4)
+ENDPROC(memcpy_retl_o2_plus_g1_plus_8)
+ENTRY(memcpy_retl_o2_plus_o4)
ba,pt %xcc, __restore_asi
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4)
-ENTRY(NG4_retl_o2_plus_o4_plus_8)
+ENDPROC(memcpy_retl_o2_plus_o4)
+ENTRY(memcpy_retl_o2_plus_o4_plus_8)
add %o4, 8, %o4
ba,pt %xcc, __restore_asi
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_8)
-ENTRY(NG4_retl_o2_plus_o4_plus_16)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_8)
+ENTRY(memcpy_retl_o2_plus_o4_plus_16)
add %o4, 16, %o4
ba,pt %xcc, __restore_asi
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_16)
-ENTRY(NG4_retl_o2_plus_o4_plus_24)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_16)
+ENTRY(memcpy_retl_o2_plus_o4_plus_24)
add %o4, 24, %o4
ba,pt %xcc, __restore_asi
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_24)
-ENTRY(NG4_retl_o2_plus_o4_plus_32)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_24)
+ENTRY(memcpy_retl_o2_plus_o4_plus_32)
add %o4, 32, %o4
ba,pt %xcc, __restore_asi
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_32)
-ENTRY(NG4_retl_o2_plus_o4_plus_40)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_32)
+ENTRY(memcpy_retl_o2_plus_o4_plus_40)
add %o4, 40, %o4
ba,pt %xcc, __restore_asi
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_40)
-ENTRY(NG4_retl_o2_plus_o4_plus_48)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_40)
+ENTRY(memcpy_retl_o2_plus_o4_plus_48)
add %o4, 48, %o4
ba,pt %xcc, __restore_asi
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_48)
-ENTRY(NG4_retl_o2_plus_o4_plus_56)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_48)
+ENTRY(memcpy_retl_o2_plus_o4_plus_56)
add %o4, 56, %o4
ba,pt %xcc, __restore_asi
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_56)
-ENTRY(NG4_retl_o2_plus_o4_plus_64)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_56)
+ENTRY(memcpy_retl_o2_plus_o4_plus_64)
add %o4, 64, %o4
ba,pt %xcc, __restore_asi
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_64)
-ENTRY(NG4_retl_o2_plus_o4_fp)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_64)
+ENTRY(memcpy_retl_o2_plus_o4_fp)
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_8_fp)
+ENDPROC(memcpy_retl_o2_plus_o4_fp)
+ENTRY(memcpy_retl_o2_plus_o4_plus_8_fp)
add %o4, 8, %o4
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_8_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_16_fp)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_8_fp)
+ENTRY(memcpy_retl_o2_plus_o4_plus_16_fp)
add %o4, 16, %o4
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_16_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_24_fp)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_16_fp)
+ENTRY(memcpy_retl_o2_plus_o4_plus_24_fp)
add %o4, 24, %o4
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_24_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_32_fp)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_24_fp)
+ENTRY(memcpy_retl_o2_plus_o4_plus_32_fp)
add %o4, 32, %o4
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_32_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_40_fp)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_32_fp)
+ENTRY(memcpy_retl_o2_plus_o4_plus_40_fp)
add %o4, 40, %o4
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_40_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_48_fp)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_40_fp)
+ENTRY(memcpy_retl_o2_plus_o4_plus_48_fp)
add %o4, 48, %o4
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_48_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_56_fp)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_48_fp)
+ENTRY(memcpy_retl_o2_plus_o4_plus_56_fp)
add %o4, 56, %o4
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_56_fp)
-ENTRY(NG4_retl_o2_plus_o4_plus_64_fp)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_56_fp)
+ENTRY(memcpy_retl_o2_plus_o4_plus_64_fp)
add %o4, 64, %o4
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
-ENDPROC(NG4_retl_o2_plus_o4_plus_64_fp)
+ENDPROC(memcpy_retl_o2_plus_o4_plus_64_fp)
#endif
diff --git a/arch/sparc/lib/NG4memcpy.S b/arch/sparc/lib/NG4memcpy.S
index e20d654..b5dacd1 100644
--- a/arch/sparc/lib/NG4memcpy.S
+++ b/arch/sparc/lib/NG4memcpy.S
@@ -126,12 +126,12 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
sub %o2, %g1, %o2
-1: EX_LD(LOAD(ldub, %o1 + 0x00, %g2), NG4_retl_o2_plus_g1)
+1: EX_LD(LOAD(ldub, %o1 + 0x00, %g2), memcpy_retl_o2_plus_g1)
add %o1, 1, %o1
subcc %g1, 1, %g1
add %o0, 1, %o0
bne,pt %icc, 1b
- EX_ST(STORE(stb, %g2, %o0 - 0x01), NG4_retl_o2_plus_g1_plus_1)
+ EX_ST(STORE(stb, %g2, %o0 - 0x01), memcpy_retl_o2_plus_g1_plus_1)
51: LOAD(prefetch, %o1 + 0x040, #n_reads_strong)
LOAD(prefetch, %o1 + 0x080, #n_reads_strong)
@@ -156,43 +156,43 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
brz,pt %g1, .Llarge_aligned
sub %o2, %g1, %o2
-1: EX_LD(LOAD(ldx, %o1 + 0x00, %g2), NG4_retl_o2_plus_g1)
+1: EX_LD(LOAD(ldx, %o1 + 0x00, %g2), memcpy_retl_o2_plus_g1)
add %o1, 8, %o1
subcc %g1, 8, %g1
add %o0, 8, %o0
bne,pt %icc, 1b
- EX_ST(STORE(stx, %g2, %o0 - 0x08), NG4_retl_o2_plus_g1_plus_8)
+ EX_ST(STORE(stx, %g2, %o0 - 0x08), memcpy_retl_o2_plus_g1_plus_8)
.Llarge_aligned:
/* len >= 0x80 && src 8-byte aligned && dest 8-byte aligned */
andn %o2, 0x3f, %o4
sub %o2, %o4, %o2
-1: EX_LD(LOAD(ldx, %o1 + 0x00, %g1), NG4_retl_o2_plus_o4)
+1: EX_LD(LOAD(ldx, %o1 + 0x00, %g1), memcpy_retl_o2_plus_o4)
add %o1, 0x40, %o1
- EX_LD(LOAD(ldx, %o1 - 0x38, %g2), NG4_retl_o2_plus_o4)
+ EX_LD(LOAD(ldx, %o1 - 0x38, %g2), memcpy_retl_o2_plus_o4)
subcc %o4, 0x40, %o4
- EX_LD(LOAD(ldx, %o1 - 0x30, %g3), NG4_retl_o2_plus_o4_plus_64)
- EX_LD(LOAD(ldx, %o1 - 0x28, GLOBAL_SPARE), NG4_retl_o2_plus_o4_plus_64)
- EX_LD(LOAD(ldx, %o1 - 0x20, %o5), NG4_retl_o2_plus_o4_plus_64)
- EX_ST(STORE_INIT(%g1, %o0), NG4_retl_o2_plus_o4_plus_64)
+ EX_LD(LOAD(ldx, %o1 - 0x30, %g3), memcpy_retl_o2_plus_o4_plus_64)
+ EX_LD(LOAD(ldx, %o1 - 0x28, GLOBAL_SPARE), memcpy_retl_o2_plus_o4_plus_64)
+ EX_LD(LOAD(ldx, %o1 - 0x20, %o5), memcpy_retl_o2_plus_o4_plus_64)
+ EX_ST(STORE_INIT(%g1, %o0), memcpy_retl_o2_plus_o4_plus_64)
add %o0, 0x08, %o0
- EX_ST(STORE_INIT(%g2, %o0), NG4_retl_o2_plus_o4_plus_56)
+ EX_ST(STORE_INIT(%g2, %o0), memcpy_retl_o2_plus_o4_plus_56)
add %o0, 0x08, %o0
- EX_LD(LOAD(ldx, %o1 - 0x18, %g2), NG4_retl_o2_plus_o4_plus_48)
- EX_ST(STORE_INIT(%g3, %o0), NG4_retl_o2_plus_o4_plus_48)
+ EX_LD(LOAD(ldx, %o1 - 0x18, %g2), memcpy_retl_o2_plus_o4_plus_48)
+ EX_ST(STORE_INIT(%g3, %o0), memcpy_retl_o2_plus_o4_plus_48)
add %o0, 0x08, %o0
- EX_LD(LOAD(ldx, %o1 - 0x10, %g3), NG4_retl_o2_plus_o4_plus_40)
- EX_ST(STORE_INIT(GLOBAL_SPARE, %o0), NG4_retl_o2_plus_o4_plus_40)
+ EX_LD(LOAD(ldx, %o1 - 0x10, %g3), memcpy_retl_o2_plus_o4_plus_40)
+ EX_ST(STORE_INIT(GLOBAL_SPARE, %o0), memcpy_retl_o2_plus_o4_plus_40)
add %o0, 0x08, %o0
- EX_LD(LOAD(ldx, %o1 - 0x08, GLOBAL_SPARE), NG4_retl_o2_plus_o4_plus_32)
- EX_ST(STORE_INIT(%o5, %o0), NG4_retl_o2_plus_o4_plus_32)
+ EX_LD(LOAD(ldx, %o1 - 0x08, GLOBAL_SPARE), memcpy_retl_o2_plus_o4_plus_32)
+ EX_ST(STORE_INIT(%o5, %o0), memcpy_retl_o2_plus_o4_plus_32)
add %o0, 0x08, %o0
- EX_ST(STORE_INIT(%g2, %o0), NG4_retl_o2_plus_o4_plus_24)
+ EX_ST(STORE_INIT(%g2, %o0), memcpy_retl_o2_plus_o4_plus_24)
add %o0, 0x08, %o0
- EX_ST(STORE_INIT(%g3, %o0), NG4_retl_o2_plus_o4_plus_16)
+ EX_ST(STORE_INIT(%g3, %o0), memcpy_retl_o2_plus_o4_plus_16)
add %o0, 0x08, %o0
- EX_ST(STORE_INIT(GLOBAL_SPARE, %o0), NG4_retl_o2_plus_o4_plus_8)
+ EX_ST(STORE_INIT(GLOBAL_SPARE, %o0), memcpy_retl_o2_plus_o4_plus_8)
add %o0, 0x08, %o0
bne,pt %icc, 1b
LOAD(prefetch, %o1 + 0x200, #n_reads_strong)
@@ -218,17 +218,17 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
sub %o2, %o4, %o2
alignaddr %o1, %g0, %g1
add %o1, %o4, %o1
- EX_LD_FP(LOAD(ldd, %g1 + 0x00, %f0), NG4_retl_o2_plus_o4)
-1: EX_LD_FP(LOAD(ldd, %g1 + 0x08, %f2), NG4_retl_o2_plus_o4)
+ EX_LD_FP(LOAD(ldd, %g1 + 0x00, %f0), memcpy_retl_o2_plus_o4)
+1: EX_LD_FP(LOAD(ldd, %g1 + 0x08, %f2), memcpy_retl_o2_plus_o4)
subcc %o4, 0x40, %o4
- EX_LD_FP(LOAD(ldd, %g1 + 0x10, %f4), NG4_retl_o2_plus_o4_plus_64)
- EX_LD_FP(LOAD(ldd, %g1 + 0x18, %f6), NG4_retl_o2_plus_o4_plus_64)
- EX_LD_FP(LOAD(ldd, %g1 + 0x20, %f8), NG4_retl_o2_plus_o4_plus_64)
- EX_LD_FP(LOAD(ldd, %g1 + 0x28, %f10), NG4_retl_o2_plus_o4_plus_64)
- EX_LD_FP(LOAD(ldd, %g1 + 0x30, %f12), NG4_retl_o2_plus_o4_plus_64)
- EX_LD_FP(LOAD(ldd, %g1 + 0x38, %f14), NG4_retl_o2_plus_o4_plus_64)
+ EX_LD_FP(LOAD(ldd, %g1 + 0x10, %f4), memcpy_retl_o2_plus_o4_plus_64)
+ EX_LD_FP(LOAD(ldd, %g1 + 0x18, %f6), memcpy_retl_o2_plus_o4_plus_64)
+ EX_LD_FP(LOAD(ldd, %g1 + 0x20, %f8), memcpy_retl_o2_plus_o4_plus_64)
+ EX_LD_FP(LOAD(ldd, %g1 + 0x28, %f10), memcpy_retl_o2_plus_o4_plus_64)
+ EX_LD_FP(LOAD(ldd, %g1 + 0x30, %f12), memcpy_retl_o2_plus_o4_plus_64)
+ EX_LD_FP(LOAD(ldd, %g1 + 0x38, %f14), memcpy_retl_o2_plus_o4_plus_64)
faligndata %f0, %f2, %f16
- EX_LD_FP(LOAD(ldd, %g1 + 0x40, %f0), NG4_retl_o2_plus_o4_plus_64)
+ EX_LD_FP(LOAD(ldd, %g1 + 0x40, %f0), memcpy_retl_o2_plus_o4_plus_64)
faligndata %f2, %f4, %f18
add %g1, 0x40, %g1
faligndata %f4, %f6, %f20
@@ -237,14 +237,14 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
faligndata %f10, %f12, %f26
faligndata %f12, %f14, %f28
faligndata %f14, %f0, %f30
- EX_ST_FP(STORE(std, %f16, %o0 + 0x00), NG4_retl_o2_plus_o4_plus_64)
- EX_ST_FP(STORE(std, %f18, %o0 + 0x08), NG4_retl_o2_plus_o4_plus_56)
- EX_ST_FP(STORE(std, %f20, %o0 + 0x10), NG4_retl_o2_plus_o4_plus_48)
- EX_ST_FP(STORE(std, %f22, %o0 + 0x18), NG4_retl_o2_plus_o4_plus_40)
- EX_ST_FP(STORE(std, %f24, %o0 + 0x20), NG4_retl_o2_plus_o4_plus_32)
- EX_ST_FP(STORE(std, %f26, %o0 + 0x28), NG4_retl_o2_plus_o4_plus_24)
- EX_ST_FP(STORE(std, %f28, %o0 + 0x30), NG4_retl_o2_plus_o4_plus_16)
- EX_ST_FP(STORE(std, %f30, %o0 + 0x38), NG4_retl_o2_plus_o4_plus_8)
+ EX_ST_FP(STORE(std, %f16, %o0 + 0x00), memcpy_retl_o2_plus_o4_plus_64)
+ EX_ST_FP(STORE(std, %f18, %o0 + 0x08), memcpy_retl_o2_plus_o4_plus_56)
+ EX_ST_FP(STORE(std, %f20, %o0 + 0x10), memcpy_retl_o2_plus_o4_plus_48)
+ EX_ST_FP(STORE(std, %f22, %o0 + 0x18), memcpy_retl_o2_plus_o4_plus_40)
+ EX_ST_FP(STORE(std, %f24, %o0 + 0x20), memcpy_retl_o2_plus_o4_plus_32)
+ EX_ST_FP(STORE(std, %f26, %o0 + 0x28), memcpy_retl_o2_plus_o4_plus_24)
+ EX_ST_FP(STORE(std, %f28, %o0 + 0x30), memcpy_retl_o2_plus_o4_plus_16)
+ EX_ST_FP(STORE(std, %f30, %o0 + 0x38), memcpy_retl_o2_plus_o4_plus_8)
add %o0, 0x40, %o0
bne,pt %icc, 1b
LOAD(prefetch, %g1 + 0x200, #n_reads_strong)
@@ -272,38 +272,38 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
andncc %o2, 0x20 - 1, %o5
be,pn %icc, 2f
sub %o2, %o5, %o2
-1: EX_LD(LOAD(ldx, %o1 + 0x00, %g1), NG4_retl_o2_plus_o5)
- EX_LD(LOAD(ldx, %o1 + 0x08, %g2), NG4_retl_o2_plus_o5)
- EX_LD(LOAD(ldx, %o1 + 0x10, GLOBAL_SPARE), NG4_retl_o2_plus_o5)
- EX_LD(LOAD(ldx, %o1 + 0x18, %o4), NG4_retl_o2_plus_o5)
+1: EX_LD(LOAD(ldx, %o1 + 0x00, %g1), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1 + 0x08, %g2), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1 + 0x10, GLOBAL_SPARE), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1 + 0x18, %o4), memcpy_retl_o2_plus_o5)
add %o1, 0x20, %o1
subcc %o5, 0x20, %o5
- EX_ST(STORE(stx, %g1, %o0 + 0x00), NG4_retl_o2_plus_o5_plus_32)
- EX_ST(STORE(stx, %g2, %o0 + 0x08), NG4_retl_o2_plus_o5_plus_24)
- EX_ST(STORE(stx, GLOBAL_SPARE, %o0 + 0x10), NG4_retl_o2_plus_o5_plus_24)
- EX_ST(STORE(stx, %o4, %o0 + 0x18), NG4_retl_o2_plus_o5_plus_8)
+ EX_ST(STORE(stx, %g1, %o0 + 0x00), memcpy_retl_o2_plus_o5_plus_32)
+ EX_ST(STORE(stx, %g2, %o0 + 0x08), memcpy_retl_o2_plus_o5_plus_24)
+ EX_ST(STORE(stx, GLOBAL_SPARE, %o0 + 0x10), memcpy_retl_o2_plus_o5_plus_24)
+ EX_ST(STORE(stx, %o4, %o0 + 0x18), memcpy_retl_o2_plus_o5_plus_8)
bne,pt %icc, 1b
add %o0, 0x20, %o0
2: andcc %o2, 0x18, %o5
be,pt %icc, 3f
sub %o2, %o5, %o2
-1: EX_LD(LOAD(ldx, %o1 + 0x00, %g1), NG4_retl_o2_plus_o5)
+1: EX_LD(LOAD(ldx, %o1 + 0x00, %g1), memcpy_retl_o2_plus_o5)
add %o1, 0x08, %o1
add %o0, 0x08, %o0
subcc %o5, 0x08, %o5
bne,pt %icc, 1b
- EX_ST(STORE(stx, %g1, %o0 - 0x08), NG4_retl_o2_plus_o5_plus_8)
+ EX_ST(STORE(stx, %g1, %o0 - 0x08), memcpy_retl_o2_plus_o5_plus_8)
3: brz,pt %o2, .Lexit
cmp %o2, 0x04
bl,pn %icc, .Ltiny
nop
- EX_LD(LOAD(lduw, %o1 + 0x00, %g1), NG4_retl_o2)
+ EX_LD(LOAD(lduw, %o1 + 0x00, %g1), memcpy_retl_o2)
add %o1, 0x04, %o1
add %o0, 0x04, %o0
subcc %o2, 0x04, %o2
bne,pn %icc, .Ltiny
- EX_ST(STORE(stw, %g1, %o0 - 0x04), NG4_retl_o2_plus_4)
+ EX_ST(STORE(stw, %g1, %o0 - 0x04), memcpy_retl_o2_plus_4)
ba,a,pt %icc, .Lexit
.Lmedium_unaligned:
/* First get dest 8 byte aligned. */
@@ -312,12 +312,12 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
brz,pt %g1, 2f
sub %o2, %g1, %o2
-1: EX_LD(LOAD(ldub, %o1 + 0x00, %g2), NG4_retl_o2_plus_g1)
+1: EX_LD(LOAD(ldub, %o1 + 0x00, %g2), memcpy_retl_o2_plus_g1)
add %o1, 1, %o1
subcc %g1, 1, %g1
add %o0, 1, %o0
bne,pt %icc, 1b
- EX_ST(STORE(stb, %g2, %o0 - 0x01), NG4_retl_o2_plus_g1_plus_1)
