LTO in AmigaOS4 gcc

	Bottom Previous Topic Next Topic
Register To Post

(1) 2 »

Raziel

LTO in AmigaOS4 gcc

Posted on: 2019/2/9 14:21 #1

Home away from home

Are LTO's supported in AmigaOS4's GCC?

Taken from here
Quote:

Link-time optimization is a type of program optimization performed by a compiler to a program at link time. Link time optimization is relevant in programming languages that compile programs on a file-by-file basis, and then link those files together (such as C and Fortran), rather than all at once (such as Java's "Just in time" (JIT) compilation[citation needed]).

Once all files have been compiled separately into object files, traditionally, a compiler links (merges) the object files into a single file, the executable. However, in the case of the GCC compiler, for example, with Link Time Optimization (LTO) enabled, GCC is able to dump its internal representation (GIMPLE) to disk, so that all the different compilation units that will go to make up a single executable can be optimized as a single module. This expands the scope of inter-procedural optimizations to encompass the whole program (or, rather, everything that is visible at link time). With link-time optimization, the compiler can apply various forms of interprocedural optimization to the whole program, allowing for deeper analysis, more optimization, and ultimately better program performance.

It's often used in small systems like the original RPi but could probably help performance on AmigaOS4 too(?)

Sorry if this has been answered already

--

game box/art scans
scummvm infrequent builds

salass00

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/9 15:26 #2

Just can't stay away

@Raziel

VBCC has link time optimizations if you use the -O4 option when compiling and linking.

As I recall, in VBCC's case it basically postpones the actual compiling work to the link stage which means that this phase of the build process will be much slower. Not very nice if you are making small changes and recompiling a lot, but might work if you only use it for release builds only and use lower optimization otherwise.

Raziel

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/9 16:00 #3

Home away from home

@salass00

OK, but what about gcc 8.x which is used mainly (i think) on AmigaOS4 these days?

--

game box/art scans
scummvm infrequent builds

Elwood

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/9 17:25 #4

Just can't stay away

I never heard OS4 programs were compiled with GCC versions newer than v4

Philippe 'Elwood' FERRUCCI
Sam460ex 1.10 Ghz
http://elwoodb.free.fr

salass00

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/9 17:38 #5

Just can't stay away

@Elwood

I've compiled quite a few programs with gcc 8.2.0 already.

Quote:

$ ppc-amigaos-gcc -v
Using built-in specs.
COLLECT_GCC=ppc-amigaos-gcc
COLLECT_LTO_WRAPPER=/usr/local/amiga/adtools-gcc-8/libexec/gcc/ppc-amigaos/8.2.0/lto-wrapper
Target: ppc-amigaos
Configured with: /home/salass00/Development/Projects/DevTools/sba1-adtools-gcc-8/gcc/repo/configure --with-bugurl=https://github.com/sba1/adtools/issues --with-pkgversion='adtools build 8.2.0' --target=ppc-amigaos --prefix=/usr/local/amiga/adtools-gcc-8 --enable-languages=c,c++ --enable-haifa --enable-sjlj-exceptions --disable-libstdcxx-pch --disable-tls --enable-threads=amigaos
Thread model: amigaos
gcc version 8.2.0 (adtools build 8.2.0)

Even if you are compiling natively there is no need to stick to the now ancient gcc 4.2.4 as there are newer versions that you can download from Aminet:

http://aminet.net/package/dev/gcc/adtools-os4
http://aminet.net/package/dev/gcc/adtools-8-os4

salass00

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/9 17:45 #6

Just can't stay away

@Raziel

Sorry, I have no experience of LTO with gcc so I can't tell if it is working or not.

I mainly just use -O2 or -O3 optimization for my programs and don't use too many of the more fancy options.

Raziel

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/9 19:23 #7

Home away from home

@salass00

Hmm, ok, guess i'll have to try it out myself then.

