dayeol
commented
3 years ago This might be related to how we manage fp in the runtime. But to make sure, can you try to run them outside of the enclave and see if they're fine?
Open neiling opened 3 years ago
dayeol
commented
3 years ago This might be related to how we manage fp in the runtime. But to make sure, can you try to run them outside of the enclave and see if they're fine?
neiling
commented
3 years ago If I run it outside the enclave, all works fine and as expected.
dkohlbre
commented
3 years ago This is pretty odd. Best guess is that the FPU state isn't clean due to some issue with our context switch and the O1 assumes it is?
Can you post the assembly for your main for both the O0 and O1 binaries?
neiling
commented
3 years ago **************************************************************
* GCC -O1 *
**************************************************************
int __stdcall main(void)
assume gp = 0x17800
int a0:4 <RETURN>
main XREF[4]: Entry Point(*),
_start_c:0001016e(*),
00017148(*),
.debug_frame::00000040(*)
00010208 41 11 c.addi sp,-0x10
assume gp = <UNKNOWN>
0001020a 06 e4 c.sdsp ra,0x8(sp)
0001020c 81 47 c.li a5,0x0
0001020e 53 07 00 f2 fmv.d.x fa4,zero
00010212 17 47 00 00 auipc a4,0x4
00010216 07 35 e7 66 fld fa0,0x66e(a4=>DAT_00014880) = 3FE0000000000000h
0001021a 17 47 00 00 auipc a4,0x4
0001021e 87 35 e7 66 fld fa1,0x66e(a4=>DAT_00014888) = 3E45798EE2308C3Ah
00010222 17 47 00 00 auipc a4,0x4
00010226 07 36 e7 66 fld fa2,0x66e(a4=>DAT_00014890) = 3FF0000000000000h
0001022a 17 47 00 00 auipc a4,0x4
0001022e 87 36 e7 66 fld fa3,0x66e(a4=>DAT_00014898) = 4010000000000000h
00010232 37 e7 f5 05 lui a4,0x5f5e
00010236 13 07 07 10 addi a4,a4,0x100
LAB_0001023a XREF[1]: 00010258(j)
0001023a d3 f7 37 d2 fcvt.d.lu fa5,a5,dyn
0001023e d3 f7 a7 02 fadd.d fa5,fa5,fa0,dyn
00010242 d3 f7 b7 12 fmul.d fa5,fa5,fa1,dyn
00010246 d3 f7 f7 12 fmul.d fa5,fa5,fa5,dyn
0001024a d3 f7 c7 02 fadd.d fa5,fa5,fa2,dyn
0001024e d3 f7 f6 1a fdiv.d fa5,fa3,fa5,dyn
00010252 53 77 f7 02 fadd.d fa4,fa4,fa5,dyn
00010256 85 07 c.addi a5,0x1
00010258 e3 91 e7 fe bne a5,a4,LAB_0001023a
0001025c 97 47 00 00 auipc a5,0x4
00010260 87 b7 c7 62 fld fa5,0x62c(a5=>DAT_00014888) = 3E45798EE2308C3Ah
00010264 d3 77 f7 12 fmul.d fa5,fa4,fa5,dyn
00010268 d3 85 07 e2 fmv.x.d a1,fa5
0001026c 17 45 00 00 auipc a0,0x4
00010270 13 05 45 60 addi a0=>s_Pi=%.15lf_00014870,a0,0x604 = "Pi=%.15lf\n"
00010274 ef 00 60 21 jal ra,printf int printf(char * __format, ...)
00010278 01 45 c.li a0,0x0
0001027a a2 60 c.ldsp ra,0x8(sp)
0001027c 41 01 c.addi sp,0x10
0001027e 82 80 ret **************************************************************
* GCC -O0 *
**************************************************************
int __stdcall main(void)
assume gp = 0x17800
int a0:4 <RETURN>
double Stack[-0x10]:8 x
double Stack[-0x18]:8 pi XREF[2]: 0001025e(*),
00010262(*)
double Stack[-0x20]:8 delta_x XREF[3]: 0001029e(*),
000102c4(*),
000102c8(*)
uint64_t Stack[-0x28]:8 N XREF[4]: 00010232(*),
00010256(*),
00010296(*),
000102bc(*)
uint64_t Stack[-0x30]:8 i XREF[3]: 00010218(*),
0001021c(*),
000102b0(*)
double Stack[-0x38]:8 sum XREF[5]: 0001023a(*),
00010240(*),
000102a2(*),
000102a8(*),
000102ac(*)
undefined8 Stack[-0x40]:8 local_40 XREF[5]: 00010236(*),
00010286(*),
0001028e(*),
00010292(*),
000102b8(*)
main XREF[4]: Entry Point(*),
_start_c:0001016e(*),
00017148(*),
.debug_frame::00000040(*)
00010208 39 71 c.addi16sp sp,-0x40
