cmd/compile: inline constant sized memclrNoHeapPointers calls on loong64

Tested that on loong64, the optimization effect is negative for
constant size cases greater than 512.
So only enable inlining for constant size cases less than 512.

goos: linux
goarch: loong64
pkg: runtime
cpu: Loongson-3A6000 @ 2500.00MHz
                      |  bench.old   |              bench.new               |
                      |    sec/op    |    sec/op     vs base                |
MemclrKnownSize1        2.4070n ± 0%   0.4004n ± 0%  -83.37% (p=0.000 n=20)
MemclrKnownSize2        2.1365n ± 0%   0.4004n ± 0%  -81.26% (p=0.000 n=20)
MemclrKnownSize4        2.4445n ± 0%   0.4004n ± 0%  -83.62% (p=0.000 n=20)
MemclrKnownSize8        2.4200n ± 0%   0.4004n ± 0%  -83.45% (p=0.000 n=20)
MemclrKnownSize16       2.8030n ± 0%   0.8007n ± 0%  -71.43% (p=0.000 n=20)
MemclrKnownSize32        2.803n ± 0%    1.602n ± 0%  -42.85% (p=0.000 n=20)
MemclrKnownSize64        3.250n ± 0%    2.402n ± 0%  -26.08% (p=0.000 n=20)
MemclrKnownSize112       6.006n ± 0%    2.819n ± 0%  -53.06% (p=0.000 n=20)
MemclrKnownSize128       6.006n ± 0%    3.240n ± 0%  -46.05% (p=0.000 n=20)
MemclrKnownSize192       6.807n ± 0%    5.205n ± 0%  -23.53% (p=0.000 n=20)
MemclrKnownSize248       7.608n ± 0%    6.301n ± 0%  -17.19% (p=0.000 n=20)
MemclrKnownSize256       7.608n ± 0%    6.707n ± 0%  -11.84% (p=0.000 n=20)
MemclrKnownSize512       13.61n ± 0%    13.61n ± 0%        ~ (p=0.374 n=20)
MemclrKnownSize1024      26.43n ± 0%    26.43n ± 0%        ~ (p=0.826 n=20)
MemclrKnownSize4096      103.3n ± 0%    103.3n ± 0%        ~ (p=1.000 n=20)
MemclrKnownSize512KiB    26.29µ ± 0%    26.29µ ± 0%   -0.00% (p=0.012 n=20)
geomean                  10.05n         5.006n       -50.18%

                      |  bench.old   |               bench.new                |
                      |     B/s      |      B/s       vs base                 |
MemclrKnownSize1        396.2Mi ± 0%   2381.9Mi ± 0%  +501.21% (p=0.000 n=20)
MemclrKnownSize2        892.8Mi ± 0%   4764.0Mi ± 0%  +433.59% (p=0.000 n=20)
MemclrKnownSize4        1.524Gi ± 0%    9.305Gi ± 0%  +510.56% (p=0.000 n=20)
MemclrKnownSize8        3.079Gi ± 0%   18.609Gi ± 0%  +504.42% (p=0.000 n=20)
MemclrKnownSize16       5.316Gi ± 0%   18.609Gi ± 0%  +250.05% (p=0.000 n=20)
MemclrKnownSize32       10.63Gi ± 0%    18.61Gi ± 0%   +75.00% (p=0.000 n=20)
MemclrKnownSize64       18.34Gi ± 0%    24.81Gi ± 0%   +35.27% (p=0.000 n=20)
MemclrKnownSize112      17.37Gi ± 0%    37.01Gi ± 0%  +113.08% (p=0.000 n=20)
MemclrKnownSize128      19.85Gi ± 0%    36.80Gi ± 0%   +85.39% (p=0.000 n=20)
MemclrKnownSize192      26.27Gi ± 0%    34.35Gi ± 0%   +30.77% (p=0.000 n=20)
MemclrKnownSize248      30.36Gi ± 0%    36.66Gi ± 0%   +20.75% (p=0.000 n=20)
MemclrKnownSize256      31.34Gi ± 0%    35.55Gi ± 0%   +13.43% (p=0.000 n=20)
MemclrKnownSize512      35.02Gi ± 0%    35.03Gi ± 0%    +0.00% (p=0.030 n=20)
MemclrKnownSize1024     36.09Gi ± 0%    36.09Gi ± 0%         ~ (p=0.101 n=20)
MemclrKnownSize4096     36.93Gi ± 0%    36.93Gi ± 0%    +0.00% (p=0.003 n=20)
MemclrKnownSize512KiB   18.57Gi ± 0%    18.57Gi ± 0%    +0.00% (p=0.041 n=20)
geomean                 10.13Gi         20.33Gi       +100.72%