+ EX_ST(STORE(stb, %g2, %o0 - 0x01), memcpy_retl_o2_plus_g1_plus_1)
2:
and %o1, 0x7, %g1
brz,pn %g1, .Lmedium_noprefetch
@@ -325,16 +325,16 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
mov 64, %g2
sub %g2, %g1, %g2
andn %o1, 0x7, %o1
- EX_LD(LOAD(ldx, %o1 + 0x00, %o4), NG4_retl_o2)
+ EX_LD(LOAD(ldx, %o1 + 0x00, %o4), memcpy_retl_o2)
sllx %o4, %g1, %o4
andn %o2, 0x08 - 1, %o5
sub %o2, %o5, %o2
-1: EX_LD(LOAD(ldx, %o1 + 0x08, %g3), NG4_retl_o2_plus_o5)
+1: EX_LD(LOAD(ldx, %o1 + 0x08, %g3), memcpy_retl_o2_plus_o5)
add %o1, 0x08, %o1
subcc %o5, 0x08, %o5
srlx %g3, %g2, GLOBAL_SPARE
or GLOBAL_SPARE, %o4, GLOBAL_SPARE
- EX_ST(STORE(stx, GLOBAL_SPARE, %o0 + 0x00), NG4_retl_o2_plus_o5_plus_8)
+ EX_ST(STORE(stx, GLOBAL_SPARE, %o0 + 0x00), memcpy_retl_o2_plus_o5_plus_8)
add %o0, 0x08, %o0
bne,pt %icc, 1b
sllx %g3, %g1, %o4
@@ -345,17 +345,17 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
ba,pt %icc, .Lsmall_unaligned
.Ltiny:
- EX_LD(LOAD(ldub, %o1 + 0x00, %g1), NG4_retl_o2)
+ EX_LD(LOAD(ldub, %o1 + 0x00, %g1), memcpy_retl_o2)
subcc %o2, 1, %o2
be,pn %icc, .Lexit
- EX_ST(STORE(stb, %g1, %o0 + 0x00), NG4_retl_o2_plus_1)
- EX_LD(LOAD(ldub, %o1 + 0x01, %g1), NG4_retl_o2)
+ EX_ST(STORE(stb, %g1, %o0 + 0x00), memcpy_retl_o2_plus_1)
+ EX_LD(LOAD(ldub, %o1 + 0x01, %g1), memcpy_retl_o2)
subcc %o2, 1, %o2
be,pn %icc, .Lexit
- EX_ST(STORE(stb, %g1, %o0 + 0x01), NG4_retl_o2_plus_1)
- EX_LD(LOAD(ldub, %o1 + 0x02, %g1), NG4_retl_o2)
+ EX_ST(STORE(stb, %g1, %o0 + 0x01), memcpy_retl_o2_plus_1)
+ EX_LD(LOAD(ldub, %o1 + 0x02, %g1), memcpy_retl_o2)
ba,pt %icc, .Lexit
- EX_ST(STORE(stb, %g1, %o0 + 0x02), NG4_retl_o2)
+ EX_ST(STORE(stb, %g1, %o0 + 0x02), memcpy_retl_o2)
.Lsmall:
andcc %g2, 0x3, %g0
@@ -363,23 +363,23 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
andn %o2, 0x4 - 1, %o5
sub %o2, %o5, %o2
1:
- EX_LD(LOAD(lduw, %o1 + 0x00, %g1), NG4_retl_o2_plus_o5)
+ EX_LD(LOAD(lduw, %o1 + 0x00, %g1), memcpy_retl_o2_plus_o5)
add %o1, 0x04, %o1
subcc %o5, 0x04, %o5
add %o0, 0x04, %o0
bne,pt %icc, 1b
- EX_ST(STORE(stw, %g1, %o0 - 0x04), NG4_retl_o2_plus_o5_plus_4)
+ EX_ST(STORE(stw, %g1, %o0 - 0x04), memcpy_retl_o2_plus_o5_plus_4)
brz,pt %o2, .Lexit
nop
ba,a,pt %icc, .Ltiny
.Lsmall_unaligned:
-1: EX_LD(LOAD(ldub, %o1 + 0x00, %g1), NG4_retl_o2)
+1: EX_LD(LOAD(ldub, %o1 + 0x00, %g1), memcpy_retl_o2)
add %o1, 1, %o1
add %o0, 1, %o0
subcc %o2, 1, %o2
bne,pt %icc, 1b
- EX_ST(STORE(stb, %g1, %o0 - 0x01), NG4_retl_o2_plus_1)
+ EX_ST(STORE(stb, %g1, %o0 - 0x01), memcpy_retl_o2_plus_1)
ba,a,pt %icc, .Lexit
nop
.size FUNC_NAME, .-FUNC_NAME
--
1.7.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH v2 3/4] arch/sparc: Optimized memcpy, memset, copy_to_user, copy_from_user for M7/M8
2017-08-07 23:52 [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures Babu Moger
2017-08-07 23:52 ` [PATCH v2 1/4] arch/sparc: Separate the exception handlers from NG4memcpy Babu Moger
2017-08-07 23:52 ` [PATCH v2 2/4] arch/sparc: Rename exception handlers Babu Moger
@ 2017-08-07 23:52 ` Babu Moger
2017-08-07 23:52 ` [PATCH v2 4/4] arch/sparc: Add accurate exception reporting in M7memcpy Babu Moger
2017-08-10 21:38 ` [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures David Miller
4 siblings, 0 replies; 7+ messages in thread
From: Babu Moger @ 2017-08-07 23:52 UTC (permalink / raw)
To: davem; +Cc: sparclinux, linux-kernel, babu.moger
New algorithm that takes advantage of the M7/M8 block init store
ASI, ie, overlapping pipelines and miss buffer filling.
Full details in code comments.
Signed-off-by: Babu Moger <babu.moger@oracle.com>
---
arch/sparc/kernel/head_64.S | 16 +-
arch/sparc/lib/M7copy_from_user.S | 41 ++
arch/sparc/lib/M7copy_to_user.S | 51 ++
arch/sparc/lib/M7memcpy.S | 923 +++++++++++++++++++++++++++++++++++++
arch/sparc/lib/M7memset.S | 352 ++++++++++++++
arch/sparc/lib/M7patch.S | 51 ++
arch/sparc/lib/Makefile | 3 +
7 files changed, 1435 insertions(+), 2 deletions(-)
create mode 100644 arch/sparc/lib/M7copy_from_user.S
create mode 100644 arch/sparc/lib/M7copy_to_user.S
create mode 100644 arch/sparc/lib/M7memcpy.S
create mode 100644 arch/sparc/lib/M7memset.S
create mode 100644 arch/sparc/lib/M7patch.S
diff --git a/arch/sparc/kernel/head_64.S b/arch/sparc/kernel/head_64.S
index 78e0211..bf9a5ac 100644
--- a/arch/sparc/kernel/head_64.S
+++ b/arch/sparc/kernel/head_64.S
@@ -603,10 +603,10 @@ niagara_tlb_fixup:
be,pt %xcc, niagara4_patch
nop
cmp %g1, SUN4V_CHIP_SPARC_M7
- be,pt %xcc, niagara4_patch
+ be,pt %xcc, sparc_m7_patch
nop
cmp %g1, SUN4V_CHIP_SPARC_M8
- be,pt %xcc, niagara4_patch
+ be,pt %xcc, sparc_m7_patch
nop
cmp %g1, SUN4V_CHIP_SPARC_SN
be,pt %xcc, niagara4_patch
@@ -621,6 +621,18 @@ niagara_tlb_fixup:
ba,a,pt %xcc, 80f
nop
+
+sparc_m7_patch:
+ call m7_patch_copyops
+ nop
+ call m7_patch_bzero
+ nop
+ call m7_patch_pageops
+ nop
+
+ ba,a,pt %xcc, 80f
+ nop
+
niagara4_patch:
call niagara4_patch_copyops
nop
diff --git a/arch/sparc/lib/M7copy_from_user.S b/arch/sparc/lib/M7copy_from_user.S
new file mode 100644
index 0000000..d0689d7
--- /dev/null
+++ b/arch/sparc/lib/M7copy_from_user.S
@@ -0,0 +1,41 @@
+/*
+ * M7copy_from_user.S: SPARC M7 optimized copy from userspace.
+ *
+ * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
+ */
+
+
+#define EX_LD(x) \
+98: x; \
+ .section __ex_table,"a"; \
+ .align 4; \
+ .word 98b, __restore_asi; \
+ .text; \
+ .align 4;
+
+#define EX_LD_FP(x) \
+98: x; \
+ .section __ex_table,"a"; \
+ .align 4; \
+ .word 98b, __restore_asi_fp; \
+ .text; \
+ .align 4;
+
+
+#ifndef ASI_AIUS
+#define ASI_AIUS 0x11
+#endif
+
+#define FUNC_NAME M7copy_from_user
+#define LOAD(type,addr,dest) type##a [addr] %asi, dest
+#define EX_RETVAL(x) 0
+
+#ifdef __KERNEL__
+#define PREAMBLE \
+ rd %asi, %g1; \
+ cmp %g1, ASI_AIUS; \
+ bne,pn %icc, raw_copy_in_user; \
+ nop
+#endif
+
+#include "M7memcpy.S"
diff --git a/arch/sparc/lib/M7copy_to_user.S b/arch/sparc/lib/M7copy_to_user.S
new file mode 100644
index 0000000..d3be132
--- /dev/null
+++ b/arch/sparc/lib/M7copy_to_user.S
@@ -0,0 +1,51 @@
+/*
+ * M7copy_to_user.S: SPARC M7 optimized copy to userspace.
+ *
+ * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
+ */
+
+
+#define EX_ST(x) \
+98: x; \
+ .section __ex_table,"a"; \
+ .align 4; \
+ .word 98b, __restore_asi; \
+ .text; \
+ .align 4;
+
+#define EX_ST_FP(x) \
+98: x; \
+ .section __ex_table,"a"; \
+ .align 4; \
+ .word 98b, __restore_asi_fp; \
+ .text; \
+ .align 4;
+
+
+#ifndef ASI_AIUS
+#define ASI_AIUS 0x11
+#endif
+
+#ifndef ASI_BLK_INIT_QUAD_LDD_AIUS
+#define ASI_BLK_INIT_QUAD_LDD_AIUS 0x23
+#endif
+
+#define FUNC_NAME M7copy_to_user
+#define STORE(type,src,addr) type##a src, [addr] %asi
+#define STORE_ASI ASI_BLK_INIT_QUAD_LDD_AIUS
+#define STORE_MRU_ASI ASI_ST_BLKINIT_MRU_S
+#define EX_RETVAL(x) 0
+
+#ifdef __KERNEL__
+ /* Writing to %asi is _expensive_ so we hardcode it.
+ * Reading %asi to check for KERNEL_DS is comparatively
+ * cheap.
+ */
+#define PREAMBLE \
+ rd %asi, %g1; \
+ cmp %g1, ASI_AIUS; \
+ bne,pn %icc, raw_copy_in_user; \
+ nop
+#endif
+
+#include "M7memcpy.S"
diff --git a/arch/sparc/lib/M7memcpy.S b/arch/sparc/lib/M7memcpy.S
new file mode 100644
index 0000000..0a0421d
--- /dev/null
+++ b/arch/sparc/lib/M7memcpy.S
@@ -0,0 +1,923 @@
+/*
+ * M7memcpy: Optimized SPARC M7 memcpy
+ *
+ * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
+ */
+
+ .file "M7memcpy.S"
+
+/*
+ * memcpy(s1, s2, len)
+ *
+ * Copy s2 to s1, always copy n bytes.
+ * Note: this C code does not work for overlapped copies.
+ *
+ * Fast assembler language version of the following C-program for memcpy
+ * which represents the `standard' for the C-library.
+ *
+ * void *
+ * memcpy(void *s, const void *s0, size_t n)
+ * {
+ * if (n != 0) {
+ * char *s1 = s;
+ * const char *s2 = s0;
+ * do {
+ * *s1++ = *s2++;
+ * } while (--n != 0);
+ * }
+ * return (s);
+ * }
+ *
+ *
+ * SPARC T7/M7 Flow :
+ *
+ * if (count < SMALL_MAX) {
+ * if count < SHORTCOPY (SHORTCOPY=3)
+ * copy bytes; exit with dst addr
+ * if src & dst aligned on word boundary but not long word boundary,
+ * copy with ldw/stw; branch to finish_up
+ * if src & dst aligned on long word boundary
+ * copy with ldx/stx; branch to finish_up
+ * if src & dst not aligned and length <= SHORTCHECK (SHORTCHECK=14)
+ * copy bytes; exit with dst addr
+ * move enough bytes to get src to word boundary
+ * if dst now on word boundary
+ * move_words:
+ * copy words; branch to finish_up
+ * if dst now on half word boundary
+ * load words, shift half words, store words; branch to finish_up
+ * if dst on byte 1
+ * load words, shift 3 bytes, store words; branch to finish_up
+ * if dst on byte 3
+ * load words, shift 1 byte, store words; branch to finish_up
+ * finish_up:
+ * copy bytes; exit with dst addr
+ * } else { More than SMALL_MAX bytes
+ * move bytes until dst is on long word boundary
+ * if( src is on long word boundary ) {
+ * if (count < MED_MAX) {
+ * finish_long: src/dst aligned on 8 bytes
+ * copy with ldx/stx in 8-way unrolled loop;
+ * copy final 0-63 bytes; exit with dst addr
+ * } else { src/dst aligned; count > MED_MAX
+ * align dst on 64 byte boundary; for main data movement:
+ * prefetch src data to L2 cache; let HW prefetch move data to L1 cache
+ * Use BIS (block initializing store) to avoid copying store cache
+ * lines from memory. But pre-store first element of each cache line
+ * ST_CHUNK lines in advance of the rest of that cache line. That
+ * gives time for replacement cache lines to be written back without
+ * excess STQ and Miss Buffer filling. Repeat until near the end,
+ * then finish up storing before going to finish_long.
+ * }
+ * } else { src/dst not aligned on 8 bytes
+ * if src is word aligned and count < MED_WMAX
+ * move words in 8-way unrolled loop
+ * move final 0-31 bytes; exit with dst addr
+ * if count < MED_UMAX
+ * use alignaddr/faligndata combined with ldd/std in 8-way
+ * unrolled loop to move data.
+ * go to unalign_done
+ * else
+ * setup alignaddr for faligndata instructions
+ * align dst on 64 byte boundary; prefetch src data to L1 cache
+ * loadx8, falign, block-store, prefetch loop
+ * (only use block-init-store when src/dst on 8 byte boundaries.)
+ * unalign_done:
+ * move remaining bytes for unaligned cases. exit with dst addr.
+ * }
+ *
+ */
+
+#include <asm/visasm.h>
+#include <asm/asi.h>
+
+#if !defined(EX_LD) && !defined(EX_ST)
+#define NON_USER_COPY
+#endif
+
+#ifndef EX_LD
+#define EX_LD(x) x
+#endif
+#ifndef EX_LD_FP
+#define EX_LD_FP(x) x
+#endif
+
+#ifndef EX_ST
+#define EX_ST(x) x
+#endif
+#ifndef EX_ST_FP
+#define EX_ST_FP(x) x
+#endif
+
+#ifndef EX_RETVAL
+#define EX_RETVAL(x) x
+#endif
+
+#ifndef LOAD
+#define LOAD(type,addr,dest) type [addr], dest
+#endif
+
+#ifndef STORE
+#define STORE(type,src,addr) type src, [addr]
+#endif
+
+/*
+ * ASI_BLK_INIT_QUAD_LDD_P/ASI_BLK_INIT_QUAD_LDD_S marks the cache
+ * line as "least recently used" which means if many threads are
+ * active, it has a high probability of being pushed out of the cache
+ * between the first initializing store and the final stores.
+ * Thus, we use ASI_ST_BLKINIT_MRU_P/ASI_ST_BLKINIT_MRU_S which
+ * marks the cache line as "most recently used" for all
+ * but the last cache line
+ */
+#ifndef STORE_ASI
+#ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA
+#define STORE_ASI ASI_BLK_INIT_QUAD_LDD_P
+#else
+#define STORE_ASI 0x80 /* ASI_P */
+#endif
+#endif
+
+#ifndef STORE_MRU_ASI
+#ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA
+#define STORE_MRU_ASI ASI_ST_BLKINIT_MRU_P
+#else
+#define STORE_MRU_ASI 0x80 /* ASI_P */
+#endif
+#endif
+
+#ifndef STORE_INIT
+#define STORE_INIT(src,addr) stxa src, [addr] STORE_ASI
+#endif
+
+#ifndef STORE_INIT_MRU
+#define STORE_INIT_MRU(src,addr) stxa src, [addr] STORE_MRU_ASI
+#endif
+
+#ifndef FUNC_NAME
+#define FUNC_NAME M7memcpy
+#endif
+
+#ifndef PREAMBLE
+#define PREAMBLE
+#endif
+
+#define BLOCK_SIZE 64
+#define SHORTCOPY 3
+#define SHORTCHECK 14
+#define SHORT_LONG 64 /* max copy for short longword-aligned case */
+ /* must be at least 64 */
+#define SMALL_MAX 128
+#define MED_UMAX 1024 /* max copy for medium un-aligned case */
+#define MED_WMAX 1024 /* max copy for medium word-aligned case */
+#define MED_MAX 1024 /* max copy for medium longword-aligned case */
+#define ST_CHUNK 24 /* ST_CHUNK - block of values for BIS Store */
+#define ALIGN_PRE 24 /* distance for aligned prefetch loop */
+
+ .register %g2,#scratch
+
+ .section ".text"
+ .global FUNC_NAME
+ .type FUNC_NAME, #function
+ .align 16
+FUNC_NAME:
+ srlx %o2, 31, %g2
+ cmp %g2, 0
+ tne %xcc, 5
+ PREAMBLE
+ mov %o0, %g1 ! save %o0
+ brz,pn %o2, .Lsmallx
+ cmp %o2, 3
+ ble,pn %icc, .Ltiny_cp
+ cmp %o2, 19
+ ble,pn %icc, .Lsmall_cp
+ or %o0, %o1, %g2
+ cmp %o2, SMALL_MAX
+ bl,pn %icc, .Lmedium_cp
+ nop
+
+.Lmedium:
+ neg %o0, %o5
+ andcc %o5, 7, %o5 ! bytes till DST 8 byte aligned
+ brz,pt %o5, .Ldst_aligned_on_8
+
+ ! %o5 has the bytes to be written in partial store.
