[Qemu-devel] [PATCH 0/2] target/ppc: make use of new gvec expanders

[Qemu-devel] [PATCH 0/2] target/ppc: make use of new gvec expanders

[Richard Henderson]

Based-on: <20190518190157.21255-1-richard.henderson@linaro.org>
Aka "tcg: misc gvec improvements".

Since Mark's initial patches, we've added (or are adding)
generic support for variable vector shifts and bitsel.


r~


Richard Henderson (2):
  target/ppc: Use vector variable shifts for VSL, VSR, VSRA
  target/ppc: Use tcg_gen_gvec_bitsel

 target/ppc/helper.h                 | 12 ----------
 target/ppc/int_helper.c             | 37 -----------------------------
 target/ppc/translate/vmx-impl.inc.c | 24 +++++++++----------
 target/ppc/translate/vsx-impl.inc.c | 24 ++-----------------
 4 files changed, 14 insertions(+), 83 deletions(-)

-- 
2.17.1

[Qemu-devel] [PATCH 1/2] target/ppc: Use vector variable shifts for VSL, VSR, VSRA

[Richard Henderson]

The gvec expanders take care of masking the shift amount
against the element width.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/ppc/helper.h                 | 12 ----------
 target/ppc/int_helper.c             | 37 -----------------------------
 target/ppc/translate/vmx-impl.inc.c | 24 +++++++++----------
 3 files changed, 12 insertions(+), 61 deletions(-)

diff --git a/target/ppc/helper.h b/target/ppc/helper.h
index 638a6e99c4..02b67a333e 100644
--- a/target/ppc/helper.h
+++ b/target/ppc/helper.h
@@ -180,18 +180,6 @@ DEF_HELPER_3(vmuloub, void, avr, avr, avr)
 DEF_HELPER_3(vmulouh, void, avr, avr, avr)
 DEF_HELPER_3(vmulouw, void, avr, avr, avr)
 DEF_HELPER_3(vmuluwm, void, avr, avr, avr)
-DEF_HELPER_3(vsrab, void, avr, avr, avr)
-DEF_HELPER_3(vsrah, void, avr, avr, avr)
-DEF_HELPER_3(vsraw, void, avr, avr, avr)
-DEF_HELPER_3(vsrad, void, avr, avr, avr)
-DEF_HELPER_3(vsrb, void, avr, avr, avr)
-DEF_HELPER_3(vsrh, void, avr, avr, avr)
-DEF_HELPER_3(vsrw, void, avr, avr, avr)
-DEF_HELPER_3(vsrd, void, avr, avr, avr)
-DEF_HELPER_3(vslb, void, avr, avr, avr)
-DEF_HELPER_3(vslh, void, avr, avr, avr)
-DEF_HELPER_3(vslw, void, avr, avr, avr)
-DEF_HELPER_3(vsld, void, avr, avr, avr)
 DEF_HELPER_3(vslo, void, avr, avr, avr)
 DEF_HELPER_3(vsro, void, avr, avr, avr)
 DEF_HELPER_3(vsrv, void, avr, avr, avr)
diff --git a/target/ppc/int_helper.c b/target/ppc/int_helper.c
index f6a088ac08..40a7035df0 100644
--- a/target/ppc/int_helper.c
+++ b/target/ppc/int_helper.c
@@ -1776,23 +1776,6 @@ VSHIFT(l, 1)
 VSHIFT(r, 0)
 #undef VSHIFT
 
-#define VSL(suffix, element, mask)                                      \
-    void helper_vsl##suffix(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)   \
-    {                                                                   \
-        int i;                                                          \
-                                                                        \
-        for (i = 0; i < ARRAY_SIZE(r->element); i++) {                  \
-            unsigned int shift = b->element[i] & mask;                  \
-                                                                        \
-            r->element[i] = a->element[i] << shift;                     \
-        }                                                               \
-    }
-VSL(b, u8, 0x7)
-VSL(h, u16, 0x0F)
-VSL(w, u32, 0x1F)
-VSL(d, u64, 0x3F)
-#undef VSL
-
 void helper_vslv(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
 {
     int i;
@@ -1965,26 +1948,6 @@ VNEG(vnegw, s32)
 VNEG(vnegd, s64)
 #undef VNEG
 
