Comments (4)
marcinkoss
commented on August 14, 2024
I did more digging and it seems that amoadd.d.aqrl does not work correctly.
This is execution log from LLVM compiled binary in REV (see code below):
111c0 >> <main>: 13 01 01 f8 addi sp, sp, -128
111c4 >> <main>: 23 3c 11 06 sd ra, 120(sp)
111c8 >> <main>: 23 38 81 06 sd s0, 112(sp)
111cc >> <main>: 13 04 01 08 addi s0, sp, 128
111d0 >> <main>: 13 05 00 00 li a0, 0
111d4 >> <main>: 23 26 a4 fe sw a0, -20(s0)
111d8 >> <main>: 13 06 20 00 li a2, 2
111dc >> <main>: 23 30 c4 fe sd a2, -32(s0)
111e0 >> <main>: 83 36 04 fe ld a3, -32(s0)
111e4 >> <main>: b7 35 01 00 lui a1, 19
111e8 >> <main>: 93 85 85 f5 addi a1, a1, -168
111ec >> <main>: af b6 d5 00 amoadd.d a3, a3, (a1)
111f0 >> <main>: 23 3c d4 fc sd a3, -40(s0)
111f4 >> <main>: 23 38 c4 fc sd a2, -48(s0)
111f8 >> <main>: 83 36 04 fd ld a3, -48(s0)
111fc >> <main>: af b6 d5 04 amoadd.d.aq a3, a3, (a1)
11200 >> <main>: 23 34 d4 fc sd a3, -56(s0)
11204 >> <main>: 23 30 c4 fc sd a2, -64(s0)
11208 >> <main>: 83 36 04 fc ld a3, -64(s0)
1120c >> <main>: af b6 d5 04 amoadd.d.aq a3, a3, (a1)
11210 >> <main>: 23 3c d4 fa sd a3, -72(s0)
11214 >> <main>: 23 38 c4 fa sd a2, -80(s0)
11218 >> <main>: 83 36 04 fb ld a3, -80(s0)
1121c >> <main>: af b6 d5 02 amoadd.d.rl a3, a3, (a1)
11220 >> <main>: 23 34 d4 fa sd a3, -88(s0)
11224 >> <main>: 23 30 c4 fa sd a2, -96(s0)
11228 >> <main>: 83 36 04 fa ld a3, -96(s0)
1122c >> <main>: af b6 d5 06 amoadd.d.aqrl a3, a3, (a1)
1122c >> <main>: af b6 d5 06 amoadd.d.aqrl a3, a3, (a1)
1122c >> <main>: af b6 d5 06 amoadd.d.aqrl a3, a3, (a1)
As you can see amoadd.d.aqrl enters some kind of loop (infinite).
Gnu gcc does not generate addmod.d.aqrl instruction for last two memory ordering parameters.
Program in C:
uint64_t atom64 = 0;
int main() {
__atomic_fetch_add(&atom64, 2, __ATOMIC_RELAXED);
__atomic_fetch_add(&atom64, 2, __ATOMIC_CONSUME);
__atomic_fetch_add(&atom64, 2, __ATOMIC_ACQUIRE);
__atomic_fetch_add(&atom64, 2, __ATOMIC_RELEASE);
__atomic_fetch_add(&atom64, 2, __ATOMIC_ACQ_REL);
__atomic_fetch_add(&atom64, 2, __ATOMIC_SEQ_CST);
return 0;
}
objdump from binary compiled with gnu gcc:
0000000000000628 <main>:
628: ff010113 addi sp,sp,-16
62c: 00813423 sd s0,8(sp)
630: 01010413 addi s0,sp,16
634: 00002797 auipc a5,0x2
638: a1c78793 addi a5,a5,-1508 # 2050 <atom16>
63c: 00200713 li a4,2
640: 00e7b02f amoadd.d zero,a4,(a5)
644: 00002797 auipc a5,0x2
648: a0c78793 addi a5,a5,-1524 # 2050 <atom16>
64c: 00200713 li a4,2
650: 04e7b02f amoadd.d.aq zero,a4,(a5)
654: 00002797 auipc a5,0x2
658: 9fc78793 addi a5,a5,-1540 # 2050 <atom16>
65c: 00200713 li a4,2
660: 04e7b02f amoadd.d.aq zero,a4,(a5)
664: 00002797 auipc a5,0x2
668: 9ec78793 addi a5,a5,-1556 # 2050 <atom16>
66c: 00200713 li a4,2
670: 0f50000f fence iorw,ow
674: 00e7b02f amoadd.d zero,a4,(a5)
678: 00002797 auipc a5,0x2
67c: 9d878793 addi a5,a5,-1576 # 2050 <atom16>
680: 00200713 li a4,2
684: 0f50000f fence iorw,ow
688: 04e7b02f amoadd.d.aq zero,a4,(a5)
68c: 00002797 auipc a5,0x2
690: 9c478793 addi a5,a5,-1596 # 2050 <atom16>
694: 00200713 li a4,2
698: 0f50000f fence iorw,ow
69c: 04e7b02f amoadd.d.aq zero,a4,(a5)
6a0: 00000793 li a5,0