Thank you

--

game box/art scans
scummvm infrequent builds

Raziel

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/11 10:26 #8

Home away from home

@thread

Just for the record LTO is NOT supported in gcc...time for a feature request

--

game box/art scans
scummvm infrequent builds

xenic

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/11 21:46 #9

Just can't stay away

@Raziel
Quote:

Just for the record LTO is NOT supported in gcc...time for a feature request

Are you sure. This would seem to indicate otherwise:

http://gcc.gnu.org/wiki/LinkTimeOptimization

Amiga X1000 with 2GB memory & OS 4.1FE + Radeon HD 5450

Raziel

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/11 21:56 #10

Home away from home

@xenic

Yes, but gcc for AmigaOS has not been build supporting LTO, so it's not in (yet)

See last comment by sba1

--

game box/art scans
scummvm infrequent builds

gregthecanuck

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/12 12:49 #11

Just popping in

@Raziel

bebbo recently got LTO support working on his gcc 6.x port for 68K. Maybe you can find some more info there?

http://eab.abime.net/showpost.php?p=1297430&postcount=1024

Raziel

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/12 15:13 #12

Home away from home

@gregthecanuck

Thank you, interesting thread, found even something that might be useful once gcc supports LTO

--

game box/art scans
scummvm infrequent builds

kas1e

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/16 17:13 #13

Home away from home

So i give it a go and build gcc with LTO enabled. All i need its to add to the /amiga/adtools/gcc-build/features.mk "lto" to enabled languages as well as add "--enabled-lto" option. As i read everywhere adding "lto" to enabled-languages there is not necessary, as once configure scripts will find "--enable-lto" then will add it for you, but then i anyway add it and it works. So, my current features.mk with c,c++,objc,objc++,fortran and lto looks like this:


MAJOR_VERSION:=$(word 1, $(subst ., , $(VERSION)))



FEATURES=/\

        --enable-languages=c,c++,objc,obj-c++,fortran,lto /\

        --enable-haifa /\

        --enable-lto /\

        --enable-sjlj-exceptions  /\

        --disable-libstdcxx-pch /\

        --disable-tls



# Check, if major version is greater than or equals to 8

ifeq ($(shell test $(MAJOR_VERSION) -ge 8; echo $$?), 0)

FEATURES+=--enable-threads=amigaos

endif

Then i clean my previous build, run "make -C native-build gcc-cross CROSS_PREFIX=/usr/local/amiga" and in end have new gcc:


$ ppc-amigaos-gcc -v

Using built-in specs.

COLLECT_GCC=ppc-amigaos-gcc

COLLECT_LTO_WRAPPER=/usr/local/amiga/libexec/gcc/ppc-amigaos/8.2.0/lto-wrapper.exe

Target: ppc-amigaos

Configured with: /amiga/adtools/gcc/repo/configure --with-bugurl=https://github.com/sba1/adtools/issues --with-pkgversion='adtools build 8.2.0' --target=ppc-amigaos --prefix=/usr/local/amiga --enable-languages=c,c++,objc,obj-c++,fortran,lto --enable-haifa --enable-lto --enable-sjlj-exceptions --disable-libstdcxx-pch --disable-tls --enable-threads=amigaos

Thread model: amigaos

gcc version 8.2.0 (adtools build 8.2.0)

Then, i trying to find some benchmarks, some testcases which will show that lto is works for sure, etc, but fail. Only find some texts with descriptions how it works, with some pseudo-asm and stuff, so to only understand that LTO is another optimisation, which can do some more low-level optimisation.