-#define VSR(suffix, element, mask)                                      \
-    void helper_vsr##suffix(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)   \
-    {                                                                   \
-        int i;                                                          \
-                                                                        \
-        for (i = 0; i < ARRAY_SIZE(r->element); i++) {                  \
-            unsigned int shift = b->element[i] & mask;                  \
-            r->element[i] = a->element[i] >> shift;                     \
-        }                                                               \
-    }
-VSR(ab, s8, 0x7)
-VSR(ah, s16, 0xF)
-VSR(aw, s32, 0x1F)
-VSR(ad, s64, 0x3F)
-VSR(b, u8, 0x7)
-VSR(h, u16, 0xF)
-VSR(w, u32, 0x1F)
-VSR(d, u64, 0x3F)
-#undef VSR
-
 void helper_vsro(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
 {
     int sh = (b->VsrB(0xf) >> 3) & 0xf;
diff --git a/target/ppc/translate/vmx-impl.inc.c b/target/ppc/translate/vmx-impl.inc.c
index 6861f4c5b9..663275b729 100644
--- a/target/ppc/translate/vmx-impl.inc.c
+++ b/target/ppc/translate/vmx-impl.inc.c
@@ -530,21 +530,21 @@ GEN_VXFORM(vmuleuw, 4, 10);
 GEN_VXFORM(vmulesb, 4, 12);
 GEN_VXFORM(vmulesh, 4, 13);
 GEN_VXFORM(vmulesw, 4, 14);
-GEN_VXFORM(vslb, 2, 4);
-GEN_VXFORM(vslh, 2, 5);
-GEN_VXFORM(vslw, 2, 6);
+GEN_VXFORM_V(vslb, MO_8, tcg_gen_gvec_shlv, 2, 4);
+GEN_VXFORM_V(vslh, MO_16, tcg_gen_gvec_shlv, 2, 5);
+GEN_VXFORM_V(vslw, MO_32, tcg_gen_gvec_shlv, 2, 6);
 GEN_VXFORM(vrlwnm, 2, 6);
 GEN_VXFORM_DUAL(vslw, PPC_ALTIVEC, PPC_NONE, \
                 vrlwnm, PPC_NONE, PPC2_ISA300)
-GEN_VXFORM(vsld, 2, 23);
-GEN_VXFORM(vsrb, 2, 8);
-GEN_VXFORM(vsrh, 2, 9);
-GEN_VXFORM(vsrw, 2, 10);
-GEN_VXFORM(vsrd, 2, 27);
-GEN_VXFORM(vsrab, 2, 12);
-GEN_VXFORM(vsrah, 2, 13);
-GEN_VXFORM(vsraw, 2, 14);
-GEN_VXFORM(vsrad, 2, 15);
+GEN_VXFORM_V(vsld, MO_64, tcg_gen_gvec_shlv, 2, 23);
+GEN_VXFORM_V(vsrb, MO_8, tcg_gen_gvec_shrv, 2, 8);
+GEN_VXFORM_V(vsrh, MO_16, tcg_gen_gvec_shrv, 2, 9);
+GEN_VXFORM_V(vsrw, MO_32, tcg_gen_gvec_shrv, 2, 10);
+GEN_VXFORM_V(vsrd, MO_64, tcg_gen_gvec_shrv, 2, 27);
+GEN_VXFORM_V(vsrab, MO_8, tcg_gen_gvec_sarv, 2, 12);
+GEN_VXFORM_V(vsrah, MO_16, tcg_gen_gvec_sarv, 2, 13);
+GEN_VXFORM_V(vsraw, MO_32, tcg_gen_gvec_sarv, 2, 14);
+GEN_VXFORM_V(vsrad, MO_64, tcg_gen_gvec_sarv, 2, 15);
 GEN_VXFORM(vsrv, 2, 28);
 GEN_VXFORM(vslv, 2, 29);
 GEN_VXFORM(vslo, 6, 16);
-- 
2.17.1

Re: [Qemu-devel] [PATCH 1/2] target/ppc: Use vector variable shifts for VSL, VSR, VSRA

[Aleksandar Markovic]

On May 18, 2019 9:21 PM, "Richard Henderson" <richard.henderson@linaro.org>
wrote:
>
> The gvec expanders take care of masking the shift amount
> against the element width.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  target/ppc/helper.h                 | 12 ----------
>  target/ppc/int_helper.c             | 37 -----------------------------
>  target/ppc/translate/vmx-impl.inc.c | 24 +++++++++----------

You changed the line

-GEN_VXFORM(vslw, 2, 6);

to be:

+GEN_VXFORM_V(vslw, MO_32, tcg_gen_gvec_shlv, 2, 6);

and left this line unchanged (even though it deals with the same vslw
instruction):

GEN_VXFORM_DUAL(vslw, PPC_ALTIVEC, PPC_NONE, \                 vrlwnm,
PPC_NONE, PPC2_ISA300)

I just want to doublecheck - is this really what you wanted to do?