+ sub %o2, %o5, %o2
+ sub %o1, %o0, %o1 ! %o1 gets the difference
+7: ! dst aligning loop
+ add %o1, %o0, %o4
+ EX_LD(LOAD(ldub, %o4, %o4)) ! load one byte
+ subcc %o5, 1, %o5
+ EX_ST(STORE(stb, %o4, %o0))
+ bgu,pt %ncc, 7b
+ add %o0, 1, %o0 ! advance dst
+ add %o1, %o0, %o1 ! restore %o1
+.Ldst_aligned_on_8:
+ andcc %o1, 7, %o5
+ brnz,pt %o5, .Lsrc_dst_unaligned_on_8
+ nop
+
+.Lsrc_dst_aligned_on_8:
+ ! check if we are copying MED_MAX or more bytes
+ set MED_MAX, %o3
+ cmp %o2, %o3 ! limit to store buffer size
+ bgu,pn %ncc, .Llarge_align8_copy
+ nop
+
+/*
+ * Special case for handling when src and dest are both long word aligned
+ * and total data to move is less than MED_MAX bytes
+ */
+.Lmedlong:
+ subcc %o2, 63, %o2 ! adjust length to allow cc test
+ ble,pn %ncc, .Lmedl63 ! skip big loop if less than 64 bytes
+ nop
+.Lmedl64:
+ EX_LD(LOAD(ldx, %o1, %o4)) ! load
+ subcc %o2, 64, %o2 ! decrement length count
+ EX_ST(STORE(stx, %o4, %o0)) ! and store
+ EX_LD(LOAD(ldx, %o1+8, %o3)) ! a block of 64 bytes
+ EX_ST(STORE(stx, %o3, %o0+8))
+ EX_LD(LOAD(ldx, %o1+16, %o4))
+ EX_ST(STORE(stx, %o4, %o0+16))
+ EX_LD(LOAD(ldx, %o1+24, %o3))
+ EX_ST(STORE(stx, %o3, %o0+24))
+ EX_LD(LOAD(ldx, %o1+32, %o4)) ! load
+ EX_ST(STORE(stx, %o4, %o0+32)) ! and store
+ EX_LD(LOAD(ldx, %o1+40, %o3)) ! a block of 64 bytes
+ add %o1, 64, %o1 ! increase src ptr by 64
+ EX_ST(STORE(stx, %o3, %o0+40))
+ EX_LD(LOAD(ldx, %o1-16, %o4))
+ add %o0, 64, %o0 ! increase dst ptr by 64
+ EX_ST(STORE(stx, %o4, %o0-16))
+ EX_LD(LOAD(ldx, %o1-8, %o3))
+ bgu,pt %ncc, .Lmedl64 ! repeat if at least 64 bytes left
+ EX_ST(STORE(stx, %o3, %o0-8))
+.Lmedl63:
+ addcc %o2, 32, %o2 ! adjust remaining count
+ ble,pt %ncc, .Lmedl31 ! to skip if 31 or fewer bytes left
+ nop
+ EX_LD(LOAD(ldx, %o1, %o4)) ! load
+ sub %o2, 32, %o2 ! decrement length count
+ EX_ST(STORE(stx, %o4, %o0)) ! and store
+ EX_LD(LOAD(ldx, %o1+8, %o3)) ! a block of 32 bytes
+ add %o1, 32, %o1 ! increase src ptr by 32
+ EX_ST(STORE(stx, %o3, %o0+8))
+ EX_LD(LOAD(ldx, %o1-16, %o4))
+ add %o0, 32, %o0 ! increase dst ptr by 32
+ EX_ST(STORE(stx, %o4, %o0-16))
+ EX_LD(LOAD(ldx, %o1-8, %o3))
+ EX_ST(STORE(stx, %o3, %o0-8))
+.Lmedl31:
+ addcc %o2, 16, %o2 ! adjust remaining count
+ ble,pt %ncc, .Lmedl15 ! skip if 15 or fewer bytes left
+ nop !
+ EX_LD(LOAD(ldx, %o1, %o4))
+ add %o1, 16, %o1 ! increase src ptr by 16
+ EX_ST(STORE(stx, %o4, %o0))
+ sub %o2, 16, %o2 ! decrease count by 16
+ EX_LD(LOAD(ldx, %o1-8, %o3))
+ add %o0, 16, %o0 ! increase dst ptr by 16
+ EX_ST(STORE(stx, %o3, %o0-8))
+.Lmedl15:
+ addcc %o2, 15, %o2 ! restore count
+ bz,pt %ncc, .Lsmallx ! exit if finished
+ cmp %o2, 8
+ blt,pt %ncc, .Lmedw7 ! skip if 7 or fewer bytes left
+ tst %o2
+ EX_LD(LOAD(ldx, %o1, %o4)) ! load 8 bytes
+ add %o1, 8, %o1 ! increase src ptr by 8
+ add %o0, 8, %o0 ! increase dst ptr by 8
+ subcc %o2, 8, %o2 ! decrease count by 8
+ bnz,pn %ncc, .Lmedw7
+ EX_ST(STORE(stx, %o4, %o0-8)) ! and store 8 bytes
+ retl
+ mov EX_RETVAL(%g1), %o0 ! restore %o0
+
+ .align 16
+.Lsrc_dst_unaligned_on_8:
+ ! DST is 8-byte aligned, src is not
+2:
+ andcc %o1, 0x3, %o5 ! test word alignment
+ bnz,pt %ncc, .Lunalignsetup ! branch to skip if not word aligned
+ nop
+
+/*
+ * Handle all cases where src and dest are aligned on word
+ * boundaries. Use unrolled loops for better performance.
+ * This option wins over standard large data move when
+ * source and destination is in cache for.Lmedium
+ * to short data moves.
+ */
+ set MED_WMAX, %o3
+ cmp %o2, %o3 ! limit to store buffer size
+ bge,pt %ncc, .Lunalignrejoin ! otherwise rejoin main loop
+ nop
+
+ subcc %o2, 31, %o2 ! adjust length to allow cc test
+ ! for end of loop
+ ble,pt %ncc, .Lmedw31 ! skip big loop if less than 16
+.Lmedw32:
+ EX_LD(LOAD(ld, %o1, %o4)) ! move a block of 32 bytes
+ sllx %o4, 32, %o5
+ EX_LD(LOAD(ld, %o1+4, %o4))
+ or %o4, %o5, %o5
+ EX_ST(STORE(stx, %o5, %o0))
+ subcc %o2, 32, %o2 ! decrement length count
+ EX_LD(LOAD(ld, %o1+8, %o4))
+ sllx %o4, 32, %o5
+ EX_LD(LOAD(ld, %o1+12, %o4))
+ or %o4, %o5, %o5
+ EX_ST(STORE(stx, %o5, %o0+8))
+ add %o1, 32, %o1 ! increase src ptr by 32
+ EX_LD(LOAD(ld, %o1-16, %o4))
+ sllx %o4, 32, %o5
+ EX_LD(LOAD(ld, %o1-12, %o4))
+ or %o4, %o5, %o5
+ EX_ST(STORE(stx, %o5, %o0+16))
+ add %o0, 32, %o0 ! increase dst ptr by 32
+ EX_LD(LOAD(ld, %o1-8, %o4))
+ sllx %o4, 32, %o5
+ EX_LD(LOAD(ld, %o1-4, %o4))
+ or %o4, %o5, %o5
+ bgu,pt %ncc, .Lmedw32 ! repeat if at least 32 bytes left
+ EX_ST(STORE(stx, %o5, %o0-8))
+.Lmedw31:
+ addcc %o2, 31, %o2 ! restore count
+
+ bz,pt %ncc, .Lsmallx ! exit if finished
+ nop
+ cmp %o2, 16
+ blt,pt %ncc, .Lmedw15
+ nop
+ EX_LD(LOAD(ld, %o1, %o4)) ! move a block of 16 bytes
+ sllx %o4, 32, %o5
+ subcc %o2, 16, %o2 ! decrement length count
+ EX_LD(LOAD(ld, %o1+4, %o4))
+ or %o4, %o5, %o5
+ EX_ST(STORE(stx, %o5, %o0))
+ add %o1, 16, %o1 ! increase src ptr by 16
+ EX_LD(LOAD(ld, %o1-8, %o4))
+ add %o0, 16, %o0 ! increase dst ptr by 16
+ sllx %o4, 32, %o5
+ EX_LD(LOAD(ld, %o1-4, %o4))
+ or %o4, %o5, %o5
+ EX_ST(STORE(stx, %o5, %o0-8))
+.Lmedw15:
+ bz,pt %ncc, .Lsmallx ! exit if finished
+ cmp %o2, 8
+ blt,pn %ncc, .Lmedw7 ! skip if 7 or fewer bytes left
+ tst %o2
+ EX_LD(LOAD(ld, %o1, %o4)) ! load 4 bytes
+ subcc %o2, 8, %o2 ! decrease count by 8
+ EX_ST(STORE(stw, %o4, %o0)) ! and store 4 bytes
+ add %o1, 8, %o1 ! increase src ptr by 8
+ EX_LD(LOAD(ld, %o1-4, %o3)) ! load 4 bytes
+ add %o0, 8, %o0 ! increase dst ptr by 8
+ EX_ST(STORE(stw, %o3, %o0-4)) ! and store 4 bytes
+ bz,pt %ncc, .Lsmallx ! exit if finished
+.Lmedw7: ! count is ge 1, less than 8
+ cmp %o2, 4 ! check for 4 bytes left
+ blt,pn %ncc, .Lsmallleft3 ! skip if 3 or fewer bytes left
+ nop !
+ EX_LD(LOAD(ld, %o1, %o4)) ! load 4 bytes
+ add %o1, 4, %o1 ! increase src ptr by 4
+ add %o0, 4, %o0 ! increase dst ptr by 4
+ subcc %o2, 4, %o2 ! decrease count by 4
+ bnz .Lsmallleft3
+ EX_ST(STORE(stw, %o4, %o0-4))! and store 4 bytes
+ retl
+ mov EX_RETVAL(%g1), %o0
+
+ .align 16
+.Llarge_align8_copy: ! Src and dst share 8 byte alignment
+ ! align dst to 64 byte boundary
+ andcc %o0, 0x3f, %o3 ! %o3 == 0 means dst is 64 byte aligned
+ brz,pn %o3, .Laligned_to_64
+ andcc %o0, 8, %o3 ! odd long words to move?
+ brz,pt %o3, .Laligned_to_16
+ nop
+ EX_LD(LOAD(ldx, %o1, %o4))
+ sub %o2, 8, %o2
+ add %o1, 8, %o1 ! increment src ptr
+ add %o0, 8, %o0 ! increment dst ptr
+ EX_ST(STORE(stx, %o4, %o0-8))
+.Laligned_to_16:
+ andcc %o0, 16, %o3 ! pair of long words to move?
+ brz,pt %o3, .Laligned_to_32
+ nop
+ EX_LD(LOAD(ldx, %o1, %o4))
+ sub %o2, 16, %o2
+ EX_ST(STORE(stx, %o4, %o0))
+ add %o1, 16, %o1 ! increment src ptr
+ EX_LD(LOAD(ldx, %o1-8, %o4))
+ add %o0, 16, %o0 ! increment dst ptr
+ EX_ST(STORE(stx, %o4, %o0-8))
+.Laligned_to_32:
+ andcc %o0, 32, %o3 ! four long words to move?
+ brz,pt %o3, .Laligned_to_64
+ nop
+ EX_LD(LOAD(ldx, %o1, %o4))
+ sub %o2, 32, %o2
+ EX_ST(STORE(stx, %o4, %o0))
+ EX_LD(LOAD(ldx, %o1+8, %o4))
+ EX_ST(STORE(stx, %o4, %o0+8))
+ EX_LD(LOAD(ldx, %o1+16, %o4))
+ EX_ST(STORE(stx, %o4, %o0+16))
+ add %o1, 32, %o1 ! increment src ptr
+ EX_LD(LOAD(ldx, %o1-8, %o4))
+ add %o0, 32, %o0 ! increment dst ptr
+ EX_ST(STORE(stx, %o4, %o0-8))
+.Laligned_to_64:
+!
+! Using block init store (BIS) instructions to avoid fetching cache
+! lines from memory. Use ST_CHUNK stores to first element of each cache
+! line (similar to prefetching) to avoid overfilling STQ or miss buffers.
+! Gives existing cache lines time to be moved out of L1/L2/L3 cache.
+! Initial stores using MRU version of BIS to keep cache line in
+! cache until we are ready to store final element of cache line.
+! Then store last element using the LRU version of BIS.
+!
+ andn %o2, 0x3f, %o5 ! %o5 is multiple of block size
+ and %o2, 0x3f, %o2 ! residue bytes in %o2
+!
+! We use STORE_MRU_ASI for the first seven stores to each cache line
+! followed by STORE_ASI (mark as LRU) for the last store. That
+! mixed approach reduces the probability that the cache line is removed
+! before we finish setting it, while minimizing the effects on
+! other cached values during a large memcpy
+!
+! ST_CHUNK batches up initial BIS operations for several cache lines
+! to allow multiple requests to not be blocked by overflowing the
+! the store miss buffer. Then the matching stores for all those
+! BIS operations are executed.
+!
+
+ sub %o0, 8, %o0 ! adjust %o0 for ASI alignment
+.Lalign_loop:
+ cmp %o5, ST_CHUNK*64
+ blu,pt %ncc, .Lalign_loop_fin
+ mov ST_CHUNK,%o3
+.Lalign_loop_start:
+ prefetch [%o1 + (ALIGN_PRE * BLOCK_SIZE)], 21
+ subcc %o3, 1, %o3
+ EX_LD(LOAD(ldx, %o1, %o4))
+ add %o1, 64, %o1
+ add %o0, 8, %o0
+ EX_ST(STORE_INIT_MRU(%o4, %o0))
+ bgu %ncc,.Lalign_loop_start
+ add %o0, 56, %o0
+
+ mov ST_CHUNK,%o3
+ sllx %o3, 6, %o4 ! ST_CHUNK*64
+ sub %o1, %o4, %o1 ! reset %o1
+ sub %o0, %o4, %o0 ! reset %o0
+
+.Lalign_loop_rest:
+ EX_LD(LOAD(ldx, %o1+8, %o4))
+ add %o0, 16, %o0
+ EX_ST(STORE_INIT_MRU(%o4, %o0))
+ EX_LD(LOAD(ldx, %o1+16, %o4))
+ add %o0, 8, %o0
+ EX_ST(STORE_INIT_MRU(%o4, %o0))
+ subcc %o3, 1, %o3
+ EX_LD(LOAD(ldx, %o1+24, %o4))
+ add %o0, 8, %o0
+ EX_ST(STORE_INIT_MRU(%o4, %o0))
+ EX_LD(LOAD(ldx, %o1+32, %o4))
+ add %o0, 8, %o0
+ EX_ST(STORE_INIT_MRU(%o4, %o0))
+ EX_LD(LOAD(ldx, %o1+40, %o4))
+ add %o0, 8, %o0
+ EX_ST(STORE_INIT_MRU(%o4, %o0))
+ EX_LD(LOAD(ldx, %o1+48, %o4))
+ add %o1, 64, %o1
+ add %o0, 8, %o0
+ EX_ST(STORE_INIT_MRU(%o4, %o0))
+ add %o0, 8, %o0
+ EX_LD(LOAD(ldx, %o1-8, %o4))
+ sub %o5, 64, %o5
+ bgu %ncc,.Lalign_loop_rest
+ ! mark cache line as LRU
+ EX_ST(STORE_INIT(%o4, %o0))
+
+ cmp %o5, ST_CHUNK*64
+ bgu,pt %ncc, .Lalign_loop_start
+ mov ST_CHUNK,%o3
+
+ cmp %o5, 0
+ beq .Lalign_done
+ nop
+.Lalign_loop_fin:
+ EX_LD(LOAD(ldx, %o1, %o4))
+ EX_ST(STORE(stx, %o4, %o0+8))
+ EX_LD(LOAD(ldx, %o1+8, %o4))
+ EX_ST(STORE(stx, %o4, %o0+8+8))
+ EX_LD(LOAD(ldx, %o1+16, %o4))
+ EX_ST(STORE(stx, %o4, %o0+8+16))
+ subcc %o5, 64, %o5
+ EX_LD(LOAD(ldx, %o1+24, %o4))
+ EX_ST(STORE(stx, %o4, %o0+8+24))
+ EX_LD(LOAD(ldx, %o1+32, %o4))
+ EX_ST(STORE(stx, %o4, %o0+8+32))
+ EX_LD(LOAD(ldx, %o1+40, %o4))
+ EX_ST(STORE(stx, %o4, %o0+8+40))
+ EX_LD(LOAD(ldx, %o1+48, %o4))
+ add %o1, 64, %o1
+ EX_ST(STORE(stx, %o4, %o0+8+48))
+ add %o0, 64, %o0
+ EX_LD(LOAD(ldx, %o1-8, %o4))
+ bgu %ncc,.Lalign_loop_fin
+ EX_ST(STORE(stx, %o4, %o0))
+
+.Lalign_done:
+ add %o0, 8, %o0 ! restore %o0 from ASI alignment
+ membar #StoreStore
+ sub %o2, 63, %o2 ! adjust length to allow cc test
+ ba .Lmedl63 ! in .Lmedl63
+ nop
+
+ .align 16
+ ! Dst is on 8 byte boundary; src is not; remaining count > SMALL_MAX
+.Lunalignsetup:
+.Lunalignrejoin:
+ mov %g1, %o3 ! save %g1 as VISEntryHalf clobbers it
+#ifdef NON_USER_COPY
+ VISEntryHalfFast(.Lmedium_vis_entry_fail_cp)
+#else
+ VISEntryHalf
+#endif
+ mov %o3, %g1 ! restore %g1
+
+ set MED_UMAX, %o3
+ cmp %o2, %o3 ! check for.Lmedium unaligned limit
+ bge,pt %ncc,.Lunalign_large
+ prefetch [%o1 + (4 * BLOCK_SIZE)], 20
+ andn %o2, 0x3f, %o5 ! %o5 is multiple of block size
+ and %o2, 0x3f, %o2 ! residue bytes in %o2
+ cmp %o2, 8 ! Insure we do not load beyond
+ bgt .Lunalign_adjust ! end of source buffer
+ andn %o1, 0x7, %o4 ! %o4 has long word aligned src address
+ add %o2, 64, %o2 ! adjust to leave loop
+ sub %o5, 64, %o5 ! early if necessary
+.Lunalign_adjust:
+ alignaddr %o1, %g0, %g0 ! generate %gsr
+ add %o1, %o5, %o1 ! advance %o1 to after blocks
+ EX_LD_FP(LOAD(ldd, %o4, %f0))
+.Lunalign_loop:
+ EX_LD_FP(LOAD(ldd, %o4+8, %f2))
+ faligndata %f0, %f2, %f16
+ EX_LD_FP(LOAD(ldd, %o4+16, %f4))
+ subcc %o5, BLOCK_SIZE, %o5
+ EX_ST_FP(STORE(std, %f16, %o0))
+ faligndata %f2, %f4, %f18
+ EX_LD_FP(LOAD(ldd, %o4+24, %f6))
+ EX_ST_FP(STORE(std, %f18, %o0+8))
+ faligndata %f4, %f6, %f20
+ EX_LD_FP(LOAD(ldd, %o4+32, %f8))
+ EX_ST_FP(STORE(std, %f20, %o0+16))
+ faligndata %f6, %f8, %f22
+ EX_LD_FP(LOAD(ldd, %o4+40, %f10))
+ EX_ST_FP(STORE(std, %f22, %o0+24))
+ faligndata %f8, %f10, %f24
+ EX_LD_FP(LOAD(ldd, %o4+48, %f12))
+ EX_ST_FP(STORE(std, %f24, %o0+32))
+ faligndata %f10, %f12, %f26
+ EX_LD_FP(LOAD(ldd, %o4+56, %f14))
+ add %o4, BLOCK_SIZE, %o4
+ EX_ST_FP(STORE(std, %f26, %o0+40))
+ faligndata %f12, %f14, %f28
+ EX_LD_FP(LOAD(ldd, %o4, %f0))
+ EX_ST_FP(STORE(std, %f28, %o0+48))
+ faligndata %f14, %f0, %f30
+ EX_ST_FP(STORE(std, %f30, %o0+56))
+ add %o0, BLOCK_SIZE, %o0
+ bgu,pt %ncc, .Lunalign_loop
+ prefetch [%o4 + (5 * BLOCK_SIZE)], 20
+ ba .Lunalign_done
+ nop
+
+.Lunalign_large:
+ andcc %o0, 0x3f, %o3 ! is dst 64-byte block aligned?