So all what i test at moment, is create such a simple test case:


#include <stdio.h>

#include <strings.h>



int *ptr;



void* my_malloc(size_t nbytes) 

{

          if (nbytes == 0)

                      return NULL;

            /* more code here */

            return ptr;

}



int main()

{



 my_malloc(100);

 my_malloc(100);

 my_malloc(100);

 my_malloc(100);



}

Then i build 2 versions of it:

With -O3 , but no LTO:

$ ppc-amigaos-gcc -O3 test_lto1.c -o test_lto1_no_lto

With -O3 , but with LTO:

$ ppc-amigaos-gcc -O3 -flto test_lto1.c -o test_lto1_with_lto

Now, size of the produced binary without LTO are 6176, size of binary produced with LTO enable are 6107. Difference are that version build with LTO are smaller on 69 bytes.

Then i disassmble it to asm code to see how it looks like, so, version WITHOUT lto, looks like this :


$ ppc-amigaos-objdump -d test_lto1_no_lto



test_lto1_no_lto:     file format elf32-amigaos





Disassembly of section .text:



01000074 <_start>:

 1000074:       94 21 ff c0     stwu    r1,-64(r1)

 1000078:       7c 08 02 a6     mflr    r0

 100007c:       3d 20 01 01     lis     r9,257

 1000080:       90 01 00 44     stw     r0,68(r1)

 1000084:       bf 21 00 24     stmw    r25,36(r1)

 1000088:       7c 79 1b 78     mr      r25,r3

 100008c:       7d bb 6b 78     mr      r27,r13

 1000090:       90 a9 10 30     stw     r5,4144(r9)

 1000094:       3d 20 01 02     lis     r9,258

 1000098:       7c bd 2b 78     mr      r29,r5

 100009c:       83 e5 02 78     lwz     r31,632(r5)

 10000a0:       39 a9 90 2c     addi    r13,r9,-28628

 10000a4:       7c 9a 23 78     mr      r26,r4

 10000a8:       80 1f 00 3c     lwz     r0,60(r31)

 10000ac:       7f e3 fb 78     mr      r3,r31

 10000b0:       7c 09 03 a6     mtctr   r0

 10000b4:       4e 80 04 21     bctrl

 10000b8:       81 5d 02 9e     lwz     r10,670(r29)

 10000bc:       3d 20 01 01     lis     r9,257

 10000c0:       3d 60 01 01     lis     r11,257

 10000c4:       3c 80 01 00     lis     r4,256

 10000c8:       93 eb 10 38     stw     r31,4152(r11)

 10000cc:       7f e3 fb 78     mr      r3,r31

 10000d0:       80 0a 00 10     lwz     r0,16(r10)

 10000d4:       38 84 10 00     addi    r4,r4,4096

 10000d8:       38 a0 00 34     li      r5,52

 10000dc:       90 09 10 44     stw     r0,4164(r9)

 10000e0:       3d 20 01 01     lis     r9,257

 10000e4:       80 1f 01 a8     lwz     r0,424(r31)

 10000e8:       91 49 10 48     stw     r10,4168(r9)

 10000ec:       7c 09 03 a6     mtctr   r0

 10000f0:       4e 80 04 21     bctrl

 10000f4:       7c 7c 1b 79     mr.     r28,r3

 10000f8:       40 82 00 4c     bne     1000144 <_start+0xd0>

 10000fc:       80 1f 02 3c     lwz     r0,572(r31)

 1000100:       3c 80 00 03     lis     r4,3

 1000104:       7f e3 fb 78     mr      r3,r31

 1000108:       60 84 80 0e     ori     r4,r4,32782

 100010c:       3b a0 00 14     li      r29,20

 1000110:       7c 09 03 a6     mtctr   r0

 1000114:       4e 80 04 21     bctrl

 1000118:       7f e3 fb 78     mr      r3,r31

 100011c:       83 ff 00 40     lwz     r31,64(r31)

 1000120:       7f e9 03 a6     mtctr   r31

 1000124:       4e 80 04 21     bctrl

 1000128:       80 01 00 44     lwz     r0,68(r1)

 100012c:       7f a3 eb 78     mr      r3,r29

 1000130:       7f 6d db 78     mr      r13,r27

 1000134:       bb 21 00 24     lmw     r25,36(r1)

