summary refs log tree commit diff stats
path: root/results/classifier/zero-shot/105/mistranslation/1918026
blob: 59faa2ac6bc20e2fdad21692776da69ab8ead5a1 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
mistranslation: 0.963
semantic: 0.886
other: 0.830
graphic: 0.817
instruction: 0.811
device: 0.770
network: 0.646
vnc: 0.567
socket: 0.540
KVM: 0.539
boot: 0.473
assembly: 0.440

RISCV64 32-bit AMOs incorrectly simulated

Version: qemu-riscv64 version 4.2.1 (Debian 1:4.2-3ubuntu6.14)

test:
  amomaxu.w a0, a1, (a0)
  ret

int32_t* value = -7;
EXPECT_EQ(-7, test(&value, -11));
EXPECT_EQ(-7, value);  // FAIL, saw -11
EXPECT_EQ(-7, test(&value, -7));
EXPECT_EQ(-7, value);  // FAIL, raw -11
EXPECT_EQ(-7, test(&value, -4));
EXPECT_EQ(-4, value);

test:
  amomax.w a0, a1, (a0)
  ret

int32_t* value = -7;
EXPECT_EQ(-7, test(&value, -11));
EXPECT_EQ(-7, value);
EXPECT_EQ(-7, test(&value, -7));
EXPECT_EQ(-7, value);
EXPECT_EQ(-7, test(&value, -4));
EXPECT_EQ(-4, value);  // FAIL, saw -7

I suspect that trans_amo<op>_w should be using tcg_gen_atomic_fetch_<op>_i32 instead of tcg_gen_atomic_fetch_<op>_tl.

Not as simple as _tl vs _i32.  That suffix should be taking
care of the sign-extension to _tl that happens after the
operation.  The size of the operation should be in the MO_TESL,
which specifies target-endian signed "long" (32-bit).

Will investigate further.

Flushing out the report to something that compiles,
the test case works for me using qemu 5.2.