RE: [PATCH 3/9] target/Hexagon: Finish conversion to tcg_gen_qemu_{ld,st}_*

[Taylor Simpson]

> -----Original Message-----
> From: Richard Henderson <richard.henderson@linaro.org>
> Sent: Tuesday, May 2, 2023 10:49 AM
> To: Taylor Simpson <tsimpson@quicinc.com>; qemu-devel@nongnu.org
> Subject: Re: [PATCH 3/9] target/Hexagon: Finish conversion to
> tcg_gen_qemu_{ld,st}_*
> 
> On 5/2/23 16:27, Taylor Simpson wrote:
> >
> >
> >> -----Original Message-----
> >> From: Richard Henderson <richard.henderson@linaro.org>
> >> Sent: Tuesday, May 2, 2023 8:58 AM
> >> To: qemu-devel@nongnu.org
> >> Cc: mrolnik@gmail.com; edgar.iglesias@gmail.com; Taylor Simpson
> >> <tsimpson@quicinc.com>; ale@rev.ng; anjo@rev.ng; laurent@vivier.eu;
> >> philmd@linaro.org; jiaxun.yang@flygoat.com; david@redhat.com;
> >> iii@linux.ibm.com; thuth@redhat.com; mark.cave-ayland@ilande.co.uk;
> >> atar4qemu@gmail.com; jcmvbkbc@gmail.com
> >> Subject: [PATCH 3/9] target/Hexagon: Finish conversion to
> >> tcg_gen_qemu_{ld,st}_*
> >>
> >> Convert away from the old interface with the implicit MemOp argument.
> >> Importantly, this removes some incorrect casts generated by
> >> idef-parser's gen_load().
> >>
> >> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> >> ---
> >>   target/hexagon/macros.h                     | 14 ++++-----
> >>   target/hexagon/genptr.c                     |  8 +++---
> >>   target/hexagon/idef-parser/parser-helpers.c | 28 +++++++++---------
> >>   target/hexagon/translate.c                  | 32 ++++++++++-----------
> >>   4 files changed, 40 insertions(+), 42 deletions(-)
> >>
> >> diff --git a/target/hexagon/genptr.c b/target/hexagon/genptr.c index
> >> 502c85ae35..244063b1d2 100644
> >> --- a/target/hexagon/genptr.c
> >> +++ b/target/hexagon/genptr.c
> >> @@ -320,14 +320,14 @@ void gen_set_byte_i64(int N, TCGv_i64 result,
> >> TCGv
> >> src)
> >>
> >>   static void gen_return(DisasContext *ctx, TCGv_i64 dst, TCGv src)
> >> @@ -
> >> 1019,7 +1019,7 @@ static void gen_vreg_load(DisasContext *ctx,
> >> intptr_t dstoff, TCGv src,
> >>           tcg_gen_andi_tl(src, src, ~((int32_t)sizeof(MMVector) - 1));
> >>       }
> >>       for (int i = 0; i < sizeof(MMVector) / 8; i++) {
> >> -        tcg_gen_qemu_ld64(tmp, src, ctx->mem_idx);
> >> +        tcg_gen_qemu_ld_i64(tmp, src, ctx->mem_idx, MO_TEUQ);
> >>           tcg_gen_addi_tl(src, src, 8);
> >>           tcg_gen_st_i64(tmp, cpu_env, dstoff + i * 8);
> >
> > Did you intend to leave the tcg_gen_st_i64 alone or should that be
> converted to tcg_gen_qemu_st64.
> >
> > There's a tcg_gen_st8_i64 in vec_to_qvec.  Does that need to be
> converted?
> 
> No, those are host stores to env, not guest stores to guest memory.
> Notice the lack of "qemu" in the function names.
> 
> > I'm curious if there's a better way to do a vector load (e.g., with
> tcg_gen_gvec_<something>) than a loop that does 8 bytes at a time.
> 
> The best you can do at the moment is tcg_gen_qemu_ld_i128.
> But there's no gvec variant to load arbitrary vector lengths.

OK, thanks.

Also,
Tested-by: Taylor Simpson <tsimpson@quicinc.com>