Message ID | 20231030190354.94671-1-richard.henderson@linaro.org |
---|---|
State | New |
Headers | show |
Series | target/arm: Fix SVE STR increment | expand |
On Mon, 30 Oct 2023 at 19:04, Richard Henderson <richard.henderson@linaro.org> wrote: > > The previous change missed updating one of the increments. > > Cc: qemu-stable@nongnu.org > Fixes: e6dd5e782be ("target/arm: Use tcg_gen_qemu_{ld, st}_i128 in gen_sve_{ld, st}r") > Signed-off-by: Richard Henderson <richard.henderson@linaro.org> > --- > target/arm/tcg/translate-sve.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c > index 7b39962f20..fc4220114d 100644 > --- a/target/arm/tcg/translate-sve.c > +++ b/target/arm/tcg/translate-sve.c > @@ -4294,7 +4294,7 @@ void gen_sve_str(DisasContext *s, TCGv_ptr base, int vofs, > t0 = tcg_temp_new_i64(); > t1 = tcg_temp_new_i64(); > t16 = tcg_temp_new_i128(); > - for (i = 0; i < len_align; i += 8) { > + for (i = 0; i < len_align; i += 16) { > tcg_gen_ld_i64(t0, base, vofs + i); > tcg_gen_ld_i64(t1, base, vofs + i + 8); > tcg_gen_concat_i64_i128(t16, t0, t1); > -- This fixes the test case I have for vector lengths of 64 bytes and below, but vector lengths above that (eg 128 bytes) still misbehave in a different way (only every other set of 8 bytes gets stored). Presumably that's a different bug somewhere else? thanks -- PMM
On 10/31/23 03:52, Peter Maydell wrote: > On Mon, 30 Oct 2023 at 19:04, Richard Henderson > <richard.henderson@linaro.org> wrote: >> >> The previous change missed updating one of the increments. >> >> Cc: qemu-stable@nongnu.org >> Fixes: e6dd5e782be ("target/arm: Use tcg_gen_qemu_{ld, st}_i128 in gen_sve_{ld, st}r") >> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> >> --- >> target/arm/tcg/translate-sve.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c >> index 7b39962f20..fc4220114d 100644 >> --- a/target/arm/tcg/translate-sve.c >> +++ b/target/arm/tcg/translate-sve.c >> @@ -4294,7 +4294,7 @@ void gen_sve_str(DisasContext *s, TCGv_ptr base, int vofs, >> t0 = tcg_temp_new_i64(); >> t1 = tcg_temp_new_i64(); >> t16 = tcg_temp_new_i128(); >> - for (i = 0; i < len_align; i += 8) { >> + for (i = 0; i < len_align; i += 16) { >> tcg_gen_ld_i64(t0, base, vofs + i); >> tcg_gen_ld_i64(t1, base, vofs + i + 8); >> tcg_gen_concat_i64_i128(t16, t0, t1); >> -- > > This fixes the test case I have for vector lengths of 64 bytes > and below, but vector lengths above that (eg 128 bytes) still > misbehave in a different way (only every other set of 8 bytes > gets stored). Presumably that's a different bug somewhere else? Oh yes -- enable-debug yields qemu-aarch64: ../src/tcg/tcg-op-ldst.c:712: tcg_gen_qemu_st_i128_chk: Assertion `(memop & MO_SIZE) == MO_128' failed. r~
diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index 7b39962f20..fc4220114d 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -4294,7 +4294,7 @@ void gen_sve_str(DisasContext *s, TCGv_ptr base, int vofs, t0 = tcg_temp_new_i64(); t1 = tcg_temp_new_i64(); t16 = tcg_temp_new_i128(); - for (i = 0; i < len_align; i += 8) { + for (i = 0; i < len_align; i += 16) { tcg_gen_ld_i64(t0, base, vofs + i); tcg_gen_ld_i64(t1, base, vofs + i + 8); tcg_gen_concat_i64_i128(t16, t0, t1);
The previous change missed updating one of the increments. Cc: qemu-stable@nongnu.org Fixes: e6dd5e782be ("target/arm: Use tcg_gen_qemu_{ld, st}_i128 in gen_sve_{ld, st}r") Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/tcg/translate-sve.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)