 1000138:       38 21 00 40     addi    r1,r1,64

 100013c:       7c 08 03 a6     mtlr    r0

 1000140:       4e 80 00 20     blr

 1000144:       80 1f 01 c0     lwz     r0,448(r31)

 1000148:       3c a0 01 00     lis     r5,256

 100014c:       7f e3 fb 78     mr      r3,r31

 1000150:       38 a5 10 10     addi    r5,r5,4112

 1000154:       7f 84 e3 78     mr      r4,r28

 1000158:       7c 09 03 a6     mtctr   r0

 100015c:       38 c0 00 01     li      r6,1

 1000160:       38 e0 00 00     li      r7,0

 1000164:       4e 80 04 21     bctrl

 1000168:       7c 7e 1b 79     mr.     r30,r3

 100016c:       41 82 00 d0     beq     100023c <_start+0x1c8>

 1000170:       3d 20 01 01     lis     r9,257

 1000174:       38 00 00 01     li      r0,1

 1000178:       81 69 10 2c     lwz     r11,4140(r9)

 100017c:       3d 20 01 01     lis     r9,257

 1000180:       3d 40 01 01     lis     r10,257

 1000184:       39 29 10 1c     addi    r9,r9,4124

 1000188:       3d 00 01 01     lis     r8,257

 100018c:       80 6b 00 00     lwz     r3,0(r11)

 1000190:       3d 60 01 01     lis     r11,257

 1000194:       3c c0 01 01     lis     r6,257

 1000198:       39 6b 10 24     addi    r11,r11,4132

 100019c:       90 01 00 08     stw     r0,8(r1)

 10001a0:       3c e0 01 01     lis     r7,257

 10001a4:       91 61 00 10     stw     r11,16(r1)

 10001a8:       3d 60 01 01     lis     r11,257

 10001ac:       7f 24 cb 78     mr      r4,r25

 10001b0:       91 21 00 0c     stw     r9,12(r1)

 10001b4:       7f 45 d3 78     mr      r5,r26

 10001b8:       38 c6 10 34     addi    r6,r6,4148

 10001bc:       93 c8 10 40     stw     r30,4160(r8)

 10001c0:       3d 00 01 00     lis     r8,256

 10001c4:       38 e7 10 3c     addi    r7,r7,4156

 10001c8:       39 08 02 78     addi    r8,r8,632

 10001cc:       80 1e 00 50     lwz     r0,80(r30)

 10001d0:       81 2a 10 50     lwz     r9,4176(r10)

 10001d4:       81 4b 10 54     lwz     r10,4180(r11)

 10001d8:       7c 09 03 a6     mtctr   r0

 10001dc:       55 4a 07 fe     clrlwi  r10,r10,31

 10001e0:       4e 80 04 21     bctrl

 10001e4:       80 1f 01 c8     lwz     r0,456(r31)

 10001e8:       7f c4 f3 78     mr      r4,r30

 10001ec:       7c 7d 1b 78     mr      r29,r3

 10001f0:       7f e3 fb 78     mr      r3,r31

 10001f4:       7c 09 03 a6     mtctr   r0

 10001f8:       4e 80 04 21     bctrl

 10001fc:       80 1f 01 ac     lwz     r0,428(r31)

 1000200:       7f 84 e3 78     mr      r4,r28

 1000204:       7f e3 fb 78     mr      r3,r31

 1000208:       7c 09 03 a6     mtctr   r0

 100020c:       4e 80 04 21     bctrl

 1000210:       7f e3 fb 78     mr      r3,r31

 1000214:       83 ff 00 40     lwz     r31,64(r31)

 1000218:       7f e9 03 a6     mtctr   r31

 100021c:       4e 80 04 21     bctrl

 1000220:       80 01 00 44     lwz     r0,68(r1)

 1000224:       7f a3 eb 78     mr      r3,r29

 1000228:       7f 6d db 78     mr      r13,r27

 100022c:       bb 21 00 24     lmw     r25,36(r1)