+ bz %ncc, .Lunalignsrc
+ sub %o3, 64, %o3 ! %o3 will be multiple of 8
+ neg %o3 ! bytes until dest is 64 byte aligned
+ sub %o2, %o3, %o2 ! update cnt with bytes to be moved
+ ! Move bytes according to source alignment
+ andcc %o1, 0x1, %o5
+ bnz %ncc, .Lunalignbyte ! check for byte alignment
+ nop
+ andcc %o1, 2, %o5 ! check for half word alignment
+ bnz %ncc, .Lunalignhalf
+ nop
+ ! Src is word aligned
+.Lunalignword:
+ EX_LD_FP(LOAD(ld, %o1, %o4)) ! load 4 bytes
+ add %o1, 8, %o1 ! increase src ptr by 8
+ EX_ST_FP(STORE(stw, %o4, %o0)) ! and store 4 bytes
+ subcc %o3, 8, %o3 ! decrease count by 8
+ EX_LD_FP(LOAD(ld, %o1-4, %o4)) ! load 4 bytes
+ add %o0, 8, %o0 ! increase dst ptr by 8
+ bnz %ncc, .Lunalignword
+ EX_ST_FP(STORE(stw, %o4, %o0-4))! and store 4 bytes
+ ba .Lunalignsrc
+ nop
+
+ ! Src is half-word aligned
+.Lunalignhalf:
+ EX_LD_FP(LOAD(lduh, %o1, %o4)) ! load 2 bytes
+ sllx %o4, 32, %o5 ! shift left
+ EX_LD_FP(LOAD(lduw, %o1+2, %o4))
+ or %o4, %o5, %o5
+ sllx %o5, 16, %o5
+ EX_LD_FP(LOAD(lduh, %o1+6, %o4))
+ or %o4, %o5, %o5
+ EX_ST_FP(STORE(stx, %o5, %o0))
+ add %o1, 8, %o1
+ subcc %o3, 8, %o3
+ bnz %ncc, .Lunalignhalf
+ add %o0, 8, %o0
+ ba .Lunalignsrc
+ nop
+
+ ! Src is Byte aligned
+.Lunalignbyte:
+ sub %o0, %o1, %o0 ! share pointer advance
+.Lunalignbyte_loop:
+ EX_LD_FP(LOAD(ldub, %o1, %o4))
+ sllx %o4, 56, %o5
+ EX_LD_FP(LOAD(lduh, %o1+1, %o4))
+ sllx %o4, 40, %o4
+ or %o4, %o5, %o5
+ EX_LD_FP(LOAD(lduh, %o1+3, %o4))
+ sllx %o4, 24, %o4
+ or %o4, %o5, %o5
+ EX_LD_FP(LOAD(lduh, %o1+5, %o4))
+ sllx %o4, 8, %o4
+ or %o4, %o5, %o5
+ EX_LD_FP(LOAD(ldub, %o1+7, %o4))
+ or %o4, %o5, %o5
+ add %o0, %o1, %o0
+ EX_ST_FP(STORE(stx, %o5, %o0))
+ sub %o0, %o1, %o0
+ subcc %o3, 8, %o3
+ bnz %ncc, .Lunalignbyte_loop
+ add %o1, 8, %o1
+ add %o0,%o1, %o0 ! restore pointer
+
+ ! Destination is now block (64 byte aligned)
+.Lunalignsrc:
+ andn %o2, 0x3f, %o5 ! %o5 is multiple of block size
+ and %o2, 0x3f, %o2 ! residue bytes in %o2
+ add %o2, 64, %o2 ! Insure we do not load beyond
+ sub %o5, 64, %o5 ! end of source buffer
+
+ andn %o1, 0x7, %o4 ! %o4 has long word aligned src address
+ alignaddr %o1, %g0, %g0 ! generate %gsr
+ add %o1, %o5, %o1 ! advance %o1 to after blocks
+
+ EX_LD_FP(LOAD(ldd, %o4, %f14))
+ add %o4, 8, %o4
+.Lunalign_sloop:
+ EX_LD_FP(LOAD(ldd, %o4, %f16))
+ faligndata %f14, %f16, %f0
+ EX_LD_FP(LOAD(ldd, %o4+8, %f18))
+ faligndata %f16, %f18, %f2
+ EX_LD_FP(LOAD(ldd, %o4+16, %f20))
+ faligndata %f18, %f20, %f4
+ EX_ST_FP(STORE(std, %f0, %o0))
+ subcc %o5, 64, %o5
+ EX_LD_FP(LOAD(ldd, %o4+24, %f22))
+ faligndata %f20, %f22, %f6
+ EX_ST_FP(STORE(std, %f2, %o0+8))
+ EX_LD_FP(LOAD(ldd, %o4+32, %f24))
+ faligndata %f22, %f24, %f8
+ EX_ST_FP(STORE(std, %f4, %o0+16))
+ EX_LD_FP(LOAD(ldd, %o4+40, %f26))
+ faligndata %f24, %f26, %f10
+ EX_ST_FP(STORE(std, %f6, %o0+24))
+ EX_LD_FP(LOAD(ldd, %o4+48, %f28))
+ faligndata %f26, %f28, %f12
+ EX_ST_FP(STORE(std, %f8, %o0+32))
+ add %o4, 64, %o4
+ EX_LD_FP(LOAD(ldd, %o4-8, %f30))
+ faligndata %f28, %f30, %f14
+ EX_ST_FP(STORE(std, %f10, %o0+40))
+ EX_ST_FP(STORE(std, %f12, %o0+48))
+ add %o0, 64, %o0
+ EX_ST_FP(STORE(std, %f14, %o0-8))
+ fsrc2 %f30, %f14
+ bgu,pt %ncc, .Lunalign_sloop
+ prefetch [%o4 + (8 * BLOCK_SIZE)], 20
+
+.Lunalign_done:
+ ! Handle trailing bytes, 64 to 127
+ ! Dest long word aligned, Src not long word aligned
+ cmp %o2, 15
+ bleu %ncc, .Lunalign_short
+
+ andn %o2, 0x7, %o5 ! %o5 is multiple of 8
+ and %o2, 0x7, %o2 ! residue bytes in %o2
+ add %o2, 8, %o2
+ sub %o5, 8, %o5 ! insure we do not load past end of src
+ andn %o1, 0x7, %o4 ! %o4 has long word aligned src address
+ add %o1, %o5, %o1 ! advance %o1 to after multiple of 8
+ EX_LD_FP(LOAD(ldd, %o4, %f0)) ! fetch partial word
+.Lunalign_by8:
+ EX_LD_FP(LOAD(ldd, %o4+8, %f2))
+ add %o4, 8, %o4
+ faligndata %f0, %f2, %f16
+ subcc %o5, 8, %o5
+ EX_ST_FP(STORE(std, %f16, %o0))
+ fsrc2 %f2, %f0
+ bgu,pt %ncc, .Lunalign_by8
+ add %o0, 8, %o0
+
+.Lunalign_short:
+#ifdef NON_USER_COPY
+ VISExitHalfFast
+#else
+ VISExitHalf
+#endif
+ ba .Lsmallrest
+ nop
+
+/*
+ * This is a special case of nested memcpy. This can happen when kernel
+ * calls unaligned memcpy back to back without saving FP registers. We need
+ * traps(context switch) to save/restore FP registers. If the kernel calls
+ * memcpy without this trap sequence we will hit FP corruption. Let's use
+ * the normal integer load/store method in this case.
+ */
+
+#ifdef NON_USER_COPY
+.Lmedium_vis_entry_fail_cp:
+ or %o0, %o1, %g2
+#endif
+.Lmedium_cp:
+ LOAD(prefetch, %o1 + 0x40, #n_reads_strong)
+ andcc %g2, 0x7, %g0
+ bne,pn %ncc, .Lmedium_unaligned_cp
+ nop
+
+.Lmedium_noprefetch_cp:
+ andncc %o2, 0x20 - 1, %o5
+ be,pn %ncc, 2f
+ sub %o2, %o5, %o2
+1: EX_LD(LOAD(ldx, %o1 + 0x00, %o3))
+ EX_LD(LOAD(ldx, %o1 + 0x08, %g2))
+ EX_LD(LOAD(ldx, %o1 + 0x10, %g7))
+ EX_LD(LOAD(ldx, %o1 + 0x18, %o4))
+ add %o1, 0x20, %o1
+ subcc %o5, 0x20, %o5
+ EX_ST(STORE(stx, %o3, %o0 + 0x00))
+ EX_ST(STORE(stx, %g2, %o0 + 0x08))
+ EX_ST(STORE(stx, %g7, %o0 + 0x10))
+ EX_ST(STORE(stx, %o4, %o0 + 0x18))
+ bne,pt %ncc, 1b
+ add %o0, 0x20, %o0
+2: andcc %o2, 0x18, %o5
+ be,pt %ncc, 3f
+ sub %o2, %o5, %o2
+1: EX_LD(LOAD(ldx, %o1 + 0x00, %o3))
+ add %o1, 0x08, %o1
+ add %o0, 0x08, %o0
+ subcc %o5, 0x08, %o5
+ bne,pt %ncc, 1b
+ EX_ST(STORE(stx, %o3, %o0 - 0x08))
+3: brz,pt %o2, .Lexit_cp
+ cmp %o2, 0x04
+ bl,pn %ncc, .Ltiny_cp
+ nop
+ EX_LD(LOAD(lduw, %o1 + 0x00, %o3))
+ add %o1, 0x04, %o1
+ add %o0, 0x04, %o0
+ subcc %o2, 0x04, %o2
+ bne,pn %ncc, .Ltiny_cp
+ EX_ST(STORE(stw, %o3, %o0 - 0x04))
+ ba,a,pt %ncc, .Lexit_cp
+
+.Lmedium_unaligned_cp:
+ /* First get dest 8 byte aligned. */
+ sub %g0, %o0, %o3
+ and %o3, 0x7, %o3
+ brz,pt %o3, 2f
+ sub %o2, %o3, %o2
+
+1: EX_LD(LOAD(ldub, %o1 + 0x00, %g2))
+ add %o1, 1, %o1
+ subcc %o3, 1, %o3
+ add %o0, 1, %o0
+ bne,pt %ncc, 1b
+ EX_ST(STORE(stb, %g2, %o0 - 0x01))
+2:
+ and %o1, 0x7, %o3
+ brz,pn %o3, .Lmedium_noprefetch_cp
+ sll %o3, 3, %o3
+ mov 64, %g2
+ sub %g2, %o3, %g2
+ andn %o1, 0x7, %o1
+ EX_LD(LOAD(ldx, %o1 + 0x00, %o4))
+ sllx %o4, %o3, %o4
+ andn %o2, 0x08 - 1, %o5
+ sub %o2, %o5, %o2
+
+1: EX_LD(LOAD(ldx, %o1 + 0x08, %g3))
+ add %o1, 0x08, %o1
+ subcc %o5, 0x08, %o5
+ srlx %g3, %g2, %g7
+ or %g7, %o4, %g7
+ EX_ST(STORE(stx, %g7, %o0 + 0x00))
+ add %o0, 0x08, %o0
+ bne,pt %ncc, 1b
+ sllx %g3, %o3, %o4
+ srl %o3, 3, %o3
+ add %o1, %o3, %o1
+ brz,pn %o2, .Lexit_cp
+ nop
+ ba,pt %ncc, .Lsmall_unaligned_cp
+
+.Ltiny_cp:
+ EX_LD(LOAD(ldub, %o1 + 0x00, %o3))
+ subcc %o2, 1, %o2
+ be,pn %ncc, .Lexit_cp
+ EX_ST(STORE(stb, %o3, %o0 + 0x00))
+ EX_LD(LOAD(ldub, %o1 + 0x01, %o3))
+ subcc %o2, 1, %o2
+ be,pn %ncc, .Lexit_cp
+ EX_ST(STORE(stb, %o3, %o0 + 0x01))
+ EX_LD(LOAD(ldub, %o1 + 0x02, %o3))
+ ba,pt %ncc, .Lexit_cp
+ EX_ST(STORE(stb, %o3, %o0 + 0x02))
+
+.Lsmall_cp:
+ andcc %g2, 0x3, %g0
+ bne,pn %ncc, .Lsmall_unaligned_cp
+ andn %o2, 0x4 - 1, %o5
+ sub %o2, %o5, %o2
+1:
+ EX_LD(LOAD(lduw, %o1 + 0x00, %o3))
+ add %o1, 0x04, %o1
+ subcc %o5, 0x04, %o5
+ add %o0, 0x04, %o0
+ bne,pt %ncc, 1b
+ EX_ST(STORE(stw, %o3, %o0 - 0x04))
+ brz,pt %o2, .Lexit_cp
+ nop
+ ba,a,pt %ncc, .Ltiny_cp
+
+.Lsmall_unaligned_cp:
+1: EX_LD(LOAD(ldub, %o1 + 0x00, %o3))
+ add %o1, 1, %o1
+ add %o0, 1, %o0
+ subcc %o2, 1, %o2
+ bne,pt %ncc, 1b
+ EX_ST(STORE(stb, %o3, %o0 - 0x01))
+ ba,a,pt %ncc, .Lexit_cp
+
+.Lsmallrest:
+ tst %o2
+ bz,pt %ncc, .Lsmallx
+ cmp %o2, 4
+ blt,pn %ncc, .Lsmallleft3
+ nop
+ sub %o2, 3, %o2
+.Lsmallnotalign4:
+ EX_LD(LOAD(ldub, %o1, %o3))! read byte
+ subcc %o2, 4, %o2 ! reduce count by 4
+ EX_ST(STORE(stb, %o3, %o0)) ! write byte
+ EX_LD(LOAD(ldub, %o1+1, %o3))! repeat for total of 4 bytes
+ add %o1, 4, %o1 ! advance SRC by 4
+ EX_ST(STORE(stb, %o3, %o0+1))
+ EX_LD(LOAD(ldub, %o1-2, %o3))
+ add %o0, 4, %o0 ! advance DST by 4
+ EX_ST(STORE(stb, %o3, %o0-2))
+ EX_LD(LOAD(ldub, %o1-1, %o3))
+ bgu,pt %ncc, .Lsmallnotalign4 ! loop til 3 or fewer bytes remain
+ EX_ST(STORE(stb, %o3, %o0-1))
+ addcc %o2, 3, %o2 ! restore count
+ bz,pt %ncc, .Lsmallx
+.Lsmallleft3: ! 1, 2, or 3 bytes remain
+ subcc %o2, 1, %o2
+ EX_LD(LOAD(ldub, %o1, %o3)) ! load one byte
+ bz,pt %ncc, .Lsmallx
+ EX_ST(STORE(stb, %o3, %o0)) ! store one byte
+ EX_LD(LOAD(ldub, %o1+1, %o3)) ! load second byte
+ subcc %o2, 1, %o2
+ bz,pt %ncc, .Lsmallx
+ EX_ST(STORE(stb, %o3, %o0+1))! store second byte
+ EX_LD(LOAD(ldub, %o1+2, %o3)) ! load third byte
+ EX_ST(STORE(stb, %o3, %o0+2)) ! store third byte
+.Lsmallx:
+ retl
+ mov EX_RETVAL(%g1), %o0
+.Lsmallfin:
+ tst %o2
+ bnz,pn %ncc, .Lsmallleft3
+ nop
+ retl
+ mov EX_RETVAL(%g1), %o0 ! restore %o0
+.Lexit_cp:
+ retl
+ mov EX_RETVAL(%g1), %o0
+ .size FUNC_NAME, .-FUNC_NAME
diff --git a/arch/sparc/lib/M7memset.S b/arch/sparc/lib/M7memset.S
new file mode 100644
index 0000000..ea88424
--- /dev/null
+++ b/arch/sparc/lib/M7memset.S
@@ -0,0 +1,352 @@
+/*
+ * M7memset.S: SPARC M7 optimized memset.
+ *
+ * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
+ */
+
+/*
+ * M7memset.S: M7 optimized memset.
+ *
+ * char *memset(sp, c, n)
+ *
+ * Set an array of n chars starting at sp to the character c.
+ * Return sp.
+ *
+ * Fast assembler language version of the following C-program for memset
+ * which represents the `standard' for the C-library.
+ *
+ * void *
+ * memset(void *sp1, int c, size_t n)
+ * {
+ * if (n != 0) {
+ * char *sp = sp1;
+ * do {
+ * *sp++ = (char)c;
+ * } while (--n != 0);
+ * }
+ * return (sp1);
+ * }
+ *
+ * The algorithm is as follows :
+ *
+ * For small 6 or fewer bytes stores, bytes will be stored.
+ *
+ * For less than 32 bytes stores, align the address on 4 byte boundary.
+ * Then store as many 4-byte chunks, followed by trailing bytes.
+ *
+ * For sizes greater than 32 bytes, align the address on 8 byte boundary.
+ * if (count >= 64) {
+ * store 8-bytes chunks to align the address on 64 byte boundary
+ * if (value to be set is zero && count >= MIN_ZERO) {
+ * Using BIS stores, set the first long word of each
+ * 64-byte cache line to zero which will also clear the
+ * other seven long words of the cache line.
+ * }
+ * else if (count >= MIN_LOOP) {
+ * Using BIS stores, set the first long word of each of
+ * ST_CHUNK cache lines (64 bytes each) before the main
+ * loop is entered.
+ * In the main loop, continue pre-setting the first long
+ * word of each cache line ST_CHUNK lines in advance while
+ * setting the other seven long words (56 bytes) of each
+ * cache line until fewer than ST_CHUNK*64 bytes remain.
+ * Then set the remaining seven long words of each cache
+ * line that has already had its first long word set.
+ * }
+ * store remaining data in 64-byte chunks until less than
+ * 64 bytes remain.
+ * }
+ * Store as many 8-byte chunks, followed by trailing bytes.
+ *
+ * BIS = Block Init Store
+ * Doing the advance store of the first element of the cache line
+ * initiates the displacement of a cache line while only using a single
+ * instruction in the pipeline. That avoids various pipeline delays,
+ * such as filling the miss buffer. The performance effect is
+ * similar to prefetching for normal stores.
+ * The special case for zero fills runs faster and uses fewer instruction
+ * cycles than the normal memset loop.
+ *
+ * We only use BIS for memset of greater than MIN_LOOP bytes because a sequence
+ * BIS stores must be followed by a membar #StoreStore. The benefit of
+ * the BIS store must be balanced against the cost of the membar operation.
+ */
+
+/*
+ * ASI_STBI_P marks the cache line as "least recently used"
+ * which means if many threads are active, it has a high chance
+ * of being pushed out of the cache between the first initializing
+ * store and the final stores.
+ * Thus, we use ASI_STBIMRU_P which marks the cache line as
+ * "most recently used" for all but the last store to the cache line.
