From 78e726b34f20c80e41285c646a33107fb6caf080 Mon Sep 17 00:00:00 2001
From: sergei <gorelkin@nanoreflex.ru>
Date: Sat, 18 Jan 2014 19:28:00 +0000
Subject: [PATCH] * i386: Somewhat optimized fpc_mul_qword routine, got rid
 from variable 'r', registers esi and edi. Also ignore overflow checking when
 both operands have their high dwords equal to zero, because in such case
 multiplication cannot overflow.

git-svn-id: trunk@26511 -
---
 rtl/i386/int64p.inc | 56 +++++++++++++++------------------------------
 1 file changed, 18 insertions(+), 38 deletions(-)

diff --git a/rtl/i386/int64p.inc b/rtl/i386/int64p.inc
index 37c3c7e278..37fbf6c920 100644
--- a/rtl/i386/int64p.inc
+++ b/rtl/i386/int64p.inc
@@ -170,7 +170,6 @@
     }
     function fpc_mul_qword(f1,f2 : qword;checkoverflow : longbool) : qword;[public,alias: 'FPC_MUL_QWORD']; compilerproc;
       var
-        r : qword;
         overflowed : boolean;
       begin
         overflowed:=false;
@@ -179,77 +178,58 @@
         asm
            movl f1+4,%edx
            movl f2+4,%ecx
-           cmpl $0,checkoverflow
-           jnz .Loverflowchecked
-
            orl %ecx,%edx
            movl f2,%edx
            movl f1,%eax
            jnz .Lqwordmultwomul
+           { if both upper dwords are =0 then it cannot overflow }
            mull %edx
            jmp .Lqwordmulready
         .Lqwordmultwomul:
+           cmpl $0,checkoverflow
+           jnz  .Loverflowchecked
            imul f1+4,%edx
            imul %eax,%ecx
            addl %edx,%ecx
            mull f2
            add %ecx,%edx
         .Lqwordmulready:
-           movl %eax,r
-           movl %edx,r+4
+           movl %eax,__RESULT
+           movl %edx,__RESULT+4
            jmp .Lend
 
         .Loverflowchecked:
            { if both upper dwords are <>0 then it overflows always }
            or %ecx,%ecx
            jz .Loverok1
-           or %edx,%edx
+           cmpl $0,f1+4
            jnz .Loverflowed
         .Loverok1:
            { overflow checked code }
-           orl %ecx,%edx
-           movl f2,%edi
-           movl f1,%esi
-           jnz .Lqwordmultwomul2
-           movl %edi,%eax
-           mull %esi
-           movl %eax,%esi
-           movl %edx,%edi
-           jmp .Lqwordmulready2
-
-        .Lqwordmultwomul2:
            movl f1+4,%eax
-           mull %edi
-           movl %eax,%edi
-           jc  .Loverflowed
-
-           movl %esi,%eax
-           mull %ecx
+           mull f2
            movl %eax,%ecx
            jc  .Loverflowed
 
-           addl %edi,%ecx
+           movl f1,%eax
+           mull f2+4
+           jc  .Loverflowed
+
+           addl %eax,%ecx
            jc  .Loverflowed
 
            movl f2,%eax
-           mull %esi
-           movl %eax,%esi
-           movl %edx,%edi
-
-           addl %ecx,%edi
-           jc  .Loverflowed
-
-        .Lqwordmulready2:
-           movl %esi,r
-           movl %edi,r+4
-           jmp .Lend
+           mull f1
+           addl %ecx,%edx
+           movl %eax,__RESULT
+           movl %edx,__RESULT+4
+           jnc  .Lend
 
         .Loverflowed:
            movb $1,overflowed
 
         .Lend:
-        end [ 'eax','edx','ecx','edi','esi' ];
-        fpc_mul_qword:=r;
+        end [ 'eax','edx','ecx'];
 
         if overflowed then
           HandleErrorFrame(215,get_frame);