 1000230:       38 21 00 40     addi    r1,r1,64

 1000234:       7c 08 03 a6     mtlr    r0

 1000238:       4e 80 00 20     blr

 100023c:       80 1f 02 3c     lwz     r0,572(r31)

 1000240:       3c 80 00 03     lis     r4,3

 1000244:       7f e3 fb 78     mr      r3,r31

 1000248:       60 84 80 0e     ori     r4,r4,32782

 100024c:       3b a0 00 14     li      r29,20

 1000250:       7c 09 03 a6     mtctr   r0

 1000254:       4e 80 04 21     bctrl

 1000258:       4b ff ff a4     b       10001fc <_start+0x188>



0100025c <my_malloc>:

 100025c:       2f 83 00 00     cmpwi   cr7,r3,0

 1000260:       41 9e 00 10     beq     cr7,1000270 <my_malloc+0x14>

 1000264:       3d 20 01 01     lis     r9,257

 1000268:       80 69 10 4c     lwz     r3,4172(r9)

 100026c:       4e 80 00 20     blr

 1000270:       38 60 00 00     li      r3,0

 1000274:       4e 80 00 20     blr



01000278 <main>:

 1000278:       38 60 00 00     li      r3,0

 100027c:       4e 80 00 20     blr

And version with lto enabled looks like this:


$ ppc-amigaos-objdump -d test_lto1_with_lto



test_lto1_with_lto:     file format elf32-amigaos





Disassembly of section .text:



01000074 <_start>:

 1000074:       94 21 ff c0     stwu    r1,-64(r1)

 1000078:       7c 08 02 a6     mflr    r0

 100007c:       3d 20 01 01     lis     r9,257

 1000080:       90 01 00 44     stw     r0,68(r1)

 1000084:       bf 21 00 24     stmw    r25,36(r1)

 1000088:       7c 79 1b 78     mr      r25,r3

 100008c:       7d bb 6b 78     mr      r27,r13

 1000090:       90 a9 10 30     stw     r5,4144(r9)

 1000094:       3d 20 01 02     lis     r9,258

 1000098:       7c bd 2b 78     mr      r29,r5

 100009c:       83 e5 02 78     lwz     r31,632(r5)

 10000a0:       39 a9 90 2c     addi    r13,r9,-28628

 10000a4:       7c 9a 23 78     mr      r26,r4

 10000a8:       80 1f 00 3c     lwz     r0,60(r31)

 10000ac:       7f e3 fb 78     mr      r3,r31

 10000b0:       7c 09 03 a6     mtctr   r0

 10000b4:       4e 80 04 21     bctrl

 10000b8:       81 5d 02 9e     lwz     r10,670(r29)

 10000bc:       3d 20 01 01     lis     r9,257

 10000c0:       3d 60 01 01     lis     r11,257

 10000c4:       3c 80 01 00     lis     r4,256

 10000c8:       93 eb 10 38     stw     r31,4152(r11)

 10000cc:       7f e3 fb 78     mr      r3,r31

 10000d0:       80 0a 00 10     lwz     r0,16(r10)

 10000d4:       38 84 10 00     addi    r4,r4,4096

 10000d8:       38 a0 00 34     li      r5,52

 10000dc:       90 09 10 44     stw     r0,4164(r9)

 10000e0:       3d 20 01 01     lis     r9,257

 10000e4:       80 1f 01 a8     lwz     r0,424(r31)

 10000e8:       91 49 10 48     stw     r10,4168(r9)

 10000ec:       7c 09 03 a6     mtctr   r0

 10000f0:       4e 80 04 21     bctrl

 10000f4:       7c 7c 1b 79     mr.     r28,r3

 10000f8:       40 82 00 4c     bne     1000144 <_start+0xd0>

 10000fc:       80 1f 02 3c     lwz     r0,572(r31)