+ */
+
+#include <asm/asi.h>
+#include <asm/page.h>
+
+#define ASI_STBI_P ASI_BLK_INIT_QUAD_LDD_P
+#define ASI_STBIMRU_P ASI_ST_BLKINIT_MRU_P
+
+
+#define ST_CHUNK 24 /* multiple of 4 due to loop unrolling */
+#define MIN_LOOP 16320
+#define MIN_ZERO 512
+
+ .section ".text"
+ .align 32
+
+/*
+ * Define clear_page(dest) as memset(dest, 0, PAGE_SIZE)
+ * (can create a more optimized version later.)
+ */
+ .globl M7clear_page
+ .globl M7clear_user_page
+M7clear_page: /* clear_page(dest) */
+M7clear_user_page:
+ set PAGE_SIZE, %o1
+ /* fall through into bzero code */
+
+ .size M7clear_page,.-M7clear_page
+ .size M7clear_user_page,.-M7clear_user_page
+
+/*
+ * Define bzero(dest, n) as memset(dest, 0, n)
+ * (can create a more optimized version later.)
+ */
+ .globl M7bzero
+M7bzero: /* bzero(dest, size) */
+ mov %o1, %o2
+ mov 0, %o1
+ /* fall through into memset code */
+
+ .size M7bzero,.-M7bzero
+
+ .global M7memset
+ .type M7memset, #function
+ .register %g3, #scratch
+M7memset:
+ mov %o0, %o5 ! copy sp1 before using it
+ cmp %o2, 7 ! if small counts, just write bytes
+ bleu,pn %ncc, .wrchar
+ and %o1, 0xff, %o1 ! o1 is (char)c
+
+ sll %o1, 8, %o3
+ or %o1, %o3, %o1 ! now o1 has 2 bytes of c
+ sll %o1, 16, %o3
+ cmp %o2, 32
+ blu,pn %ncc, .wdalign
+ or %o1, %o3, %o1 ! now o1 has 4 bytes of c
+
+ sllx %o1, 32, %o3
+ or %o1, %o3, %o1 ! now o1 has 8 bytes of c
+
+.dbalign:
+ andcc %o5, 7, %o3 ! is sp1 aligned on a 8 byte bound?
+ bz,pt %ncc, .blkalign ! already long word aligned
+ sub %o3, 8, %o3 ! -(bytes till long word aligned)
+
+ add %o2, %o3, %o2 ! update o2 with new count
+ ! Set -(%o3) bytes till sp1 long word aligned
+1: stb %o1, [%o5] ! there is at least 1 byte to set
+ inccc %o3 ! byte clearing loop
+ bl,pt %ncc, 1b
+ inc %o5
+
+ ! Now sp1 is long word aligned (sp1 is found in %o5)
+.blkalign:
+ cmp %o2, 64 ! check if there are 64 bytes to set
+ blu,pn %ncc, .wrshort
+ mov %o2, %o3
+
+ andcc %o5, 63, %o3 ! is sp1 block aligned?
+ bz,pt %ncc, .blkwr ! now block aligned
+ sub %o3, 64, %o3 ! o3 is -(bytes till block aligned)
+ add %o2, %o3, %o2 ! o2 is the remainder
+
+ ! Store -(%o3) bytes till dst is block (64 byte) aligned.
+ ! Use long word stores.
+ ! Recall that dst is already long word aligned
+1:
+ addcc %o3, 8, %o3
+ stx %o1, [%o5]
+ bl,pt %ncc, 1b
+ add %o5, 8, %o5
+
+ ! Now sp1 is block aligned
+.blkwr:
+ andn %o2, 63, %o4 ! calculate size of blocks in bytes
+ brz,pn %o1, .wrzero ! special case if c == 0
+ and %o2, 63, %o3 ! %o3 = bytes left after blk stores.
+
+ set MIN_LOOP, %g1
+ cmp %o4, %g1 ! check there are enough bytes to set
+ blu,pn %ncc, .short_set ! to justify cost of membar
+ ! must be > pre-cleared lines
+ nop
+
+ ! initial cache-clearing stores
+ ! get store pipeline moving
+ rd %asi, %g3 ! save %asi to be restored later
+ wr %g0, ASI_STBIMRU_P, %asi
+
+ ! Primary memset loop for large memsets
+.wr_loop:
+ sub %o5, 8, %o5 ! adjust %o5 for ASI store alignment
+ mov ST_CHUNK, %g1
+.wr_loop_start:
+ stxa %o1, [%o5+8]%asi
+ subcc %g1, 4, %g1
+ stxa %o1, [%o5+8+64]%asi
+ add %o5, 256, %o5
+ stxa %o1, [%o5+8-128]%asi
+ bgu %ncc, .wr_loop_start
+ stxa %o1, [%o5+8-64]%asi
+
+ sub %o5, ST_CHUNK*64, %o5 ! reset %o5
+ mov ST_CHUNK, %g1
+
+.wr_loop_rest:
+ stxa %o1, [%o5+8+8]%asi
+ sub %o4, 64, %o4
+ stxa %o1, [%o5+16+8]%asi
+ subcc %g1, 1, %g1
+ stxa %o1, [%o5+24+8]%asi
+ stxa %o1, [%o5+32+8]%asi
+ stxa %o1, [%o5+40+8]%asi
+ add %o5, 64, %o5
+ stxa %o1, [%o5-8]%asi
+ bgu %ncc, .wr_loop_rest
+ stxa %o1, [%o5]ASI_STBI_P
+
+ ! If more than ST_CHUNK*64 bytes remain to set, continue
+ ! setting the first long word of each cache line in advance
+ ! to keep the store pipeline moving.
+
+ cmp %o4, ST_CHUNK*64
+ bge,pt %ncc, .wr_loop_start
+ mov ST_CHUNK, %g1
+
+ brz,a,pn %o4, .asi_done
+ add %o5, 8, %o5 ! restore %o5 offset
+
+.wr_loop_small:
+ stxa %o1, [%o5+8]%asi
+ stxa %o1, [%o5+8+8]%asi
+ stxa %o1, [%o5+16+8]%asi
+ stxa %o1, [%o5+24+8]%asi
+ stxa %o1, [%o5+32+8]%asi
+ subcc %o4, 64, %o4
+ stxa %o1, [%o5+40+8]%asi
+ add %o5, 64, %o5
+ stxa %o1, [%o5-8]%asi
+ bgu,pt %ncc, .wr_loop_small
+ stxa %o1, [%o5]ASI_STBI_P
+
+ ba .asi_done
+ add %o5, 8, %o5 ! restore %o5 offset
+
+ ! Special case loop for zero fill memsets
+ ! For each 64 byte cache line, single STBI to first element
+ ! clears line
+.wrzero:
+ cmp %o4, MIN_ZERO ! check if enough bytes to set
+ ! to pay %asi + membar cost
+ blu %ncc, .short_set
+ nop
+ sub %o4, 256, %o4
+
+.wrzero_loop:
+ mov 64, %g3
+ stxa %o1, [%o5]ASI_STBI_P
+ subcc %o4, 256, %o4
+ stxa %o1, [%o5+%g3]ASI_STBI_P
+ add %o5, 256, %o5
+ sub %g3, 192, %g3
+ stxa %o1, [%o5+%g3]ASI_STBI_P
+ add %g3, 64, %g3
+ bge,pt %ncc, .wrzero_loop
+ stxa %o1, [%o5+%g3]ASI_STBI_P
+ add %o4, 256, %o4
+
+ brz,pn %o4, .bsi_done
+ nop
+
+.wrzero_small:
+ stxa %o1, [%o5]ASI_STBI_P
+ subcc %o4, 64, %o4
+ bgu,pt %ncc, .wrzero_small
+ add %o5, 64, %o5
+ ba,a .bsi_done
+
+.asi_done:
+ wr %g3, 0x0, %asi ! restored saved %asi
+.bsi_done:
+ membar #StoreStore ! required by use of Block Store Init
+
+.short_set:
+ cmp %o4, 64 ! check if 64 bytes to set
+ blu %ncc, 5f
+ nop
+4: ! set final blocks of 64 bytes
+ stx %o1, [%o5]
+ stx %o1, [%o5+8]
+ stx %o1, [%o5+16]
+ stx %o1, [%o5+24]
+ subcc %o4, 64, %o4
+ stx %o1, [%o5+32]
+ stx %o1, [%o5+40]
+ add %o5, 64, %o5
+ stx %o1, [%o5-16]
+ bgu,pt %ncc, 4b
+ stx %o1, [%o5-8]
+
+5:
+ ! Set the remaining long words
+.wrshort:
+ subcc %o3, 8, %o3 ! Can we store any long words?
+ blu,pn %ncc, .wrchars
+ and %o2, 7, %o2 ! calc bytes left after long words
+6:
+ subcc %o3, 8, %o3
+ stx %o1, [%o5] ! store the long words
+ bgeu,pt %ncc, 6b
+ add %o5, 8, %o5
+
+.wrchars: ! check for extra chars
+ brnz %o2, .wrfin
+ nop
+ retl
+ nop
+
+.wdalign:
+ andcc %o5, 3, %o3 ! is sp1 aligned on a word boundary
+ bz,pn %ncc, .wrword
+ andn %o2, 3, %o3 ! create word sized count in %o3
+
+ dec %o2 ! decrement count
+ stb %o1, [%o5] ! clear a byte
+ b .wdalign
+ inc %o5 ! next byte
+
+.wrword:
+ subcc %o3, 4, %o3
+ st %o1, [%o5] ! 4-byte writing loop
+ bnz,pt %ncc, .wrword
+ add %o5, 4, %o5
+
+ and %o2, 3, %o2 ! leftover count, if any
+
+.wrchar:
+ ! Set the remaining bytes, if any
+ brz %o2, .exit
+ nop
+.wrfin:
+ deccc %o2
+ stb %o1, [%o5]
+ bgu,pt %ncc, .wrfin
+ inc %o5
+.exit:
+ retl ! %o0 was preserved
+ nop
+
+ .size M7memset,.-M7memset
diff --git a/arch/sparc/lib/M7patch.S b/arch/sparc/lib/M7patch.S
new file mode 100644
index 0000000..9000b7b
--- /dev/null
+++ b/arch/sparc/lib/M7patch.S
@@ -0,0 +1,51 @@
+/*
+ * M7patch.S: Patch generic routines with M7 variant.
+ *
+ * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
+ */
+
+#include <linux/linkage.h>
+
+#define BRANCH_ALWAYS 0x10680000
+#define NOP 0x01000000
+#define NG_DO_PATCH(OLD, NEW) \
+ sethi %hi(NEW), %g1; \
+ or %g1, %lo(NEW), %g1; \
+ sethi %hi(OLD), %g2; \
+ or %g2, %lo(OLD), %g2; \
+ sub %g1, %g2, %g1; \
+ sethi %hi(BRANCH_ALWAYS), %g3; \
+ sll %g1, 11, %g1; \
+ srl %g1, 11 + 2, %g1; \
+ or %g3, %lo(BRANCH_ALWAYS), %g3; \
+ or %g3, %g1, %g3; \
+ stw %g3, [%g2]; \
+ sethi %hi(NOP), %g3; \
+ or %g3, %lo(NOP), %g3; \
+ stw %g3, [%g2 + 0x4]; \
+ flush %g2;
+
+ENTRY(m7_patch_copyops)
+ NG_DO_PATCH(memcpy, M7memcpy)
+ NG_DO_PATCH(raw_copy_from_user, M7copy_from_user)
+ NG_DO_PATCH(raw_copy_to_user, M7copy_to_user)
+ retl
+ nop
+ENDPROC(m7_patch_copyops)
+
+ENTRY(m7_patch_bzero)
+ NG_DO_PATCH(memset, M7memset)
+ NG_DO_PATCH(__bzero, M7bzero)
+ NG_DO_PATCH(__clear_user, NGclear_user)
+ NG_DO_PATCH(tsb_init, NGtsb_init)
+ retl
+ nop
+ENDPROC(m7_patch_bzero)
+
+ENTRY(m7_patch_pageops)
+ NG_DO_PATCH(copy_user_page, NG4copy_user_page)
+ NG_DO_PATCH(_clear_page, M7clear_page)
+ NG_DO_PATCH(clear_user_page, M7clear_user_page)
+ retl
+ nop
+ENDPROC(m7_patch_pageops)
diff --git a/arch/sparc/lib/Makefile b/arch/sparc/lib/Makefile
index 37930c0..a1a2d39 100644
--- a/arch/sparc/lib/Makefile
+++ b/arch/sparc/lib/Makefile
@@ -38,6 +38,9 @@ lib-$(CONFIG_SPARC64) += NG4patch.o NG4copy_page.o NG4clear_page.o NG4memset.o
lib-$(CONFIG_SPARC64) += Memcpy_utils.o
+lib-$(CONFIG_SPARC64) += M7memcpy.o M7copy_from_user.o M7copy_to_user.o
+lib-$(CONFIG_SPARC64) += M7patch.o M7memset.o
+
lib-$(CONFIG_SPARC64) += GENmemcpy.o GENcopy_from_user.o GENcopy_to_user.o
lib-$(CONFIG_SPARC64) += GENpatch.o GENpage.o GENbzero.o
--
1.7.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH v2 4/4] arch/sparc: Add accurate exception reporting in M7memcpy
2017-08-07 23:52 [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures Babu Moger
` (2 preceding siblings ...)
2017-08-07 23:52 ` [PATCH v2 3/4] arch/sparc: Optimized memcpy, memset, copy_to_user, copy_from_user for M7/M8 Babu Moger
@ 2017-08-07 23:52 ` Babu Moger
2017-08-10 21:38 ` [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures David Miller
4 siblings, 0 replies; 7+ messages in thread
From: Babu Moger @ 2017-08-07 23:52 UTC (permalink / raw)
To: davem; +Cc: sparclinux, linux-kernel, babu.moger
Add accurate exception reporting in M7memcpy
Signed-off-by: Babu Moger <babu.moger@oracle.com>
---
arch/sparc/lib/M7copy_from_user.S | 11 +-
arch/sparc/lib/M7copy_to_user.S | 10 +-
arch/sparc/lib/M7memcpy.S | 396 ++++++++++++++++++------------------
arch/sparc/lib/Memcpy_utils.S | 182 +++++++++++++++++
4 files changed, 390 insertions(+), 209 deletions(-)
diff --git a/arch/sparc/lib/M7copy_from_user.S b/arch/sparc/lib/M7copy_from_user.S
index d0689d7..66464b3 100644
--- a/arch/sparc/lib/M7copy_from_user.S
+++ b/arch/sparc/lib/M7copy_from_user.S
@@ -5,23 +5,22 @@
*/
-#define EX_LD(x) \
+#define EX_LD(x, y) \
98: x; \
.section __ex_table,"a"; \
.align 4; \
- .word 98b, __restore_asi; \
+ .word 98b, y; \
.text; \
.align 4;
-#define EX_LD_FP(x) \
+#define EX_LD_FP(x, y) \
98: x; \
.section __ex_table,"a"; \
.align 4; \
- .word 98b, __restore_asi_fp; \
+ .word 98b, y##_fp; \
.text; \
.align 4;
-
#ifndef ASI_AIUS
#define ASI_AIUS 0x11
#endif
@@ -35,7 +34,7 @@
rd %asi, %g1; \
cmp %g1, ASI_AIUS; \
bne,pn %icc, raw_copy_in_user; \
- nop
+ nop
#endif
#include "M7memcpy.S"
diff --git a/arch/sparc/lib/M7copy_to_user.S b/arch/sparc/lib/M7copy_to_user.S
index d3be132..a60ac46 100644
--- a/arch/sparc/lib/M7copy_to_user.S
+++ b/arch/sparc/lib/M7copy_to_user.S
@@ -5,19 +5,19 @@
*/
-#define EX_ST(x) \
+#define EX_ST(x, y) \
98: x; \
.section __ex_table,"a"; \
.align 4; \
- .word 98b, __restore_asi; \
+ .word 98b, y; \
.text; \
.align 4;
-#define EX_ST_FP(x) \
+#define EX_ST_FP(x, y) \
98: x; \
.section __ex_table,"a"; \
.align 4; \
- .word 98b, __restore_asi_fp; \
+ .word 98b, y##_fp; \
.text; \
.align 4;
@@ -45,7 +45,7 @@
rd %asi, %g1; \
cmp %g1, ASI_AIUS; \
bne,pn %icc, raw_copy_in_user; \
- nop
+ nop
#endif
#include "M7memcpy.S"
diff --git a/arch/sparc/lib/M7memcpy.S b/arch/sparc/lib/M7memcpy.S
index 0a0421d..d016fc2 100644
--- a/arch/sparc/lib/M7memcpy.S
+++ b/arch/sparc/lib/M7memcpy.S
@@ -96,17 +96,17 @@
#endif
#ifndef EX_LD
-#define EX_LD(x) x
+#define EX_LD(x,y) x
#endif
#ifndef EX_LD_FP
-#define EX_LD_FP(x) x
+#define EX_LD_FP(x,y) x
#endif
#ifndef EX_ST
-#define EX_ST(x) x
+#define EX_ST(x,y) x
#endif
#ifndef EX_ST_FP
-#define EX_ST_FP(x) x
+#define EX_ST_FP(x,y) x
#endif
#ifndef EX_RETVAL
@@ -206,9 +206,9 @@ FUNC_NAME:
sub %o1, %o0, %o1 ! %o1 gets the difference
7: ! dst aligning loop
add %o1, %o0, %o4
- EX_LD(LOAD(ldub, %o4, %o4)) ! load one byte
+ EX_LD(LOAD(ldub, %o4, %o4), memcpy_retl_o2_plus_o5) ! load one byte
subcc %o5, 1, %o5
- EX_ST(STORE(stb, %o4, %o0))
+ EX_ST(STORE(stb, %o4, %o0), memcpy_retl_o2_plus_o5_plus_1)
bgu,pt %ncc, 7b
add %o0, 1, %o0 ! advance dst
add %o1, %o0, %o1 ! restore %o1
@@ -233,64 +233,64 @@ FUNC_NAME:
ble,pn %ncc, .Lmedl63 ! skip big loop if less than 64 bytes
nop
.Lmedl64:
- EX_LD(LOAD(ldx, %o1, %o4)) ! load
+ EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl_o2_plus_63) ! load
subcc %o2, 64, %o2 ! decrement length count
- EX_ST(STORE(stx, %o4, %o0)) ! and store
- EX_LD(LOAD(ldx, %o1+8, %o3)) ! a block of 64 bytes
- EX_ST(STORE(stx, %o3, %o0+8))
- EX_LD(LOAD(ldx, %o1+16, %o4))
- EX_ST(STORE(stx, %o4, %o0+16))
- EX_LD(LOAD(ldx, %o1+24, %o3))
- EX_ST(STORE(stx, %o3, %o0+24))
- EX_LD(LOAD(ldx, %o1+32, %o4)) ! load
- EX_ST(STORE(stx, %o4, %o0+32)) ! and store
- EX_LD(LOAD(ldx, %o1+40, %o3)) ! a block of 64 bytes
+ EX_ST(STORE(stx, %o4, %o0), memcpy_retl_o2_plus_63_64) ! and store
+ EX_LD(LOAD(ldx, %o1+8, %o3), memcpy_retl_o2_plus_63_56) ! a block of 64
+ EX_ST(STORE(stx, %o3, %o0+8), memcpy_retl_o2_plus_63_56)
+ EX_LD(LOAD(ldx, %o1+16, %o4), memcpy_retl_o2_plus_63_48)
+ EX_ST(STORE(stx, %o4, %o0+16), memcpy_retl_o2_plus_63_48)
+ EX_LD(LOAD(ldx, %o1+24, %o3), memcpy_retl_o2_plus_63_40)
+ EX_ST(STORE(stx, %o3, %o0+24), memcpy_retl_o2_plus_63_40)
+ EX_LD(LOAD(ldx, %o1+32, %o4), memcpy_retl_o2_plus_63_32)! load and store
+ EX_ST(STORE(stx, %o4, %o0+32), memcpy_retl_o2_plus_63_32)
+ EX_LD(LOAD(ldx, %o1+40, %o3), memcpy_retl_o2_plus_63_24)! a block of 64
add %o1, 64, %o1 ! increase src ptr by 64
- EX_ST(STORE(stx, %o3, %o0+40))
- EX_LD(LOAD(ldx, %o1-16, %o4))
+ EX_ST(STORE(stx, %o3, %o0+40), memcpy_retl_o2_plus_63_24)
+ EX_LD(LOAD(ldx, %o1-16, %o4), memcpy_retl_o2_plus_63_16)
add %o0, 64, %o0 ! increase dst ptr by 64
- EX_ST(STORE(stx, %o4, %o0-16))
- EX_LD(LOAD(ldx, %o1-8, %o3))
+ EX_ST(STORE(stx, %o4, %o0-16), memcpy_retl_o2_plus_63_16)
+ EX_LD(LOAD(ldx, %o1-8, %o3), memcpy_retl_o2_plus_63_8)
bgu,pt %ncc, .Lmedl64 ! repeat if at least 64 bytes left
- EX_ST(STORE(stx, %o3, %o0-8))
+ EX_ST(STORE(stx, %o3, %o0-8), memcpy_retl_o2_plus_63_8)
.Lmedl63:
addcc %o2, 32, %o2 ! adjust remaining count
ble,pt %ncc, .Lmedl31 ! to skip if 31 or fewer bytes left
nop
- EX_LD(LOAD(ldx, %o1, %o4)) ! load
+ EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl_o2_plus_31) ! load
sub %o2, 32, %o2 ! decrement length count
- EX_ST(STORE(stx, %o4, %o0)) ! and store
- EX_LD(LOAD(ldx, %o1+8, %o3)) ! a block of 32 bytes
+ EX_ST(STORE(stx, %o4, %o0), memcpy_retl_o2_plus_31_32) ! and store
+ EX_LD(LOAD(ldx, %o1+8, %o3), memcpy_retl_o2_plus_31_24) ! a block of 32
add %o1, 32, %o1 ! increase src ptr by 32
- EX_ST(STORE(stx, %o3, %o0+8))
- EX_LD(LOAD(ldx, %o1-16, %o4))
+ EX_ST(STORE(stx, %o3, %o0+8), memcpy_retl_o2_plus_31_24)
+ EX_LD(LOAD(ldx, %o1-16, %o4), memcpy_retl_o2_plus_31_16)
add %o0, 32, %o0 ! increase dst ptr by 32
- EX_ST(STORE(stx, %o4, %o0-16))
- EX_LD(LOAD(ldx, %o1-8, %o3))
- EX_ST(STORE(stx, %o3, %o0-8))
+ EX_ST(STORE(stx, %o4, %o0-16), memcpy_retl_o2_plus_31_16)
+ EX_LD(LOAD(ldx, %o1-8, %o3), memcpy_retl_o2_plus_31_8)
+ EX_ST(STORE(stx, %o3, %o0-8), memcpy_retl_o2_plus_31_8)
.Lmedl31:
addcc %o2, 16, %o2 ! adjust remaining count
ble,pt %ncc, .Lmedl15 ! skip if 15 or fewer bytes left
nop !
