summaryrefslogtreecommitdiffstats
path: root/results/classifier/semantic-bugs/assembly/1612
diff options
context:
space:
mode:
authorChristian Krinitsin <mail@krinitsin.com>2025-07-03 19:39:53 +0200
committerChristian Krinitsin <mail@krinitsin.com>2025-07-03 19:39:53 +0200
commitdee4dcba78baf712cab403d47d9db319ab7f95d6 (patch)
tree418478faf06786701a56268672f73d6b0b4eb239 /results/classifier/semantic-bugs/assembly/1612
parent4d9e26c0333abd39bdbd039dcdb30ed429c475ba (diff)
downloademulator-bug-study-dee4dcba78baf712cab403d47d9db319ab7f95d6.tar.gz
emulator-bug-study-dee4dcba78baf712cab403d47d9db319ab7f95d6.zip
restructure results
Diffstat (limited to 'results/classifier/semantic-bugs/assembly/1612')
-rw-r--r--results/classifier/semantic-bugs/assembly/161264
1 files changed, 0 insertions, 64 deletions
diff --git a/results/classifier/semantic-bugs/assembly/1612 b/results/classifier/semantic-bugs/assembly/1612
deleted file mode 100644
index aa2f4e52..00000000
--- a/results/classifier/semantic-bugs/assembly/1612
+++ /dev/null
@@ -1,64 +0,0 @@
-assembly: 0.776
-instruction: 0.670
-graphic: 0.648
-semantic: 0.641
-device: 0.599
-other: 0.486
-network: 0.456
-socket: 0.453
-boot: 0.407
-vnc: 0.367
-mistranslation: 0.229
-KVM: 0.225
-
-SVE first-faulting gather loads return incorrect data
-Description of problem:
-The results of `ldff1(b|h|w|d)` seem to be incorrect when `<Zt> == <Zm>`. The first element is duplicated throughout the vector while the FFR indicates that all elements were successfully loaded. This happens since https://gitlab.com/qemu-project/qemu/-/commit/50de9b78cec06e6d16e92a114a505779359ca532, and still happens on the latest master.
-Steps to reproduce:
-1. This assembly sequence loads data with an `ldff1d` instruction (and also loads the ffr). Note that with `ldff1d`, `<Zt> == <Zm>`.
-
-asmtest.s
-```
- .type asmtest, @function
- .balign 16
- .global asmtest
-asmtest:
- setffr
- ptrue p0.d
- index z1.d, #0, #1
- ldff1d z1.d, p0/z, [x0, z1.d, LSL #3]
- rdffr p1.b
- st1d {z1.d}, p0, [x1]
- str p1, [x2]
- ret
-```
-
-This harness for convenience intialises some data and checks the element at index 1, which should be 1.
-
-test.c
-```
-#include <arm_sve.h>
-#include <stdio.h>
-
-void asmtest(int64_t const * data, svint64_t * loaded, svbool_t * ffr);
-
-int main() {
- const int64_t data[] = {42, 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};
- svint64_t loaded;
- svbool_t ffr;
-
- asmtest(data, &loaded, &ffr);
-
- // Check value of element at index 1
- svuint64_t lanes = svindex_u64(0, 1);
- svbool_t lane = svcmpeq_n_u64(svptrue_b64(), lanes, 1);
- printf("%ld\n", svaddv_s64(lane, loaded));
-}
-```
-
-2. ```clang-15 -fuse-ld=lld -march=armv8-a+sve2 -target aarch64-unknown-linux-gnu -static *.c *.s -o svldffgathertest```
-3. ```qemu-aarch64 svldffgathertest``` - the value printed should be 1, but it can be seen that all values in the loaded vector are 42.
-Additional information:
-The above code was successfully tested on real SVE hardware. Normal gathers work fine in QEMU, as does a non-gather first-fault load.