 1000100:       3c 80 00 03     lis     r4,3

 1000104:       7f e3 fb 78     mr      r3,r31

 1000108:       60 84 80 0e     ori     r4,r4,32782

 100010c:       3b a0 00 14     li      r29,20

 1000110:       7c 09 03 a6     mtctr   r0

 1000114:       4e 80 04 21     bctrl

 1000118:       7f e3 fb 78     mr      r3,r31

 100011c:       83 ff 00 40     lwz     r31,64(r31)

 1000120:       7f e9 03 a6     mtctr   r31

 1000124:       4e 80 04 21     bctrl

 1000128:       80 01 00 44     lwz     r0,68(r1)

 100012c:       7f a3 eb 78     mr      r3,r29

 1000130:       7f 6d db 78     mr      r13,r27

 1000134:       bb 21 00 24     lmw     r25,36(r1)

 1000138:       38 21 00 40     addi    r1,r1,64

 100013c:       7c 08 03 a6     mtlr    r0

 1000140:       4e 80 00 20     blr

 1000144:       80 1f 01 c0     lwz     r0,448(r31)

 1000148:       3c a0 01 00     lis     r5,256

 100014c:       7f e3 fb 78     mr      r3,r31

 1000150:       38 a5 10 10     addi    r5,r5,4112

 1000154:       7f 84 e3 78     mr      r4,r28

 1000158:       7c 09 03 a6     mtctr   r0

 100015c:       38 c0 00 01     li      r6,1

 1000160:       38 e0 00 00     li      r7,0

 1000164:       4e 80 04 21     bctrl

 1000168:       7c 7e 1b 79     mr.     r30,r3

 100016c:       41 82 00 d0     beq     100023c <_start+0x1c8>

 1000170:       3d 20 01 01     lis     r9,257

 1000174:       38 00 00 01     li      r0,1

 1000178:       81 69 10 2c     lwz     r11,4140(r9)

 100017c:       3d 20 01 01     lis     r9,257

 1000180:       3d 40 01 01     lis     r10,257

 1000184:       39 29 10 1c     addi    r9,r9,4124

 1000188:       3d 00 01 01     lis     r8,257

 100018c:       80 6b 00 00     lwz     r3,0(r11)

 1000190:       3d 60 01 01     lis     r11,257

 1000194:       3c c0 01 01     lis     r6,257

 1000198:       39 6b 10 24     addi    r11,r11,4132

 100019c:       90 01 00 08     stw     r0,8(r1)

 10001a0:       3c e0 01 01     lis     r7,257

 10001a4:       91 61 00 10     stw     r11,16(r1)

 10001a8:       3d 60 01 01     lis     r11,257

 10001ac:       7f 24 cb 78     mr      r4,r25

 10001b0:       91 21 00 0c     stw     r9,12(r1)

 10001b4:       7f 45 d3 78     mr      r5,r26

 10001b8:       38 c6 10 34     addi    r6,r6,4148

 10001bc:       93 c8 10 40     stw     r30,4160(r8)

 10001c0:       3d 00 01 00     lis     r8,256

 10001c4:       38 e7 10 3c     addi    r7,r7,4156

 10001c8:       39 08 02 5c     addi    r8,r8,604

 10001cc:       80 1e 00 50     lwz     r0,80(r30)

 10001d0:       81 2a 10 4c     lwz     r9,4172(r10)

 10001d4:       81 4b 10 50     lwz     r10,4176(r11)

 10001d8:       7c 09 03 a6     mtctr   r0

 10001dc:       55 4a 07 fe     clrlwi  r10,r10,31

 10001e0:       4e 80 04 21     bctrl

 10001e4:       80 1f 01 c8     lwz     r0,456(r31)

