runtime: VZEROUPPER at the end of FilterNilAVX512

VZEROUPPER at the end of FilterNilAVX512 as we're leaving AVX
context. Also explicitly zero Z15, as the high bits are not
guaranteed zero as of CL 728240.

Change-Id: I000a199206d0b8fd4905c8a699e934551bbd3c60
Reviewed-on: https://go-review.googlesource.com/c/go/+/728740
Reviewed-by: Michael Knyszek <mknyszek@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

diff --git a/src/internal/runtime/gc/scan/filter_amd64.s b/src/internal/runtime/gc/scan/filter_amd64.s
index 47330b62eeaa02bb1c0dd8fbed85935d6defdee8..2897d3c026f79325df16a59787e1f7709d158553 100644 (file)
--- a/src/internal/runtime/gc/scan/filter_amd64.s
+++ b/src/internal/runtime/gc/scan/filter_amd64.s
@@ -16,6 +16,7 @@ TEXT ·FilterNilAVX512(SB), NOSPLIT, $0-20
        SUBL R10, R12   // R12 = n - scanned
        CMPL R12, $8    // Compare (n - scanned) with 8
        JLT scalar_loop // If (n - scanned) < 8, jump to the scalar cleanup
+       VPXOR X15, X15, X15     // Zero the high bits of Z15
 
 vector_loop:
        LEAQ (R8)(R10*8), R13   // R13 = buf[scanned:] address
@@ -61,4 +62,5 @@ scalar_increment_i:
 
 end:
        MOVL R11, ret+16(FP)
+       VZEROUPPER
        RET