Trying to go at this the other way (looking at the C code calling an external
routine) for confirmation. Here's my source code:
#include <stdio.h>
extern "OS" long long FOO();
void MyFunc()
{
long long bar;
bar = FOO();
printf("lld\n", bar);
}
I added the printf to make certain the store into bar did not get optimized
out. Here's the relevant object code (OPT(2),ARCH(8), unedited):
000011 | * bar = FOO();
00010E 5820 4000 000011 | L
r2,=Q(@STATIC)(,r4,0)
000112 A719 0000 000011 | LGHI r1,H'0'
000116 58E2 5000 000011 | L
r14,FOO_ptr(r2,r5,0)
00011A 98F0 E008 000011 | LM
r15,r0,&EPA_&WSA(r14,8)
00011E 5000 C1F4 000011 | ST
r0,_CEECAA_(,r12,500)
000122 0DEF 000011 | BASR r14,r15
000124 EBEF 0020 001C 000011 | RLLG r14,r15,32
000012 | *
000013 | * printf("lld\n", bar);
00012A E360 D0D8 0004 000013 | LG
r6,#SPILL0(,r13,216)
000130 C00A 0000 0000 000011 | NIHF r0,F'0'
000136 58F2 6004 000013 | L
r15,printf_ptr(r2,r6,4)
00013A C0EB 0000 0000 000011 | NILF r14,F'0'
000140 C020 0000 0068 000013 | LARL r2,F'104'
000146 4110 D0BC 000013 | LA
r1,#MX_TEMP1(,r13,188)
00014A B981 00E0 000011 | OGR r14,r0
00014E 50E0 D0C4 000013 | ST
r14,#MX_TEMP1(,r13,196)
000152 5020 D0BC 000013 | ST
r2,#MX_TEMP1(,r13,188)
000156 EB2E 0020 000A 000013 | SRAG r2,r14,32
00015C 98F0 F008 000013 | LM
r15,r0,&EPA_&WSA(r15,8)
000160 5000 C1F4 000013 | ST
r0,_CEECAA_(,r12,500)
000164 5020 D0C0 000013 | ST
r2,#MX_TEMP1(,r13,192)
000168 0DEF 000013 | BASR r14,r15
I see the C picking up low bits of R15 into the high bits of R14 (RLLG, NILF)
and the low bits of R0 into the low bits of R14 (NIHF, OGR).
Unless someone disagrees with my interpretation I think that confirms high bits
in R15, low bits in R0. Backwards from what I might have expected had someone
just said "R15 and R0" as R15 is the "normal" location for a return value so I
would associate that with the "long" part.
Some convoluted stuff there. The compiler goes to all the trouble of assembling
the 64 bit value in R14 and then ends up doing two separate stores into
#MX_TEMP1+192&196, one from R14 and another after having moved the high bits of
R14 into the low bits of R2 (SRAG).
Charles
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
