From e137b3a9b71a2711c1f68c8a8b9c0a7407fbcc4b Mon Sep 17 00:00:00 2001 From: Lepton Wu Date: Mon, 21 Oct 2019 20:22:18 -0700 Subject: [PATCH 1/1] mapi: Inline call x86_current_tls. This saves one return and a simple benchmark which calls glGetString repeatedly on my desktop shows it improves calls per second from 118M to 128M. Signed-off-by: Lepton Wu Reviewed-by: Kristian H. Kristensen --- src/mapi/entry_x86_tls.h | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) diff --git a/src/mapi/entry_x86_tls.h b/src/mapi/entry_x86_tls.h index 545b5a3c786..4b3d6bd02ad 100644 --- a/src/mapi/entry_x86_tls.h +++ b/src/mapi/entry_x86_tls.h @@ -56,9 +56,13 @@ __asm__(".balign 16\n" ".balign 16\n" \ func ":" -#define STUB_ASM_CODE(slot) \ - "call x86_current_tls\n\t" \ - "movl %gs:(%eax), %eax\n\t" \ +#define STUB_ASM_CODE(slot) \ + "call 1f\n" \ + "1:\n\t" \ + "popl %eax\n\t" \ + "addl $_GLOBAL_OFFSET_TABLE_+[.-1b], %eax\n\t" \ + "movl " ENTRY_CURRENT_TABLE "@GOTNTPOFF(%eax), %eax\n\t" \ + "movl %gs:(%eax), %eax\n\t" \ "jmp *(4 * " slot ")(%eax)" #define MAPI_TMP_STUB_ASM_GCC -- 2.30.2