assume gp = <UNKNOWN>
0001020a 06 fc c.sdsp ra,0x38(sp)
0001020c 22 f8 c.sdsp s0,0x30(sp)
0001020e 80 00 c.addi4spn s0,sp,0x40
00010210 b7 e7 f5 05 lui a5,0x5f5e
00010214 93 87 07 10 addi a5,a5,0x100
00010218 23 38 f4 fc sd a5,-0x30=>i(s0)
0001021c 83 37 04 fd ld a5,-0x30=>i(s0)
00010220 d3 f7 37 d2 fcvt.d.lu fa5,a5,dyn
00010224 97 47 00 00 auipc a5,0x4
00010228 93 87 c7 6b addi a5,a5,0x6bc
0001022c 98 23 c.fld fa4,0x0(a5=>DAT_000148e0) = 3FF0000000000000h
0001022e d3 77 f7 1a fdiv.d fa5,fa4,fa5,dyn
00010232 27 3c f4 fc fsd fa5,-0x28=>N(s0)
00010236 23 30 04 fc sd zero,-0x40=>local_40(s0)
0001023a 23 34 04 fc sd zero,-0x38=>sum(s0)
0001023e bd a0 c.j LAB_000102ac
LAB_00010240 XREF[1]: 000102b4(j)
00010240 83 37 84 fc ld a5,-0x38=>sum(s0)
00010244 53 f7 37 d2 fcvt.d.lu fa4,a5,dyn
00010248 97 47 00 00 auipc a5,0x4
0001024c 93 87 07 6a addi a5,a5,0x6a0
00010250 9c 23 c.fld fa5,0x0(a5=>DAT_000148e8) = 3FE0000000000000h
00010252 d3 77 f7 02 fadd.d fa5,fa4,fa5,dyn
00010256 07 37 84 fd fld fa4,-0x28=>N(s0)
0001025a d3 77 f7 12 fmul.d fa5,fa4,fa5,dyn
0001025e 27 34 f4 fe fsd fa5,-0x18=>pi(s0)
00010262 87 37 84 fe fld fa5,-0x18=>pi(s0)
00010266 53 f7 f7 12 fmul.d fa4,fa5,fa5,dyn
0001026a 97 47 00 00 auipc a5,0x4
0001026e 93 87 67 67 addi a5,a5,0x676
00010272 9c 23 c.fld fa5,0x0(a5=>DAT_000148e0) = 3FF0000000000000h
00010274 d3 77 f7 02 fadd.d fa5,fa4,fa5,dyn
00010278 97 47 00 00 auipc a5,0x4
0001027c 93 87 87 67 addi a5,a5,0x678
00010280 98 23 c.fld fa4,0x0(a5=>DAT_000148f0) = 4010000000000000h
00010282 d3 77 f7 1a fdiv.d fa5,fa4,fa5,dyn
00010286 07 37 04 fc fld fa4,-0x40=>local_40(s0)
0001028a d3 77 f7 02 fadd.d fa5,fa4,fa5,dyn
0001028e 27 30 f4 fc fsd fa5,-0x40=>local_40(s0)
00010292 07 37 04 fc fld fa4,-0x40=>local_40(s0)
00010296 87 37 84 fd fld fa5,-0x28=>N(s0)
0001029a d3 77 f7 12 fmul.d fa5,fa4,fa5,dyn
0001029e 27 30 f4 fe fsd fa5,-0x20=>delta_x(s0)
000102a2 83 37 84 fc ld a5,-0x38=>sum(s0)
000102a6 85 07 c.addi a5,0x1
000102a8 23 34 f4 fc sd a5,-0x38=>sum(s0)
LAB_000102ac XREF[1]: 0001023e(j)
000102ac 03 37 84 fc ld a4,-0x38=>sum(s0)
000102b0 83 37 04 fd ld a5,-0x30=>i(s0)
000102b4 e3 66 f7 f8 bltu a4,a5,LAB_00010240
000102b8 07 37 04 fc fld fa4,-0x40=>local_40(s0)
000102bc 87 37 84 fd fld fa5,-0x28=>N(s0)
000102c0 d3 77 f7 12 fmul.d fa5,fa4,fa5,dyn
000102c4 27 30 f4 fe fsd fa5,-0x20=>delta_x(s0)
000102c8 83 35 04 fe ld a1,-0x20=>delta_x(s0)
000102cc 17 45 00 00 auipc a0,0x4
000102d0 13 05 45 60 addi a0=>s_Pi=%.15lf_000148d0,a0,0x604 = "Pi=%.15lf\n"
000102d4 ef 00 a0 21 jal ra,printf int printf(char * __format, ...)
000102d8 81 47 c.li a5,0x0
000102da 3e 85 c.mv a0,a5
000102dc e2 70 c.ldsp ra,0x38(sp)
000102de 42 74 c.ldsp s0,0x30(sp)
000102e0 21 61 c.addi16sp sp,0x40
000102e2 82 80 ret
Describe the bug Keystone version: v1.0.0 riscv64-unknown-linux-musl-gcc version: 10.2.0
Different results of floating point multiplication when the code was compiled with higher optimization.
Screenshots or Error Log
Additional context
Hello everyone,
I observed a strange behavior related to floating point multiplication and higher compiler optimization.
The following code should approximate calculate pi. If I compile the code with -O0 all works fine, and I get every time the same result. If I compile it with e.g. -O1 the result is not a number, or it looks like a random value. I have not done any deeper debugging yet. Maybe that's a known issue and some can explain me what here happens. Thanks a lot!