- EX_LD(LOAD(ldx, %o1, %o4))
+ EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl_o2_plus_15)
add %o1, 16, %o1 ! increase src ptr by 16
- EX_ST(STORE(stx, %o4, %o0))
+ EX_ST(STORE(stx, %o4, %o0), memcpy_retl_o2_plus_15)
sub %o2, 16, %o2 ! decrease count by 16
- EX_LD(LOAD(ldx, %o1-8, %o3))
+ EX_LD(LOAD(ldx, %o1-8, %o3), memcpy_retl_o2_plus_15_8)
add %o0, 16, %o0 ! increase dst ptr by 16
- EX_ST(STORE(stx, %o3, %o0-8))
+ EX_ST(STORE(stx, %o3, %o0-8), memcpy_retl_o2_plus_15_8)
.Lmedl15:
addcc %o2, 15, %o2 ! restore count
bz,pt %ncc, .Lsmallx ! exit if finished
cmp %o2, 8
blt,pt %ncc, .Lmedw7 ! skip if 7 or fewer bytes left
tst %o2
- EX_LD(LOAD(ldx, %o1, %o4)) ! load 8 bytes
+ EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl_o2) ! load 8 bytes
add %o1, 8, %o1 ! increase src ptr by 8
add %o0, 8, %o0 ! increase dst ptr by 8
subcc %o2, 8, %o2 ! decrease count by 8
bnz,pn %ncc, .Lmedw7
- EX_ST(STORE(stx, %o4, %o0-8)) ! and store 8 bytes
+ EX_ST(STORE(stx, %o4, %o0-8), memcpy_retl_o2_plus_8) ! and store 8
retl
mov EX_RETVAL(%g1), %o0 ! restore %o0
@@ -318,30 +318,30 @@ FUNC_NAME:
! for end of loop
ble,pt %ncc, .Lmedw31 ! skip big loop if less than 16
.Lmedw32:
- EX_LD(LOAD(ld, %o1, %o4)) ! move a block of 32 bytes
+ EX_LD(LOAD(ld, %o1, %o4), memcpy_retl_o2_plus_31)! move a block of 32
sllx %o4, 32, %o5
- EX_LD(LOAD(ld, %o1+4, %o4))
+ EX_LD(LOAD(ld, %o1+4, %o4), memcpy_retl_o2_plus_31)
or %o4, %o5, %o5
- EX_ST(STORE(stx, %o5, %o0))
+ EX_ST(STORE(stx, %o5, %o0), memcpy_retl_o2_plus_31)
subcc %o2, 32, %o2 ! decrement length count
- EX_LD(LOAD(ld, %o1+8, %o4))
+ EX_LD(LOAD(ld, %o1+8, %o4), memcpy_retl_o2_plus_31_24)
sllx %o4, 32, %o5
- EX_LD(LOAD(ld, %o1+12, %o4))
+ EX_LD(LOAD(ld, %o1+12, %o4), memcpy_retl_o2_plus_31_24)
or %o4, %o5, %o5
- EX_ST(STORE(stx, %o5, %o0+8))
+ EX_ST(STORE(stx, %o5, %o0+8), memcpy_retl_o2_plus_31_24)
add %o1, 32, %o1 ! increase src ptr by 32
- EX_LD(LOAD(ld, %o1-16, %o4))
+ EX_LD(LOAD(ld, %o1-16, %o4), memcpy_retl_o2_plus_31_16)
sllx %o4, 32, %o5
- EX_LD(LOAD(ld, %o1-12, %o4))
+ EX_LD(LOAD(ld, %o1-12, %o4), memcpy_retl_o2_plus_31_16)
or %o4, %o5, %o5
- EX_ST(STORE(stx, %o5, %o0+16))
+ EX_ST(STORE(stx, %o5, %o0+16), memcpy_retl_o2_plus_31_16)
add %o0, 32, %o0 ! increase dst ptr by 32
- EX_LD(LOAD(ld, %o1-8, %o4))
+ EX_LD(LOAD(ld, %o1-8, %o4), memcpy_retl_o2_plus_31_8)
sllx %o4, 32, %o5
- EX_LD(LOAD(ld, %o1-4, %o4))
+ EX_LD(LOAD(ld, %o1-4, %o4), memcpy_retl_o2_plus_31_8)
or %o4, %o5, %o5
bgu,pt %ncc, .Lmedw32 ! repeat if at least 32 bytes left
- EX_ST(STORE(stx, %o5, %o0-8))
+ EX_ST(STORE(stx, %o5, %o0-8), memcpy_retl_o2_plus_31_8)
.Lmedw31:
addcc %o2, 31, %o2 ! restore count
@@ -350,42 +350,42 @@ FUNC_NAME:
cmp %o2, 16
blt,pt %ncc, .Lmedw15
nop
- EX_LD(LOAD(ld, %o1, %o4)) ! move a block of 16 bytes
+ EX_LD(LOAD(ld, %o1, %o4), memcpy_retl_o2)! move a block of 16 bytes
sllx %o4, 32, %o5
subcc %o2, 16, %o2 ! decrement length count
- EX_LD(LOAD(ld, %o1+4, %o4))
+ EX_LD(LOAD(ld, %o1+4, %o4), memcpy_retl_o2_plus_16)
or %o4, %o5, %o5
- EX_ST(STORE(stx, %o5, %o0))
+ EX_ST(STORE(stx, %o5, %o0), memcpy_retl_o2_plus_16)
add %o1, 16, %o1 ! increase src ptr by 16
- EX_LD(LOAD(ld, %o1-8, %o4))
+ EX_LD(LOAD(ld, %o1-8, %o4), memcpy_retl_o2_plus_8)
add %o0, 16, %o0 ! increase dst ptr by 16
sllx %o4, 32, %o5
- EX_LD(LOAD(ld, %o1-4, %o4))
+ EX_LD(LOAD(ld, %o1-4, %o4), memcpy_retl_o2_plus_8)
or %o4, %o5, %o5
- EX_ST(STORE(stx, %o5, %o0-8))
+ EX_ST(STORE(stx, %o5, %o0-8), memcpy_retl_o2_plus_8)
.Lmedw15:
bz,pt %ncc, .Lsmallx ! exit if finished
cmp %o2, 8
blt,pn %ncc, .Lmedw7 ! skip if 7 or fewer bytes left
tst %o2
- EX_LD(LOAD(ld, %o1, %o4)) ! load 4 bytes
+ EX_LD(LOAD(ld, %o1, %o4), memcpy_retl_o2) ! load 4 bytes
subcc %o2, 8, %o2 ! decrease count by 8
- EX_ST(STORE(stw, %o4, %o0)) ! and store 4 bytes
+ EX_ST(STORE(stw, %o4, %o0), memcpy_retl_o2_plus_8)! and store 4 bytes
add %o1, 8, %o1 ! increase src ptr by 8
- EX_LD(LOAD(ld, %o1-4, %o3)) ! load 4 bytes
+ EX_LD(LOAD(ld, %o1-4, %o3), memcpy_retl_o2_plus_4) ! load 4 bytes
add %o0, 8, %o0 ! increase dst ptr by 8
- EX_ST(STORE(stw, %o3, %o0-4)) ! and store 4 bytes
+ EX_ST(STORE(stw, %o3, %o0-4), memcpy_retl_o2_plus_4)! and store 4 bytes
bz,pt %ncc, .Lsmallx ! exit if finished
.Lmedw7: ! count is ge 1, less than 8
cmp %o2, 4 ! check for 4 bytes left
blt,pn %ncc, .Lsmallleft3 ! skip if 3 or fewer bytes left
nop !
- EX_LD(LOAD(ld, %o1, %o4)) ! load 4 bytes
+ EX_LD(LOAD(ld, %o1, %o4), memcpy_retl_o2) ! load 4 bytes
add %o1, 4, %o1 ! increase src ptr by 4
add %o0, 4, %o0 ! increase dst ptr by 4
subcc %o2, 4, %o2 ! decrease count by 4
bnz .Lsmallleft3
- EX_ST(STORE(stw, %o4, %o0-4))! and store 4 bytes
+ EX_ST(STORE(stw, %o4, %o0-4), memcpy_retl_o2_plus_4)! and store 4 bytes
retl
mov EX_RETVAL(%g1), %o0
@@ -397,37 +397,37 @@ FUNC_NAME:
andcc %o0, 8, %o3 ! odd long words to move?
brz,pt %o3, .Laligned_to_16
nop
- EX_LD(LOAD(ldx, %o1, %o4))
+ EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl_o2)
sub %o2, 8, %o2
add %o1, 8, %o1 ! increment src ptr
add %o0, 8, %o0 ! increment dst ptr
- EX_ST(STORE(stx, %o4, %o0-8))
+ EX_ST(STORE(stx, %o4, %o0-8), memcpy_retl_o2_plus_8)
.Laligned_to_16:
andcc %o0, 16, %o3 ! pair of long words to move?
brz,pt %o3, .Laligned_to_32
nop
- EX_LD(LOAD(ldx, %o1, %o4))
+ EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl_o2)
sub %o2, 16, %o2
- EX_ST(STORE(stx, %o4, %o0))
+ EX_ST(STORE(stx, %o4, %o0), memcpy_retl_o2_plus_16)
add %o1, 16, %o1 ! increment src ptr
- EX_LD(LOAD(ldx, %o1-8, %o4))
+ EX_LD(LOAD(ldx, %o1-8, %o4), memcpy_retl_o2_plus_8)
add %o0, 16, %o0 ! increment dst ptr
- EX_ST(STORE(stx, %o4, %o0-8))
+ EX_ST(STORE(stx, %o4, %o0-8), memcpy_retl_o2_plus_8)
.Laligned_to_32:
andcc %o0, 32, %o3 ! four long words to move?
brz,pt %o3, .Laligned_to_64
nop
- EX_LD(LOAD(ldx, %o1, %o4))
+ EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl_o2)
sub %o2, 32, %o2
- EX_ST(STORE(stx, %o4, %o0))
- EX_LD(LOAD(ldx, %o1+8, %o4))
- EX_ST(STORE(stx, %o4, %o0+8))
- EX_LD(LOAD(ldx, %o1+16, %o4))
- EX_ST(STORE(stx, %o4, %o0+16))
+ EX_ST(STORE(stx, %o4, %o0), memcpy_retl_o2_plus_32)
+ EX_LD(LOAD(ldx, %o1+8, %o4), memcpy_retl_o2_plus_24)
+ EX_ST(STORE(stx, %o4, %o0+8), memcpy_retl_o2_plus_24)
+ EX_LD(LOAD(ldx, %o1+16, %o4), memcpy_retl_o2_plus_16)
+ EX_ST(STORE(stx, %o4, %o0+16), memcpy_retl_o2_plus_16)
add %o1, 32, %o1 ! increment src ptr
- EX_LD(LOAD(ldx, %o1-8, %o4))
+ EX_LD(LOAD(ldx, %o1-8, %o4), memcpy_retl_o2_plus_8)
add %o0, 32, %o0 ! increment dst ptr
- EX_ST(STORE(stx, %o4, %o0-8))
+ EX_ST(STORE(stx, %o4, %o0-8), memcpy_retl_o2_plus_8)
.Laligned_to_64:
!