6a4: 00078513 mv a0,a5
6a8: 00813403 ld s0,8(sp)
6ac: 01010113 addi sp,sp,16
6b0: 00008067 ret
and objdump of binary compiled with llvm:
00000000000111c0 <main>:
111c0: 13 01 01 f8 addi sp, sp, -128
111c4: 23 3c 11 06 sd ra, 120(sp)
111c8: 23 38 81 06 sd s0, 112(sp)
111cc: 13 04 01 08 addi s0, sp, 128
111d0: 13 05 00 00 li a0, 0
111d4: 23 26 a4 fe sw a0, -20(s0)
111d8: 13 06 20 00 li a2, 2
111dc: 23 30 c4 fe sd a2, -32(s0)
111e0: 83 36 04 fe ld a3, -32(s0)
111e4: b7 35 01 00 lui a1, 19
111e8: 93 85 85 f5 addi a1, a1, -168
111ec: af b6 d5 00 amoadd.d a3, a3, (a1)
111f0: 23 3c d4 fc sd a3, -40(s0)
111f4: 23 38 c4 fc sd a2, -48(s0)
111f8: 83 36 04 fd ld a3, -48(s0)
111fc: af b6 d5 04 amoadd.d.aq a3, a3, (a1)
11200: 23 34 d4 fc sd a3, -56(s0)
11204: 23 30 c4 fc sd a2, -64(s0)
11208: 83 36 04 fc ld a3, -64(s0)
1120c: af b6 d5 04 amoadd.d.aq a3, a3, (a1)
11210: 23 3c d4 fa sd a3, -72(s0)
11214: 23 38 c4 fa sd a2, -80(s0)
11218: 83 36 04 fb ld a3, -80(s0)
1121c: af b6 d5 02 amoadd.d.rl a3, a3, (a1)
11220: 23 34 d4 fa sd a3, -88(s0)
11224: 23 30 c4 fa sd a2, -96(s0)
11228: 83 36 04 fa ld a3, -96(s0)
1122c: af b6 d5 06 amoadd.d.aqrl a3, a3, (a1)
11230: 23 3c d4 f8 sd a3, -104(s0)
11234: 23 38 c4 f8 sd a2, -112(s0)
11238: 03 36 04 f9 ld a2, -112(s0)
1123c: af b5 c5 06 amoadd.d.aqrl a1, a2, (a1)
11240: 23 34 b4 f8 sd a1, -120(s0)
11244: 83 30 81 07 ld ra, 120(sp)
11248: 03 34 01 07 ld s0, 112(sp)
1124c: 13 01 01 08 addi sp, sp, 128
11250: 67 80 00 00 ret
from rev.
ukasz
commented on August 14, 2024
Even if atomic operations cannot be "atomic" yet, we would like to see their behavior emulated, so they are functionally working on a single thread.
from rev.
jleidel
commented on August 14, 2024
the single threaded versions are functionally emulated using only the embedded memory infrastructure; currently adding support for these on the amo branch
from rev.
jleidel
commented on August 14, 2024
fixed
from rev.
Related Issues (20)
- munmap test fails with memH enabled HOT 3
- Double addition ends as NaN HOT 7
- simple constructor crash HOT 4
- Argument passing / atoi conversion sensitivity to compiler output HOT 9
- missing thread execution for RV32I HOT 1
- Make spikeCheck modular based on the startSymbol passed in the rev config
- Statistics not being correctly accumulated or reported HOT 2
- fcvt.w.s does not support rounding modes HOT 8
- [FEATURE REQUEST] ifence HOT 1
- [FEATURE REQUEST] Support cache flush instructions HOT 1
- A problem with FPU register dependency tracking HOT 12
- Out of Memory error when using rev_mmap (with the memH memory backend) HOT 2
- c.addi16sp instruction is incorrectly printed as c.lui by Rev Tracer HOT 1
- Tracer is printing compressed register move instructions as c.jr instructions HOT 3
- [FEATURE REQUEST] Documentation HOT 3
- rev_read operation slowdown after large number of lines read HOT 2
- Stack buffer initialization corrupts memory. HOT 22
- lwsp instruction execution generating bad address outside of stack HOT 3
- dump_stack() system call returning only zeros
- [FEATURE REQUEST] Different testing levels HOT 4
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from rev.