Change-Id: I460a56f7ccc9f820ca2c1934c1c517b9614809ac
Reviewed-on: https://go-review.googlesource.com/c/go/+/621355
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Reviewed-by: abner chenc <chenguoqi@loongson.cn>
Reviewed-by: Michael Pratt <mpratt@google.com>

1 files changed, 6 insertions, 1 deletions

diff --git a/test/codegen/slices.go b/test/codegen/slices.go
index a38fe77e3f..9e8990c586 100644
--- a/test/codegen/slices.go
+++ b/test/codegen/slices.go
@@ -47,6 +47,7 @@ func SliceExtensionConst(s []int) []int {
 	// amd64:-`.*runtime\.makeslice`
 	// amd64:-`.*runtime\.panicmakeslicelen`
 	// amd64:"MOVUPS\tX15"
+	// loong64:-`.*runtime\.memclrNoHeapPointers`
 	// ppc64x:-`.*runtime\.memclrNoHeapPointers`
 	// ppc64x:-`.*runtime\.makeslice`
 	// ppc64x:-`.*runtime\.panicmakeslicelen`
@@ -58,6 +59,7 @@ func SliceExtensionConstInt64(s []int) []int {
 	// amd64:-`.*runtime\.makeslice`
 	// amd64:-`.*runtime\.panicmakeslicelen`
 	// amd64:"MOVUPS\tX15"
+	// loong64:-`.*runtime\.memclrNoHeapPointers`
 	// ppc64x:-`.*runtime\.memclrNoHeapPointers`
 	// ppc64x:-`.*runtime\.makeslice`
 	// ppc64x:-`.*runtime\.panicmakeslicelen`
@@ -69,6 +71,7 @@ func SliceExtensionConstUint64(s []int) []int {
 	// amd64:-`.*runtime\.makeslice`
 	// amd64:-`.*runtime\.panicmakeslicelen`
 	// amd64:"MOVUPS\tX15"
+	// loong64:-`.*runtime\.memclrNoHeapPointers`
 	// ppc64x:-`.*runtime\.memclrNoHeapPointers`
 	// ppc64x:-`.*runtime\.makeslice`
 	// ppc64x:-`.*runtime\.panicmakeslicelen`
@@ -80,16 +83,18 @@ func SliceExtensionConstUint(s []int) []int {
 	// amd64:-`.*runtime\.makeslice`
 	// amd64:-`.*runtime\.panicmakeslicelen`
 	// amd64:"MOVUPS\tX15"
+	// loong64:-`.*runtime\.memclrNoHeapPointers`
 	// ppc64x:-`.*runtime\.memclrNoHeapPointers`
 	// ppc64x:-`.*runtime\.makeslice`
 	// ppc64x:-`.*runtime\.panicmakeslicelen`
 	return append(s, make([]int, uint(1<<2))...)
 }
 
-// On ppc64x continue to use memclrNoHeapPointers
+// On ppc64x and loong64 continue to use memclrNoHeapPointers
 // for sizes >= 512.
 func SliceExtensionConst512(s []int) []int {
 	// amd64:-`.*runtime\.memclrNoHeapPointers`
+	// loong64:`.*runtime\.memclrNoHeapPointers`
 	// ppc64x:`.*runtime\.memclrNoHeapPointers`
 	return append(s, make([]int, 1<<9)...)
 }