 10001e8:       7f c4 f3 78     mr      r4,r30

 10001ec:       7c 7d 1b 78     mr      r29,r3

 10001f0:       7f e3 fb 78     mr      r3,r31

 10001f4:       7c 09 03 a6     mtctr   r0

 10001f8:       4e 80 04 21     bctrl

 10001fc:       80 1f 01 ac     lwz     r0,428(r31)

 1000200:       7f 84 e3 78     mr      r4,r28

 1000204:       7f e3 fb 78     mr      r3,r31

 1000208:       7c 09 03 a6     mtctr   r0

 100020c:       4e 80 04 21     bctrl

 1000210:       7f e3 fb 78     mr      r3,r31

 1000214:       83 ff 00 40     lwz     r31,64(r31)

 1000218:       7f e9 03 a6     mtctr   r31

 100021c:       4e 80 04 21     bctrl

 1000220:       80 01 00 44     lwz     r0,68(r1)

 1000224:       7f a3 eb 78     mr      r3,r29

 1000228:       7f 6d db 78     mr      r13,r27

 100022c:       bb 21 00 24     lmw     r25,36(r1)

 1000230:       38 21 00 40     addi    r1,r1,64

 1000234:       7c 08 03 a6     mtlr    r0

 1000238:       4e 80 00 20     blr

 100023c:       80 1f 02 3c     lwz     r0,572(r31)

 1000240:       3c 80 00 03     lis     r4,3

 1000244:       7f e3 fb 78     mr      r3,r31

 1000248:       60 84 80 0e     ori     r4,r4,32782

 100024c:       3b a0 00 14     li      r29,20

 1000250:       7c 09 03 a6     mtctr   r0

 1000254:       4e 80 04 21     bctrl

 1000258:       4b ff ff a4     b       10001fc <_start+0x188>



0100025c <main>:

 100025c:       38 60 00 00     li      r3,0

 1000260:       4e 80 00 20     blr

If you bored to compare, can say that start() about the same (some values only changes in some place). While in case with LTO i can't see my_malloc() function code at all. Is it lto optimisation somehow remove it, or whf, dunno. But what for sure, the same happens if i build the same test code from win32. I.e. same differences in disassembled code, and the same disappeared my_malloc(). Probabaly that can mean that -flto works the same for us too.

So now need to make some real lto test code, which will point out that it 100% works (or not).

Any idea of simple test case ?

Join us to improve dopus5!
AmigaOS4 on youtube

broadblues

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/16 17:52 #14

Home away from home

@kas1e

Your test case is stupidly simple (no insult intended) it does nothing, calling my_malloc() has no efffect (result is unused) so it's been 'inlined then removed' by the -03 optimisation, then I expect the LTO optimsation has noticed that my_malloc() ends up not being called and removed that too. Normal optimsiation wouldn't remove the function as there would be no way to know if it was called from another module till link time.

Quote:

Any idea of simple test case ?

LTO isn't really going to do too much until you have multiple object files which do something meaningful, that can be benchmarked.

If you split out your my_malloc() into another src file so that there are two objects file before linking you might see a bigger difference as the non LTO case would not be able to inline / remove it as it coudn't know if there were side effects to the function call. LTO (I would guess) would end up somewhat similar to the current result, depending on just what link time optimisations are possible.

Blender For OS4.x : Blues : Walker Broad

kas1e

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/16 18:34 #15

Home away from home

@Andy
Quote:

LTO isn't really going to do too much until you have multiple object files which do something meaningful, that can be benchmarked.

Yeah, will try tomorrow to build some fat by objects game with fps counter

Quote:

Your test case is stupidly simple (no insult intended) it does nothing, calling my_malloc() has no efffect (result is unused) so it's been 'inlined then removed' by the -03 optimisation, then I expect the LTO optimsation has noticed that my_malloc() ends up not being called and removed that too. Normal optimsiation wouldn't remove the function as there would be no way to know if it was called from another module till link time.

Sure test case suck, but that point out that -flto is smart enough to make some optimisation even on single object file, just to even have 100bytes less exe.