! Using block init store (BIS) instructions to avoid fetching cache
@@ -461,10 +461,10 @@ FUNC_NAME:
.Lalign_loop_start:
prefetch [%o1 + (ALIGN_PRE * BLOCK_SIZE)], 21
subcc %o3, 1, %o3
- EX_LD(LOAD(ldx, %o1, %o4))
+ EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl_o2_plus_o5)
add %o1, 64, %o1
add %o0, 8, %o0
- EX_ST(STORE_INIT_MRU(%o4, %o0))
+ EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy_retl_o2_plus_o5)
bgu %ncc,.Lalign_loop_start
add %o0, 56, %o0
@@ -474,32 +474,32 @@ FUNC_NAME:
sub %o0, %o4, %o0 ! reset %o0
.Lalign_loop_rest:
- EX_LD(LOAD(ldx, %o1+8, %o4))
+ EX_LD(LOAD(ldx, %o1+8, %o4), memcpy_retl_o2_plus_o5)
add %o0, 16, %o0
- EX_ST(STORE_INIT_MRU(%o4, %o0))
- EX_LD(LOAD(ldx, %o1+16, %o4))
+ EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1+16, %o4), memcpy_retl_o2_plus_o5)
add %o0, 8, %o0
- EX_ST(STORE_INIT_MRU(%o4, %o0))
+ EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy_retl_o2_plus_o5)
subcc %o3, 1, %o3
- EX_LD(LOAD(ldx, %o1+24, %o4))
+ EX_LD(LOAD(ldx, %o1+24, %o4), memcpy_retl_o2_plus_o5)
add %o0, 8, %o0
- EX_ST(STORE_INIT_MRU(%o4, %o0))
- EX_LD(LOAD(ldx, %o1+32, %o4))
+ EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1+32, %o4), memcpy_retl_o2_plus_o5)
add %o0, 8, %o0
- EX_ST(STORE_INIT_MRU(%o4, %o0))
- EX_LD(LOAD(ldx, %o1+40, %o4))
+ EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1+40, %o4), memcpy_retl_o2_plus_o5)
add %o0, 8, %o0
- EX_ST(STORE_INIT_MRU(%o4, %o0))
- EX_LD(LOAD(ldx, %o1+48, %o4))
+ EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1+48, %o4), memcpy_retl_o2_plus_o5)
add %o1, 64, %o1
add %o0, 8, %o0
- EX_ST(STORE_INIT_MRU(%o4, %o0))
+ EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy_retl_o2_plus_o5)
add %o0, 8, %o0
- EX_LD(LOAD(ldx, %o1-8, %o4))
+ EX_LD(LOAD(ldx, %o1-8, %o4), memcpy_retl_o2_plus_o5)
sub %o5, 64, %o5
bgu %ncc,.Lalign_loop_rest
! mark cache line as LRU
- EX_ST(STORE_INIT(%o4, %o0))
+ EX_ST(STORE_INIT(%o4, %o0), memcpy_retl_o2_plus_o5_plus_64)
cmp %o5, ST_CHUNK*64
bgu,pt %ncc, .Lalign_loop_start
@@ -509,26 +509,26 @@ FUNC_NAME:
beq .Lalign_done
nop
.Lalign_loop_fin:
- EX_LD(LOAD(ldx, %o1, %o4))
- EX_ST(STORE(stx, %o4, %o0+8))
- EX_LD(LOAD(ldx, %o1+8, %o4))
- EX_ST(STORE(stx, %o4, %o0+8+8))
- EX_LD(LOAD(ldx, %o1+16, %o4))
- EX_ST(STORE(stx, %o4, %o0+8+16))
+ EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl_o2_plus_o5)
+ EX_ST(STORE(stx, %o4, %o0+8), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1+8, %o4), memcpy_retl_o2_plus_o5)
+ EX_ST(STORE(stx, %o4, %o0+8+8), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1+16, %o4), memcpy_retl_o2_plus_o5)
+ EX_ST(STORE(stx, %o4, %o0+8+16), memcpy_retl_o2_plus_o5)
subcc %o5, 64, %o5
- EX_LD(LOAD(ldx, %o1+24, %o4))
- EX_ST(STORE(stx, %o4, %o0+8+24))
- EX_LD(LOAD(ldx, %o1+32, %o4))
- EX_ST(STORE(stx, %o4, %o0+8+32))
- EX_LD(LOAD(ldx, %o1+40, %o4))
- EX_ST(STORE(stx, %o4, %o0+8+40))
- EX_LD(LOAD(ldx, %o1+48, %o4))
+ EX_LD(LOAD(ldx, %o1+24, %o4), memcpy_retl_o2_plus_o5_64)
+ EX_ST(STORE(stx, %o4, %o0+8+24), memcpy_retl_o2_plus_o5_64)
+ EX_LD(LOAD(ldx, %o1+32, %o4), memcpy_retl_o2_plus_o5_64)
+ EX_ST(STORE(stx, %o4, %o0+8+32), memcpy_retl_o2_plus_o5_64)
+ EX_LD(LOAD(ldx, %o1+40, %o4), memcpy_retl_o2_plus_o5_64)
+ EX_ST(STORE(stx, %o4, %o0+8+40), memcpy_retl_o2_plus_o5_64)
+ EX_LD(LOAD(ldx, %o1+48, %o4), memcpy_retl_o2_plus_o5_64)
add %o1, 64, %o1
- EX_ST(STORE(stx, %o4, %o0+8+48))
+ EX_ST(STORE(stx, %o4, %o0+8+48), memcpy_retl_o2_plus_o5_64)
add %o0, 64, %o0
- EX_LD(LOAD(ldx, %o1-8, %o4))
+ EX_LD(LOAD(ldx, %o1-8, %o4), memcpy_retl_o2_plus_o5_64)
bgu %ncc,.Lalign_loop_fin
- EX_ST(STORE(stx, %o4, %o0))
+ EX_ST(STORE(stx, %o4, %o0), memcpy_retl_o2_plus_o5_64)
.Lalign_done:
add %o0, 8, %o0 ! restore %o0 from ASI alignment
@@ -563,34 +563,34 @@ FUNC_NAME:
.Lunalign_adjust:
alignaddr %o1, %g0, %g0 ! generate %gsr
add %o1, %o5, %o1 ! advance %o1 to after blocks
- EX_LD_FP(LOAD(ldd, %o4, %f0))
+ EX_LD_FP(LOAD(ldd, %o4, %f0), memcpy_retl_o2_plus_o5)
.Lunalign_loop:
- EX_LD_FP(LOAD(ldd, %o4+8, %f2))
+ EX_LD_FP(LOAD(ldd, %o4+8, %f2), memcpy_retl_o2_plus_o5)
faligndata %f0, %f2, %f16
- EX_LD_FP(LOAD(ldd, %o4+16, %f4))
+ EX_LD_FP(LOAD(ldd, %o4+16, %f4), memcpy_retl_o2_plus_o5)
subcc %o5, BLOCK_SIZE, %o5
- EX_ST_FP(STORE(std, %f16, %o0))
+ EX_ST_FP(STORE(std, %f16, %o0), memcpy_retl_o2_plus_o5_plus_64)
faligndata %f2, %f4, %f18
- EX_LD_FP(LOAD(ldd, %o4+24, %f6))
- EX_ST_FP(STORE(std, %f18, %o0+8))
+ EX_LD_FP(LOAD(ldd, %o4+24, %f6), memcpy_retl_o2_plus_o5_plus_56)
+ EX_ST_FP(STORE(std, %f18, %o0+8), memcpy_retl_o2_plus_o5_plus_56)
faligndata %f4, %f6, %f20
- EX_LD_FP(LOAD(ldd, %o4+32, %f8))
- EX_ST_FP(STORE(std, %f20, %o0+16))
+ EX_LD_FP(LOAD(ldd, %o4+32, %f8), memcpy_retl_o2_plus_o5_plus_48)
+ EX_ST_FP(STORE(std, %f20, %o0+16), memcpy_retl_o2_plus_o5_plus_48)
faligndata %f6, %f8, %f22
- EX_LD_FP(LOAD(ldd, %o4+40, %f10))
- EX_ST_FP(STORE(std, %f22, %o0+24))
+ EX_LD_FP(LOAD(ldd, %o4+40, %f10), memcpy_retl_o2_plus_o5_plus_40)
+ EX_ST_FP(STORE(std, %f22, %o0+24), memcpy_retl_o2_plus_o5_plus_40)
faligndata %f8, %f10, %f24
- EX_LD_FP(LOAD(ldd, %o4+48, %f12))
- EX_ST_FP(STORE(std, %f24, %o0+32))
+ EX_LD_FP(LOAD(ldd, %o4+48, %f12), memcpy_retl_o2_plus_o5_plus_32)
+ EX_ST_FP(STORE(std, %f24, %o0+32), memcpy_retl_o2_plus_o5_plus_32)
faligndata %f10, %f12, %f26
- EX_LD_FP(LOAD(ldd, %o4+56, %f14))
+ EX_LD_FP(LOAD(ldd, %o4+56, %f14), memcpy_retl_o2_plus_o5_plus_24)
add %o4, BLOCK_SIZE, %o4
- EX_ST_FP(STORE(std, %f26, %o0+40))
+ EX_ST_FP(STORE(std, %f26, %o0+40), memcpy_retl_o2_plus_o5_plus_24)
faligndata %f12, %f14, %f28
- EX_LD_FP(LOAD(ldd, %o4, %f0))
- EX_ST_FP(STORE(std, %f28, %o0+48))
+ EX_LD_FP(LOAD(ldd, %o4, %f0), memcpy_retl_o2_plus_o5_plus_16)
+ EX_ST_FP(STORE(std, %f28, %o0+48), memcpy_retl_o2_plus_o5_plus_16)
faligndata %f14, %f0, %f30
- EX_ST_FP(STORE(std, %f30, %o0+56))
+ EX_ST_FP(STORE(std, %f30, %o0+56), memcpy_retl_o2_plus_o5_plus_8)
add %o0, BLOCK_SIZE, %o0
bgu,pt %ncc, .Lunalign_loop
prefetch [%o4 + (5 * BLOCK_SIZE)], 20
@@ -612,27 +612,27 @@ FUNC_NAME:
nop
! Src is word aligned
.Lunalignword:
- EX_LD_FP(LOAD(ld, %o1, %o4)) ! load 4 bytes
+ EX_LD_FP(LOAD(ld, %o1, %o4), memcpy_retl_o2_plus_o3) ! load 4 bytes
add %o1, 8, %o1 ! increase src ptr by 8
- EX_ST_FP(STORE(stw, %o4, %o0)) ! and store 4 bytes
+ EX_ST_FP(STORE(stw, %o4, %o0), memcpy_retl_o2_plus_o3) ! and store 4
subcc %o3, 8, %o3 ! decrease count by 8
- EX_LD_FP(LOAD(ld, %o1-4, %o4)) ! load 4 bytes
+ EX_LD_FP(LOAD(ld, %o1-4, %o4), memcpy_retl_o2_plus_o3_plus_4)! load 4
add %o0, 8, %o0 ! increase dst ptr by 8
bnz %ncc, .Lunalignword
- EX_ST_FP(STORE(stw, %o4, %o0-4))! and store 4 bytes
+ EX_ST_FP(STORE(stw, %o4, %o0-4), memcpy_retl_o2_plus_o3_plus_4)
ba .Lunalignsrc
nop
! Src is half-word aligned
.Lunalignhalf:
- EX_LD_FP(LOAD(lduh, %o1, %o4)) ! load 2 bytes
+ EX_LD_FP(LOAD(lduh, %o1, %o4), memcpy_retl_o2_plus_o3) ! load 2 bytes
sllx %o4, 32, %o5 ! shift left
- EX_LD_FP(LOAD(lduw, %o1+2, %o4))
+ EX_LD_FP(LOAD(lduw, %o1+2, %o4), memcpy_retl_o2_plus_o3)
or %o4, %o5, %o5
sllx %o5, 16, %o5
- EX_LD_FP(LOAD(lduh, %o1+6, %o4))
+ EX_LD_FP(LOAD(lduh, %o1+6, %o4), memcpy_retl_o2_plus_o3)
or %o4, %o5, %o5
- EX_ST_FP(STORE(stx, %o5, %o0))
+ EX_ST_FP(STORE(stx, %o5, %o0), memcpy_retl_o2_plus_o3)
add %o1, 8, %o1
subcc %o3, 8, %o3
bnz %ncc, .Lunalignhalf
@@ -644,21 +644,21 @@ FUNC_NAME:
.Lunalignbyte:
sub %o0, %o1, %o0 ! share pointer advance
.Lunalignbyte_loop:
- EX_LD_FP(LOAD(ldub, %o1, %o4))
+ EX_LD_FP(LOAD(ldub, %o1, %o4), memcpy_retl_o2_plus_o3)
sllx %o4, 56, %o5
- EX_LD_FP(LOAD(lduh, %o1+1, %o4))
+ EX_LD_FP(LOAD(lduh, %o1+1, %o4), memcpy_retl_o2_plus_o3)
sllx %o4, 40, %o4
or %o4, %o5, %o5
- EX_LD_FP(LOAD(lduh, %o1+3, %o4))
+ EX_LD_FP(LOAD(lduh, %o1+3, %o4), memcpy_retl_o2_plus_o3)
sllx %o4, 24, %o4
or %o4, %o5, %o5
- EX_LD_FP(LOAD(lduh, %o1+5, %o4))
+ EX_LD_FP(LOAD(lduh, %o1+5, %o4), memcpy_retl_o2_plus_o3)
sllx %o4, 8, %o4
or %o4, %o5, %o5
- EX_LD_FP(LOAD(ldub, %o1+7, %o4))
+ EX_LD_FP(LOAD(ldub, %o1+7, %o4), memcpy_retl_o2_plus_o3)
or %o4, %o5, %o5
add %o0, %o1, %o0
- EX_ST_FP(STORE(stx, %o5, %o0))
+ EX_ST_FP(STORE(stx, %o5, %o0), memcpy_retl_o2_plus_o3)
sub %o0, %o1, %o0
subcc %o3, 8, %o3
bnz %ncc, .Lunalignbyte_loop
@@ -676,36 +676,36 @@ FUNC_NAME:
alignaddr %o1, %g0, %g0 ! generate %gsr
add %o1, %o5, %o1 ! advance %o1 to after blocks
- EX_LD_FP(LOAD(ldd, %o4, %f14))
+ EX_LD_FP(LOAD(ldd, %o4, %f14), memcpy_retl_o2_plus_o5)
add %o4, 8, %o4
.Lunalign_sloop:
- EX_LD_FP(LOAD(ldd, %o4, %f16))
+ EX_LD_FP(LOAD(ldd, %o4, %f16), memcpy_retl_o2_plus_o5)
faligndata %f14, %f16, %f0
- EX_LD_FP(LOAD(ldd, %o4+8, %f18))
+ EX_LD_FP(LOAD(ldd, %o4+8, %f18), memcpy_retl_o2_plus_o5)
faligndata %f16, %f18, %f2
- EX_LD_FP(LOAD(ldd, %o4+16, %f20))
+ EX_LD_FP(LOAD(ldd, %o4+16, %f20), memcpy_retl_o2_plus_o5)
faligndata %f18, %f20, %f4
- EX_ST_FP(STORE(std, %f0, %o0))
+ EX_ST_FP(STORE(std, %f0, %o0), memcpy_retl_o2_plus_o5)
subcc %o5, 64, %o5
- EX_LD_FP(LOAD(ldd, %o4+24, %f22))
+ EX_LD_FP(LOAD(ldd, %o4+24, %f22), memcpy_retl_o2_plus_o5_plus_56)
faligndata %f20, %f22, %f6
- EX_ST_FP(STORE(std, %f2, %o0+8))
- EX_LD_FP(LOAD(ldd, %o4+32, %f24))
+ EX_ST_FP(STORE(std, %f2, %o0+8), memcpy_retl_o2_plus_o5_plus_56)
+ EX_LD_FP(LOAD(ldd, %o4+32, %f24), memcpy_retl_o2_plus_o5_plus_48)
faligndata %f22, %f24, %f8
- EX_ST_FP(STORE(std, %f4, %o0+16))
- EX_LD_FP(LOAD(ldd, %o4+40, %f26))
+ EX_ST_FP(STORE(std, %f4, %o0+16), memcpy_retl_o2_plus_o5_plus_48)
+ EX_LD_FP(LOAD(ldd, %o4+40, %f26), memcpy_retl_o2_plus_o5_plus_40)
faligndata %f24, %f26, %f10
- EX_ST_FP(STORE(std, %f6, %o0+24))
- EX_LD_FP(LOAD(ldd, %o4+48, %f28))
+ EX_ST_FP(STORE(std, %f6, %o0+24), memcpy_retl_o2_plus_o5_plus_40)
+ EX_LD_FP(LOAD(ldd, %o4+48, %f28), memcpy_retl_o2_plus_o5_plus_40)
faligndata %f26, %f28, %f12
- EX_ST_FP(STORE(std, %f8, %o0+32))
+ EX_ST_FP(STORE(std, %f8, %o0+32), memcpy_retl_o2_plus_o5_plus_40)
add %o4, 64, %o4
- EX_LD_FP(LOAD(ldd, %o4-8, %f30))
+ EX_LD_FP(LOAD(ldd, %o4-8, %f30), memcpy_retl_o2_plus_o5_plus_40)
faligndata %f28, %f30, %f14
- EX_ST_FP(STORE(std, %f10, %o0+40))
- EX_ST_FP(STORE(std, %f12, %o0+48))
+ EX_ST_FP(STORE(std, %f10, %o0+40), memcpy_retl_o2_plus_o5_plus_40)
+ EX_ST_FP(STORE(std, %f12, %o0+48), memcpy_retl_o2_plus_o5_plus_40)
add %o0, 64, %o0
- EX_ST_FP(STORE(std, %f14, %o0-8))
+ EX_ST_FP(STORE(std, %f14, %o0-8), memcpy_retl_o2_plus_o5_plus_40)
fsrc2 %f30, %f14
bgu,pt %ncc, .Lunalign_sloop
prefetch [%o4 + (8 * BLOCK_SIZE)], 20
@@ -722,13 +722,13 @@ FUNC_NAME:
sub %o5, 8, %o5 ! insure we do not load past end of src
andn %o1, 0x7, %o4 ! %o4 has long word aligned src address
add %o1, %o5, %o1 ! advance %o1 to after multiple of 8
- EX_LD_FP(LOAD(ldd, %o4, %f0)) ! fetch partial word
+ EX_LD_FP(LOAD(ldd, %o4, %f0), memcpy_retl_o2_plus_o5)! fetch partialword
.Lunalign_by8:
- EX_LD_FP(LOAD(ldd, %o4+8, %f2))
+ EX_LD_FP(LOAD(ldd, %o4+8, %f2), memcpy_retl_o2_plus_o5)
add %o4, 8, %o4
faligndata %f0, %f2, %f16
subcc %o5, 8, %o5
- EX_ST_FP(STORE(std, %f16, %o0))
+ EX_ST_FP(STORE(std, %f16, %o0), memcpy_retl_o2_plus_o5)
fsrc2 %f2, %f0
bgu,pt %ncc, .Lunalign_by8
add %o0, 8, %o0
@@ -764,37 +764,37 @@ FUNC_NAME:
andncc %o2, 0x20 - 1, %o5
be,pn %ncc, 2f
sub %o2, %o5, %o2
-1: EX_LD(LOAD(ldx, %o1 + 0x00, %o3))
- EX_LD(LOAD(ldx, %o1 + 0x08, %g2))
- EX_LD(LOAD(ldx, %o1 + 0x10, %g7))
- EX_LD(LOAD(ldx, %o1 + 0x18, %o4))
+1: EX_LD(LOAD(ldx, %o1 + 0x00, %o3), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1 + 0x08, %g2), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1 + 0x10, %g7), memcpy_retl_o2_plus_o5)
+ EX_LD(LOAD(ldx, %o1 + 0x18, %o4), memcpy_retl_o2_plus_o5)
add %o1, 0x20, %o1
subcc %o5, 0x20, %o5
- EX_ST(STORE(stx, %o3, %o0 + 0x00))
- EX_ST(STORE(stx, %g2, %o0 + 0x08))
- EX_ST(STORE(stx, %g7, %o0 + 0x10))
- EX_ST(STORE(stx, %o4, %o0 + 0x18))
+ EX_ST(STORE(stx, %o3, %o0 + 0x00), memcpy_retl_o2_plus_o5_plus_32)
+ EX_ST(STORE(stx, %g2, %o0 + 0x08), memcpy_retl_o2_plus_o5_plus_24)
+ EX_ST(STORE(stx, %g7, %o0 + 0x10), memcpy_retl_o2_plus_o5_plus_24)
+ EX_ST(STORE(stx, %o4, %o0 + 0x18), memcpy_retl_o2_plus_o5_plus_8)
bne,pt %ncc, 1b
add %o0, 0x20, %o0
2: andcc %o2, 0x18, %o5
be,pt %ncc, 3f
sub %o2, %o5, %o2
-1: EX_LD(LOAD(ldx, %o1 + 0x00, %o3))
+1: EX_LD(LOAD(ldx, %o1 + 0x00, %o3), memcpy_retl_o2_plus_o5)
add %o1, 0x08, %o1
add %o0, 0x08, %o0
subcc %o5, 0x08, %o5
bne,pt %ncc, 1b
- EX_ST(STORE(stx, %o3, %o0 - 0x08))
+ EX_ST(STORE(stx, %o3, %o0 - 0x08), memcpy_retl_o2_plus_o5_plus_8)
3: brz,pt %o2, .Lexit_cp
cmp %o2, 0x04
bl,pn %ncc, .Ltiny_cp
nop
- EX_LD(LOAD(lduw, %o1 + 0x00, %o3))
+ EX_LD(LOAD(lduw, %o1 + 0x00, %o3), memcpy_retl_o2)
add %o1, 0x04, %o1
add %o0, 0x04, %o0
subcc %o2, 0x04, %o2
bne,pn %ncc, .Ltiny_cp
- EX_ST(STORE(stw, %o3, %o0 - 0x04))
+ EX_ST(STORE(stw, %o3, %o0 - 0x04), memcpy_retl_o2_plus_4)
ba,a,pt %ncc, .Lexit_cp
.Lmedium_unaligned_cp:
@@ -804,12 +804,12 @@ FUNC_NAME:
brz,pt %o3, 2f
sub %o2, %o3, %o2
-1: EX_LD(LOAD(ldub, %o1 + 0x00, %g2))
+1: EX_LD(LOAD(ldub, %o1 + 0x00, %g2), memcpy_retl_o2_plus_g1)
add %o1, 1, %o1
subcc %o3, 1, %o3
add %o0, 1, %o0
bne,pt %ncc, 1b
- EX_ST(STORE(stb, %g2, %o0 - 0x01))
+ EX_ST(STORE(stb, %g2, %o0 - 0x01), memcpy_retl_o2_plus_g1_plus_1)
2:
and %o1, 0x7, %o3
brz,pn %o3, .Lmedium_noprefetch_cp
@@ -817,17 +817,17 @@ FUNC_NAME:
mov 64, %g2
sub %g2, %o3, %g2
andn %o1, 0x7, %o1
- EX_LD(LOAD(ldx, %o1 + 0x00, %o4))
+ EX_LD(LOAD(ldx, %o1 + 0x00, %o4), memcpy_retl_o2)
sllx %o4, %o3, %o4
andn %o2, 0x08 - 1, %o5
sub %o2, %o5, %o2
-1: EX_LD(LOAD(ldx, %o1 + 0x08, %g3))
+1: EX_LD(LOAD(ldx, %o1 + 0x08, %g3), memcpy_retl_o2_plus_o5)
add %o1, 0x08, %o1
subcc %o5, 0x08, %o5
srlx %g3, %g2, %g7
or %g7, %o4, %g7
- EX_ST(STORE(stx, %g7, %o0 + 0x00))
+ EX_ST(STORE(stx, %g7, %o0 + 0x00), memcpy_retl_o2_plus_o5_plus_8)
add %o0, 0x08, %o0
bne,pt %ncc, 1b
sllx %g3, %o3, %o4
