[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH 2/2] ppc: Add implementations for pre-increment loads/stores
|
From: |
Paulo César Pereira de Andrade |
|
Subject: |
Re: [PATCH 2/2] ppc: Add implementations for pre-increment loads/stores |
|
Date: |
Wed, 24 Jan 2024 16:08:22 -0300 |
Em dom., 7 de jan. de 2024 às 12:27, Paul Cercueil
<paul@crapouillou.net> escreveu:
>
> The PowerPC instruction set has instructions to load/store with
> pre-increments with a few exceptions.
>
> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
> ---
> lib/jit_ppc-cpu.c | 33 +++++++++++++++++++++++++++++++++
> lib/jit_ppc-fpu.c | 8 ++++++++
> lib/jit_ppc.c | 8 ++++----
> 3 files changed, 45 insertions(+), 4 deletions(-)
>
> diff --git a/lib/jit_ppc-cpu.c b/lib/jit_ppc-cpu.c
> index fc79944..1a221b6 100644
> --- a/lib/jit_ppc-cpu.c
> +++ b/lib/jit_ppc-cpu.c
> @@ -300,6 +300,7 @@ static void
> _FXS(jit_state_t*,int,int,int,int,int,int,int);
> # define LWZUX(d,a,b) FX(31,d,a,b,55)
> # define LWZX(d,a,b) FX(31,d,a,b,23)
> # define LD(d,a,s) FDs(58,d,a,s)
> +# define LDU(d,a,s) FDs(58,d,a,s|1)
> # define LDX(d,a,b) FX(31,d,a,b,21)
> # define MCRF(d,s) FXL(19,((d)<<2),((s)<<2),0)
> # if DEBUG
> @@ -890,6 +891,26 @@ static void
> _ldxr_l(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t);
> # define ldxi_l(r0,r1,i0) _ldxi_l(_jit,r0,r1,i0)
> static void _ldxi_l(jit_state_t*,jit_int32_t,jit_int32_t,jit_word_t);
> # endif
> +# define ldxbi_c(r0,r1,i0) generic_ldxbi_c(r0,r1,i0)
> +# define ldxbi_uc(r0,r1,i0) LBZU(r0,r1,i0)
> +# define ldxbi_s(r0,r1,i0) LHAU(r0,r1,i0)
> +# define ldxbi_us(r0,r1,i0) LHZU(r0,r1,i0)
> +# if __WORDSIZE == 32
> +# define ldxbi_i(r0,r1,i0) LWZU(r0,r1,i0)
> +# else
> +# define ldxbi_i(r0,r1,i0) generic_ldxbi_i(r0,r1,i0)
> +# define ldxbi_ui(r0,r1,i0) LWZU(r0,r1,i0)
> +# define ldxbi_l(r0,r1,i0) LDU(r0,r1,i0)
> +# endif
> +# define ldxai_c(r0,r1,i0) generic_ldxai_c(r0,r1,i0)
> +# define ldxai_uc(r0,r1,i0) generic_ldxai_uc(r0,r1,i0)
> +# define ldxai_s(r0,r1,i0) generic_ldxai_s(r0,r1,i0)
> +# define ldxai_us(r0,r1,i0) generic_ldxai_us(r0,r1,i0)
> +# define ldxai_i(r0,r1,i0) generic_ldxai_i(r0,r1,i0)
> +# if __WORDSIZE == 64
> +# define ldxai_ui(r0,r1,i0) generic_ldxai_ui(r0,r1,i0)
> +# define ldxai_l(r0,r1,i0) generic_ldxai_l(r0,r1,i0)
> +# endif
> # define str_c(r0,r1) STBX(r1, _R0_REGNO, r0)
> # define sti_c(i0,r0) _sti_c(_jit,i0,r0)
> static void _sti_c(jit_state_t*,jit_word_t,jit_int32_t);
> @@ -920,6 +941,18 @@ static void
> _stxr_l(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t);
> # define stxi_l(i0,r0,r1) _stxi_l(_jit,i0,r0,r1)
> static void _stxi_l(jit_state_t*,jit_word_t,jit_int32_t,jit_int32_t);
> # endif
> +# define stxbi_c(i0,r0,r1) STBU(r1,r0,i0)
