runtime: improve CALLFN macro for ppc64x

The previous CALLFN macro was copying a single byte at a
time which is extremely inefficient on ppc64x. This changes
the macro so it copies 8 bytes at a time.

benchmark in reflect:
name                      old time/op    new time/op    delta
Call-8                       177ns ± 0%     165ns ± 0%    -6.78%  (p=1.000 n=1+1)
CallArgCopy/size=128-8       194ns ± 0%     140ns ± 0%   -27.84%  (p=1.000 n=1+1)
CallArgCopy/size=256-8       253ns ± 0%     159ns ± 0%   -37.15%  (p=1.000 n=1+1)
CallArgCopy/size=1024-8      612ns ± 0%     222ns ± 0%   -63.73%  (p=1.000 n=1+1)
CallArgCopy/size=4096-8     2.14µs ± 0%    0.53µs ± 0%   -75.01%  (p=1.000 n=1+1)
CallArgCopy/size=65536-8    33.0µs ± 0%     7.3µs ± 0%   -77.72%  (p=1.000 n=1+1)

Change-Id: I71f6ee788264e61bb072264d21b77b83592c9dca
Reviewed-on: https://go-review.googlesource.com/134635
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
Reviewed-by: Michael Munday <mike.munday@ibm.com>

diff --git a/src/runtime/asm_ppc64x.s b/src/runtime/asm_ppc64x.s
index 57877c0194c7dc403b4bac71f2d886bfdbb65c88..b180cb06ab9d0b3a5001f27f92aa098222729f31 100644 (file)
--- a/src/runtime/asm_ppc64x.s
+++ b/src/runtime/asm_ppc64x.s
@@ -390,15 +390,36 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-24;              \
        /* copy arguments to stack */           \
        MOVD    arg+16(FP), R3;                 \
        MOVWZ   argsize+24(FP), R4;                     \
-       MOVD    R1, R5;                         \
-       ADD     $(FIXED_FRAME-1), R5;                   \
-       SUB     $1, R3;                         \
-       ADD     R5, R4;                         \
-       CMP     R5, R4;                         \
-       BEQ     4(PC);                          \
-       MOVBZU  1(R3), R6;                      \
-       MOVBZU  R6, 1(R5);                      \
-       BR      -4(PC);                         \
+       MOVD    R1, R5;                         \
+       CMP     R4, $8;                         \
+       BLT     tailsetup;                      \
+       /* copy 8 at a time if possible */      \
+       ADD     $(FIXED_FRAME-8), R5;                   \
+       SUB     $8, R3;                         \
+top: \
+       MOVDU   8(R3), R7;                      \
+       MOVDU   R7, 8(R5);                      \
+       SUB     $8, R4;                         \
+       CMP     R4, $8;                         \
+       BGE     top;                            \
+       /* handle remaining bytes */    \
+       CMP     $0, R4;                 \
+       BEQ     callfn;                 \
+       ADD     $7, R3;                 \
+       ADD     $7, R5;                 \
+       BR      tail;                   \
+tailsetup: \
+       CMP     $0, R4;                 \
+       BEQ     callfn;                 \
+       ADD     $(FIXED_FRAME-1), R5;   \
+       SUB     $1, R3;                 \
+tail: \
+       MOVBU   1(R3), R6;              \
+       MOVBU   R6, 1(R5);              \
+       SUB     $1, R4;                 \
+       CMP     $0, R4;                 \
+       BGT     tail;                   \
+callfn: \
        /* call function */                     \
        MOVD    f+8(FP), R11;                   \
        MOVD    (R11), R12;                     \