@@ -838,17 +838,17 @@ FUNC_NAME:
ba,pt %ncc, .Lsmall_unaligned_cp
.Ltiny_cp:
- EX_LD(LOAD(ldub, %o1 + 0x00, %o3))
+ EX_LD(LOAD(ldub, %o1 + 0x00, %o3), memcpy_retl_o2)
subcc %o2, 1, %o2
be,pn %ncc, .Lexit_cp
- EX_ST(STORE(stb, %o3, %o0 + 0x00))
- EX_LD(LOAD(ldub, %o1 + 0x01, %o3))
+ EX_ST(STORE(stb, %o3, %o0 + 0x00), memcpy_retl_o2_plus_1)
+ EX_LD(LOAD(ldub, %o1 + 0x01, %o3), memcpy_retl_o2)
subcc %o2, 1, %o2
be,pn %ncc, .Lexit_cp
- EX_ST(STORE(stb, %o3, %o0 + 0x01))
- EX_LD(LOAD(ldub, %o1 + 0x02, %o3))
+ EX_ST(STORE(stb, %o3, %o0 + 0x01), memcpy_retl_o2_plus_1)
+ EX_LD(LOAD(ldub, %o1 + 0x02, %o3), memcpy_retl_o2)
ba,pt %ncc, .Lexit_cp
- EX_ST(STORE(stb, %o3, %o0 + 0x02))
+ EX_ST(STORE(stb, %o3, %o0 + 0x02), memcpy_retl_o2)
.Lsmall_cp:
andcc %g2, 0x3, %g0
@@ -856,23 +856,23 @@ FUNC_NAME:
andn %o2, 0x4 - 1, %o5
sub %o2, %o5, %o2
1:
- EX_LD(LOAD(lduw, %o1 + 0x00, %o3))
+ EX_LD(LOAD(lduw, %o1 + 0x00, %o3), memcpy_retl_o2_plus_o5)
add %o1, 0x04, %o1
subcc %o5, 0x04, %o5
add %o0, 0x04, %o0
bne,pt %ncc, 1b
- EX_ST(STORE(stw, %o3, %o0 - 0x04))
+ EX_ST(STORE(stw, %o3, %o0 - 0x04), memcpy_retl_o2_plus_o5_plus_4)
brz,pt %o2, .Lexit_cp
nop
ba,a,pt %ncc, .Ltiny_cp
.Lsmall_unaligned_cp:
-1: EX_LD(LOAD(ldub, %o1 + 0x00, %o3))
+1: EX_LD(LOAD(ldub, %o1 + 0x00, %o3), memcpy_retl_o2)
add %o1, 1, %o1
add %o0, 1, %o0
subcc %o2, 1, %o2
bne,pt %ncc, 1b
- EX_ST(STORE(stb, %o3, %o0 - 0x01))
+ EX_ST(STORE(stb, %o3, %o0 - 0x01), memcpy_retl_o2_plus_1)
ba,a,pt %ncc, .Lexit_cp
.Lsmallrest:
@@ -883,31 +883,31 @@ FUNC_NAME:
nop
sub %o2, 3, %o2
.Lsmallnotalign4:
- EX_LD(LOAD(ldub, %o1, %o3))! read byte
+ EX_LD(LOAD(ldub, %o1, %o3), memcpy_retl_o2_plus_3)! read byte
subcc %o2, 4, %o2 ! reduce count by 4
- EX_ST(STORE(stb, %o3, %o0)) ! write byte
- EX_LD(LOAD(ldub, %o1+1, %o3))! repeat for total of 4 bytes
+ EX_ST(STORE(stb, %o3, %o0), memcpy_retl_o2_plus_7)! write byte & repeat
+ EX_LD(LOAD(ldub, %o1+1, %o3), memcpy_retl_o2_plus_6)! for total of 4
add %o1, 4, %o1 ! advance SRC by 4
- EX_ST(STORE(stb, %o3, %o0+1))
- EX_LD(LOAD(ldub, %o1-2, %o3))
+ EX_ST(STORE(stb, %o3, %o0+1), memcpy_retl_o2_plus_6)
+ EX_LD(LOAD(ldub, %o1-2, %o3), memcpy_retl_o2_plus_5)
add %o0, 4, %o0 ! advance DST by 4
- EX_ST(STORE(stb, %o3, %o0-2))
- EX_LD(LOAD(ldub, %o1-1, %o3))
+ EX_ST(STORE(stb, %o3, %o0-2), memcpy_retl_o2_plus_5)
+ EX_LD(LOAD(ldub, %o1-1, %o3), memcpy_retl_o2_plus_4)
bgu,pt %ncc, .Lsmallnotalign4 ! loop til 3 or fewer bytes remain
- EX_ST(STORE(stb, %o3, %o0-1))
+ EX_ST(STORE(stb, %o3, %o0-1), memcpy_retl_o2_plus_4)
addcc %o2, 3, %o2 ! restore count
bz,pt %ncc, .Lsmallx
.Lsmallleft3: ! 1, 2, or 3 bytes remain
subcc %o2, 1, %o2
- EX_LD(LOAD(ldub, %o1, %o3)) ! load one byte
+ EX_LD(LOAD(ldub, %o1, %o3), memcpy_retl_o2_plus_1) ! load one byte
bz,pt %ncc, .Lsmallx
- EX_ST(STORE(stb, %o3, %o0)) ! store one byte
- EX_LD(LOAD(ldub, %o1+1, %o3)) ! load second byte
+ EX_ST(STORE(stb, %o3, %o0), memcpy_retl_o2_plus_1) ! store one byte
+ EX_LD(LOAD(ldub, %o1+1, %o3), memcpy_retl_o2) ! load second byte
subcc %o2, 1, %o2
bz,pt %ncc, .Lsmallx
- EX_ST(STORE(stb, %o3, %o0+1))! store second byte
- EX_LD(LOAD(ldub, %o1+2, %o3)) ! load third byte
- EX_ST(STORE(stb, %o3, %o0+2)) ! store third byte
+ EX_ST(STORE(stb, %o3, %o0+1), memcpy_retl_o2_plus_1)! store second byte
+ EX_LD(LOAD(ldub, %o1+2, %o3), memcpy_retl_o2) ! load third byte
+ EX_ST(STORE(stb, %o3, %o0+2), memcpy_retl_o2) ! store third byte
.Lsmallx:
retl
mov EX_RETVAL(%g1), %o0
diff --git a/arch/sparc/lib/Memcpy_utils.S b/arch/sparc/lib/Memcpy_utils.S
index bcc5d77..64fbac2 100644
--- a/arch/sparc/lib/Memcpy_utils.S
+++ b/arch/sparc/lib/Memcpy_utils.S
@@ -24,14 +24,128 @@ ENTRY(memcpy_retl_o2_plus_1)
ba,pt %xcc, __restore_asi
add %o2, 1, %o0
ENDPROC(memcpy_retl_o2_plus_1)
+ENTRY(memcpy_retl_o2_plus_3)
+ ba,pt %xcc, __restore_asi
+ add %o2, 3, %o0
+ENDPROC(memcpy_retl_o2_plus_3)
ENTRY(memcpy_retl_o2_plus_4)
ba,pt %xcc, __restore_asi
add %o2, 4, %o0
ENDPROC(memcpy_retl_o2_plus_4)
+ENTRY(memcpy_retl_o2_plus_5)
+ ba,pt %xcc, __restore_asi
+ add %o2, 5, %o0
+ENDPROC(memcpy_retl_o2_plus_5)
+ENTRY(memcpy_retl_o2_plus_6)
+ ba,pt %xcc, __restore_asi
+ add %o2, 6, %o0
+ENDPROC(memcpy_retl_o2_plus_6)
+ENTRY(memcpy_retl_o2_plus_7)
+ ba,pt %xcc, __restore_asi
+ add %o2, 7, %o0
+ENDPROC(memcpy_retl_o2_plus_7)
+ENTRY(memcpy_retl_o2_plus_8)
+ ba,pt %xcc, __restore_asi
+ add %o2, 8, %o0
+ENDPROC(memcpy_retl_o2_plus_8)
+ENTRY(memcpy_retl_o2_plus_15)
+ ba,pt %xcc, __restore_asi
+ add %o2, 15, %o0
+ENDPROC(memcpy_retl_o2_plus_15)
+ENTRY(memcpy_retl_o2_plus_15_8)
+ add %o2, 15, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 8, %o0
+ENDPROC(memcpy_retl_o2_plus_15_8)
+ENTRY(memcpy_retl_o2_plus_16)
+ ba,pt %xcc, __restore_asi
+ add %o2, 16, %o0
+ENDPROC(memcpy_retl_o2_plus_16)
+ENTRY(memcpy_retl_o2_plus_24)
+ ba,pt %xcc, __restore_asi
+ add %o2, 24, %o0
+ENDPROC(memcpy_retl_o2_plus_24)
+ENTRY(memcpy_retl_o2_plus_31)
+ ba,pt %xcc, __restore_asi
+ add %o2, 31, %o0
+ENDPROC(memcpy_retl_o2_plus_31)
+ENTRY(memcpy_retl_o2_plus_32)
+ ba,pt %xcc, __restore_asi
+ add %o2, 32, %o0
+ENDPROC(memcpy_retl_o2_plus_32)
+ENTRY(memcpy_retl_o2_plus_31_32)
+ add %o2, 31, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 32, %o0
+ENDPROC(memcpy_retl_o2_plus_31_32)
+ENTRY(memcpy_retl_o2_plus_31_24)
+ add %o2, 31, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 24, %o0
+ENDPROC(memcpy_retl_o2_plus_31_24)
+ENTRY(memcpy_retl_o2_plus_31_16)
+ add %o2, 31, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 16, %o0
+ENDPROC(memcpy_retl_o2_plus_31_16)
+ENTRY(memcpy_retl_o2_plus_31_8)
+ add %o2, 31, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 8, %o0
+ENDPROC(memcpy_retl_o2_plus_31_8)
+ENTRY(memcpy_retl_o2_plus_63)
+ ba,pt %xcc, __restore_asi
+ add %o2, 63, %o0
+ENDPROC(memcpy_retl_o2_plus_63)
+ENTRY(memcpy_retl_o2_plus_63_64)
+ add %o2, 63, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 64, %o0
+ENDPROC(memcpy_retl_o2_plus_63_64)
+ENTRY(memcpy_retl_o2_plus_63_56)
+ add %o2, 63, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 56, %o0
+ENDPROC(memcpy_retl_o2_plus_63_56)
+ENTRY(memcpy_retl_o2_plus_63_48)
+ add %o2, 63, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 48, %o0
+ENDPROC(memcpy_retl_o2_plus_63_48)
+ENTRY(memcpy_retl_o2_plus_63_40)
+ add %o2, 63, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 40, %o0
+ENDPROC(memcpy_retl_o2_plus_63_40)
+ENTRY(memcpy_retl_o2_plus_63_32)
+ add %o2, 63, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 32, %o0
+ENDPROC(memcpy_retl_o2_plus_63_32)
+ENTRY(memcpy_retl_o2_plus_63_24)
+ add %o2, 63, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 24, %o0
+ENDPROC(memcpy_retl_o2_plus_63_24)
+ENTRY(memcpy_retl_o2_plus_63_16)
+ add %o2, 63, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 16, %o0
+ENDPROC(memcpy_retl_o2_plus_63_16)
+ENTRY(memcpy_retl_o2_plus_63_8)
+ add %o2, 63, %o2
+ ba,pt %xcc, __restore_asi
+ add %o2, 8, %o0
+ENDPROC(memcpy_retl_o2_plus_63_8)
ENTRY(memcpy_retl_o2_plus_o5)
ba,pt %xcc, __restore_asi
add %o2, %o5, %o0
ENDPROC(memcpy_retl_o2_plus_o5)
+ENTRY(memcpy_retl_o2_plus_o5_plus_1)
+ add %o5, 1, %o5
+ ba,pt %xcc, __restore_asi
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_plus_1)
ENTRY(memcpy_retl_o2_plus_o5_plus_4)
add %o5, 4, %o5
ba,pt %xcc, __restore_asi
@@ -57,6 +171,11 @@ ENTRY(memcpy_retl_o2_plus_o5_plus_32)
ba,pt %xcc, __restore_asi
add %o2, %o5, %o0
ENDPROC(memcpy_retl_o2_plus_o5_plus_32)
+ENTRY(memcpy_retl_o2_plus_o5_64)
+ add %o5, 32, %o5
+ ba,pt %xcc, __restore_asi
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_64)
ENTRY(memcpy_retl_o2_plus_g1)
ba,pt %xcc, __restore_asi
add %o2, %g1, %o0
@@ -115,6 +234,25 @@ ENTRY(memcpy_retl_o2_plus_o4_plus_64)
ba,pt %xcc, __restore_asi
add %o2, %o4, %o0
ENDPROC(memcpy_retl_o2_plus_o4_plus_64)
+ENTRY(memcpy_retl_o2_plus_o5_plus_64)
+ add %o5, 64, %o5
+ ba,pt %xcc, __restore_asi
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_plus_64)
+ENTRY(memcpy_retl_o2_plus_o3_fp)
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o3, %o0
+ENDPROC(memcpy_retl_o2_plus_o3_fp)
+ENTRY(memcpy_retl_o2_plus_o3_plus_1_fp)
+ add %o3, 1, %o3
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o3, %o0
+ENDPROC(memcpy_retl_o2_plus_o3_plus_1_fp)
+ENTRY(memcpy_retl_o2_plus_o3_plus_4_fp)
+ add %o3, 4, %o3
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o3, %o0
+ENDPROC(memcpy_retl_o2_plus_o3_plus_4_fp)
ENTRY(memcpy_retl_o2_plus_o4_fp)
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
@@ -159,5 +297,49 @@ ENTRY(memcpy_retl_o2_plus_o4_plus_64_fp)
ba,pt %xcc, __restore_asi_fp
add %o2, %o4, %o0
ENDPROC(memcpy_retl_o2_plus_o4_plus_64_fp)
+ENTRY(memcpy_retl_o2_plus_o5_fp)
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_fp)
+ENTRY(memcpy_retl_o2_plus_o5_plus_64_fp)
+ add %o5, 64, %o5
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_plus_64_fp)
+ENTRY(memcpy_retl_o2_plus_o5_plus_56_fp)
+ add %o5, 56, %o5
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_plus_56_fp)
+ENTRY(memcpy_retl_o2_plus_o5_plus_48_fp)
+ add %o5, 48, %o5
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_plus_48_fp)
+ENTRY(memcpy_retl_o2_plus_o5_plus_40_fp)
+ add %o5, 40, %o5
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_plus_40_fp)
+ENTRY(memcpy_retl_o2_plus_o5_plus_32_fp)
+ add %o5, 32, %o5
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_plus_32_fp)
+ENTRY(memcpy_retl_o2_plus_o5_plus_24_fp)
+ add %o5, 24, %o5
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_plus_24_fp)
+ENTRY(memcpy_retl_o2_plus_o5_plus_16_fp)
+ add %o5, 16, %o5
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_plus_16_fp)
+ENTRY(memcpy_retl_o2_plus_o5_plus_8_fp)
+ add %o5, 8, %o5
+ ba,pt %xcc, __restore_asi_fp
+ add %o2, %o5, %o0
+ENDPROC(memcpy_retl_o2_plus_o5_plus_8_fp)
#endif
--
1.7.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures
2017-08-07 23:52 [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures Babu Moger
` (3 preceding siblings ...)
2017-08-07 23:52 ` [PATCH v2 4/4] arch/sparc: Add accurate exception reporting in M7memcpy Babu Moger
@ 2017-08-10 21:38 ` David Miller
2017-08-12 0:25 ` Babu Moger
4 siblings, 1 reply; 7+ messages in thread
From: David Miller @ 2017-08-10 21:38 UTC (permalink / raw)
To: babu.moger; +Cc: sparclinux, linux-kernel
From: Babu Moger <babu.moger@oracle.com>
Date: Mon, 7 Aug 2017 17:52:48 -0600
> This series of patches updates the memcpy, memset, copy_to_user,
> copy_from_user etc for SPARC M7/M8 architecture.
This doesn't build, you cannot assume the existence of "%ncc", it is a
recent addition.
Furthermore there is no need to ever use %ncc in v9 targetted code
anyways.
I'll fix that up, but this was a really disappointing build failure
to hit.
Meanwhile, two questions:
1) Is this also faster on T4 as well? If it is, we can just get rid
of the T4 routines and use this on those chips as well.
2) There has been a lot of discussion and consideration put into how
a memcpy/memset routine might be really great for the local cpu
but overall pessimize performance for other cpus either locally
on the same core (contention for physical resources such as
ports to the store buffer and/or L3 cache) or on other cores.
Has any such study been done into these issues wrt. this new code?
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures
2017-08-10 21:38 ` [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures David Miller
@ 2017-08-12 0:25 ` Babu Moger
0 siblings, 0 replies; 7+ messages in thread
From: Babu Moger @ 2017-08-12 0:25 UTC (permalink / raw)
To: David Miller; +Cc: sparclinux, linux-kernel
David, Thanks for applying.
On 8/10/2017 4:38 PM, David Miller wrote:
> From: Babu Moger <babu.moger@oracle.com>
> Date: Mon, 7 Aug 2017 17:52:48 -0600
>
>> This series of patches updates the memcpy, memset, copy_to_user,
>> copy_from_user etc for SPARC M7/M8 architecture.
> This doesn't build, you cannot assume the existence of "%ncc", it is a
> recent addition.
>
> Furthermore there is no need to ever use %ncc in v9 targetted code
> anyways.
>
> I'll fix that up, but this was a really disappointing build failure
> to hit.
Thank you..
>
> Meanwhile, two questions:
>
> 1) Is this also faster on T4 as well? If it is, we can just get rid
> of the T4 routines and use this on those chips as well.
At the time of this work, our focus was mostly on T7 and T8. We did not
test this code on T4.
For T4 and other older configs we used NG4 versions. I would think it
would require some
changes to make it work on T4.
> 2) There has been a lot of discussion and consideration put into how
> a memcpy/memset routine might be really great for the local cpu
> but overall pessimize performance for other cpus either locally
> on the same core (contention for physical resources such as
> ports to the store buffer and/or L3 cache) or on other cores.
>
> Has any such study been done into these issues wrt. this new code?
No, we have not done this kind of study.
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2017-08-12 0:25 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-08-07 23:52 [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures Babu Moger
2017-08-07 23:52 ` [PATCH v2 1/4] arch/sparc: Separate the exception handlers from NG4memcpy Babu Moger
2017-08-07 23:52 ` [PATCH v2 2/4] arch/sparc: Rename exception handlers Babu Moger
2017-08-07 23:52 ` [PATCH v2 3/4] arch/sparc: Optimized memcpy, memset, copy_to_user, copy_from_user for M7/M8 Babu Moger
2017-08-07 23:52 ` [PATCH v2 4/4] arch/sparc: Add accurate exception reporting in M7memcpy Babu Moger
2017-08-10 21:38 ` [PATCH v2 0/4] Update memcpy, memset etc. for M7/M8 architectures David Miller
2017-08-12 0:25 ` Babu Moger
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).