Quote:

If you split out your my_malloc() into another src file so that there are two objects file before linking you might see a bigger difference as the non LTO case would not be able to inline / remove it as it coudn't know if there were side effects to the function call. LTO (I would guess) would end up somewhat similar to the current result, depending on just what link time optimisations are possible.

Tried that:


$ cat my_malloc.c



#include <stdio.h>

#include <strings.h>



int *ptr;



void* my_malloc(size_t nbytes)

{

                  if (nbytes == 0)

                   return NULL;

              /* more code here */

              return ptr;

}





$ cat main.c



#include <stdio.h>

#include <strings.h>



#include <stdio.h>

#include <strings.h>



void* my_malloc(size_t nbytes);



int main()

{



         my_malloc(100);

         my_malloc(100);

         my_malloc(100);

         my_malloc(100);



}

And then:


ppc-amigaos-gcc -flto -O3 -c my_malloc.c -o my_malloc_with_lto.o

ppc-amigaos-gcc -flto -O3 -c main.c -o main_with_lto.o

ppc-amigaos-gcc -flto -O3 main_with_lto.o my_malloc_with_lto.o -o test_with_lto



ppc-amigaos-gcc -O3 -c my_malloc.c -o my_malloc_no_lto.o

ppc-amigaos-gcc -O3 -c main.c -o main_no_lto.o

ppc-amigaos-gcc -O3 main_no_lto.o my_malloc_no_lto.o -o test_no_lto

And now, binary where no lto used, start to be bigger than when it single object version: 6247 vs 6176, while lto version in both cases be it 1 object or two, are 6107. So in case with single object version it better on 69 bytes, and in case with 2 objects, its better on 140 bytes (seems like you expect).

Join us to improve dopus5!
AmigaOS4 on youtube

kas1e

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/18 17:52 #16

Home away from home

Soo.. Good news !

I rebuild foobillard++ with -O3 -flto for all objects and -O3 -flto for final linking line , and compare it with pure -O3 compile without lto. And results are for the same test (same size of game's window, same settings, just rerun few times 2 different binaries):

pure -O3 max: 64,1, avg: 60,19
-O3 -flto max: 72.7, avg: 68.57

What mean it works damn and give already quite a boost for first tested game ! +8fps , its about 12-13% speed increase !

Also, in the case with no lto, binary size are 16.318.671, and with lto are 15.863.290, so its not only faster by fps, but also less on 0.5mb.

Probabaly that huge boost can be explained by the fact, that everything compiled statically : SDL2, gl4es, glu, sdl2_image, sdl2_mixer, and all those 3d party libs, so -flto optimize them too on the linking stage by taking from those .a all the objects.

Will try to rebuild barony, frickingshark, prototype, neverball/neverput and quake3 , to see how they will behave with -flto.

Edited by kas1e on 2019/2/18 18:54:06

Join us to improve dopus5!
AmigaOS4 on youtube

thellier

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/19 12:49 #17

Not too shy to talk

Hello

Nice

Kas1e, is your lto gcc available somewhere ?

Many Thanks

kas1e

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/19 14:02 #18

Home away from home

@thellier
At moment it just on my cross-compiler cygwin's setup. But its really easy to compile it from adtools repo by adding that --enable-lto option

Join us to improve dopus5!
AmigaOS4 on youtube

thellier

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/20 8:17 #19

Not too shy to talk

OK

I have a 32 bit cygwin cross-compiling environement too: is your gcc 32 bits ?

Alain

kas1e

Re: LTO in AmigaOS4 gcc

Posted on: 2019/2/20 11:46 #20

Home away from home

@thellier
My one are x86_64-pc-cygwin, gcc 7.4.0

Join us to improve dopus5!
AmigaOS4 on youtube

Register To Post	(1) 2 »
	Top Previous Topic Next Topic

Currently Active Users Viewing This Thread: 1 ( 0 members and 1 Anonymous Users )