> +# define stxbi_s(i0,r0,r1) STHU(r1,r0,i0)
> +# define stxbi_i(i0,r0,r1) STWU(r1,r0,i0)
> +# if __WORDSIZE == 64
> +# define stxbi_l(i0,r0,r1) STDU(r1,r0,i0)
> +# endif
> +# define stxai_c(i0,r0,r1) generic_stxai_c(i0,r0,r1)
> +# define stxai_s(i0,r0,r1) generic_stxai_s(i0,r0,r1)
> +# define stxai_i(i0,r0,r1) generic_stxai_i(i0,r0,r1)
> +# if __WORDSIZE == 64
> +# define stxai_l(i0,r0,r1) generic_stxai_l(i0,r0,r1)
> +# endif
> # define jmpr(r0) _jmpr(_jit,r0)
> static void _jmpr(jit_state_t*,jit_int32_t);
> # define jmpi(i0) _jmpi(_jit,i0)
> diff --git a/lib/jit_ppc-fpu.c b/lib/jit_ppc-fpu.c
> index 605bd4f..0eb24f2 100644
> --- a/lib/jit_ppc-fpu.c
> +++ b/lib/jit_ppc-fpu.c
> @@ -422,6 +422,14 @@ static void _sti_d(jit_state_t*,jit_word_t,jit_int32_t);
> static void _stxr_d(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t);
> # define stxi_d(i0,r0,r1) _stxi_d(_jit,i0,r0,r1)
> static void _stxi_d(jit_state_t*,jit_word_t,jit_int32_t,jit_int32_t);
> +# define ldxbi_f(r0,r1,i0) LFSU(r0,r1,i0)
> +# define ldxbi_d(r0,r1,i0) LFDU(r0,r1,i0)
> +# define ldxai_f(r0,r1,i0) generic_ldxai_f(r0,r1,i0)
> +# define ldxai_d(r0,r1,i0) generic_ldxai_d(r0,r1,i0)
> +# define stxbi_f(i0,r0,r1) STFSU(r1,r0,i0)
> +# define stxbi_d(i0,r0,r1) STFDU(r1,r0,i0)
> +# define stxai_f(i0,r0,r1) generic_stxai_f(i0,r0,r1)
> +# define stxai_d(i0,r0,r1) generic_stxai_d(i0,r0,r1)
> #endif
>
> #if CODE
> diff --git a/lib/jit_ppc.c b/lib/jit_ppc.c
> index 3b53aa6..af292f1 100644
> --- a/lib/jit_ppc.c
> +++ b/lib/jit_ppc.c
> @@ -1264,13 +1264,13 @@ _emit_code(jit_state_t *_jit)
> break
> #define case_rrx(name, type) \
> case jit_code_##name##i##type: \
> - generic_##name##i##type(rn(node->u.w), \
> - rn(node->v.w), node->w.w); \
> + name##i##type(rn(node->u.w), \
> + rn(node->v.w), node->w.w); \
> break
> #define case_xrr(name, type) \
> case jit_code_##name##i##type: \
> - generic_##name##i##type(node->u.w, rn(node->v.w), \
> - rn(node->w.w)); \
> + name##i##type(node->u.w, rn(node->v.w), \
> + rn(node->w.w)); \
> break
> #define case_rrrr(name, type) \
> case jit_code_##name##r##type: \
> --
> 2.43.0
Hi Paul,
Many thanks for the patch. I used it as the basis to write a new version.
Your patch would no longer apply due to other changes, it also would
not build in 64 due to the missing LDU definition, now added.
The new patch also adds assertions for the invalid encodings, and use an
an actual function, instead of a macro, to check the range of the
pre-increment, and if the instruction cannot be encoded, use a temporary.
Verified on ppc aix 32 bit, ppc linux 32 and 64 bit, and linux ppcle 64 bit.
The ppc port is now the first to be ready for the next Lightning release.
Thanks!
Paulo