Thanks,
Aleksandar

>  3 files changed, 12 insertions(+), 61 deletions(-)
>
> diff --git a/target/ppc/helper.h b/target/ppc/helper.h
> index 638a6e99c4..02b67a333e 100644
> --- a/target/ppc/helper.h
> +++ b/target/ppc/helper.h
> @@ -180,18 +180,6 @@ DEF_HELPER_3(vmuloub, void, avr, avr, avr)
>  DEF_HELPER_3(vmulouh, void, avr, avr, avr)
>  DEF_HELPER_3(vmulouw, void, avr, avr, avr)
>  DEF_HELPER_3(vmuluwm, void, avr, avr, avr)
> -DEF_HELPER_3(vsrab, void, avr, avr, avr)
> -DEF_HELPER_3(vsrah, void, avr, avr, avr)
> -DEF_HELPER_3(vsraw, void, avr, avr, avr)
> -DEF_HELPER_3(vsrad, void, avr, avr, avr)
> -DEF_HELPER_3(vsrb, void, avr, avr, avr)
> -DEF_HELPER_3(vsrh, void, avr, avr, avr)
> -DEF_HELPER_3(vsrw, void, avr, avr, avr)
> -DEF_HELPER_3(vsrd, void, avr, avr, avr)
> -DEF_HELPER_3(vslb, void, avr, avr, avr)
> -DEF_HELPER_3(vslh, void, avr, avr, avr)
> -DEF_HELPER_3(vslw, void, avr, avr, avr)
> -DEF_HELPER_3(vsld, void, avr, avr, avr)
>  DEF_HELPER_3(vslo, void, avr, avr, avr)
>  DEF_HELPER_3(vsro, void, avr, avr, avr)
>  DEF_HELPER_3(vsrv, void, avr, avr, avr)
> diff --git a/target/ppc/int_helper.c b/target/ppc/int_helper.c
> index f6a088ac08..40a7035df0 100644
> --- a/target/ppc/int_helper.c
> +++ b/target/ppc/int_helper.c
> @@ -1776,23 +1776,6 @@ VSHIFT(l, 1)
>  VSHIFT(r, 0)
>  #undef VSHIFT
>
> -#define VSL(suffix, element, mask)                                      \
> -    void helper_vsl##suffix(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)   \
> -    {                                                                   \
> -        int i;                                                          \
> -                                                                        \
> -        for (i = 0; i < ARRAY_SIZE(r->element); i++) {                  \
> -            unsigned int shift = b->element[i] & mask;                  \
> -                                                                        \
> -            r->element[i] = a->element[i] << shift;                     \
> -        }                                                               \
> -    }
> -VSL(b, u8, 0x7)
> -VSL(h, u16, 0x0F)
> -VSL(w, u32, 0x1F)
> -VSL(d, u64, 0x3F)
> -#undef VSL
> -
>  void helper_vslv(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
>  {
>      int i;
> @@ -1965,26 +1948,6 @@ VNEG(vnegw, s32)
>  VNEG(vnegd, s64)
>  #undef VNEG
>
> -#define VSR(suffix, element, mask)                                      \
> -    void helper_vsr##suffix(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)   \
> -    {                                                                   \
> -        int i;                                                          \
> -                                                                        \
> -        for (i = 0; i < ARRAY_SIZE(r->element); i++) {                  \
> -            unsigned int shift = b->element[i] & mask;                  \
> -            r->element[i] = a->element[i] >> shift;                     \
> -        }                                                               \
> -    }
> -VSR(ab, s8, 0x7)
> -VSR(ah, s16, 0xF)
> -VSR(aw, s32, 0x1F)
> -VSR(ad, s64, 0x3F)
> -VSR(b, u8, 0x7)
> -VSR(h, u16, 0xF)
> -VSR(w, u32, 0x1F)
> -VSR(d, u64, 0x3F)
> -#undef VSR
> -
>  void helper_vsro(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
>  {
>      int sh = (b->VsrB(0xf) >> 3) & 0xf;
> diff --git a/target/ppc/translate/vmx-impl.inc.c
b/target/ppc/translate/vmx-impl.inc.c
> index 6861f4c5b9..663275b729 100644
> --- a/target/ppc/translate/vmx-impl.inc.c
> +++ b/target/ppc/translate/vmx-impl.inc.c
> @@ -530,21 +530,21 @@ GEN_VXFORM(vmuleuw, 4, 10);
>  GEN_VXFORM(vmulesb, 4, 12);
>  GEN_VXFORM(vmulesh, 4, 13);
>  GEN_VXFORM(vmulesw, 4, 14);
> -GEN_VXFORM(vslb, 2, 4);
> -GEN_VXFORM(vslh, 2, 5);
> -GEN_VXFORM(vslw, 2, 6);
> +GEN_VXFORM_V(vslb, MO_8, tcg_gen_gvec_shlv, 2, 4);
> +GEN_VXFORM_V(vslh, MO_16, tcg_gen_gvec_shlv, 2, 5);
> +GEN_VXFORM_V(vslw, MO_32, tcg_gen_gvec_shlv, 2, 6);
>  GEN_VXFORM(vrlwnm, 2, 6);
>  GEN_VXFORM_DUAL(vslw, PPC_ALTIVEC, PPC_NONE, \
>                  vrlwnm, PPC_NONE, PPC2_ISA300)
> -GEN_VXFORM(vsld, 2, 23);
> -GEN_VXFORM(vsrb, 2, 8);
> -GEN_VXFORM(vsrh, 2, 9);
> -GEN_VXFORM(vsrw, 2, 10);
> -GEN_VXFORM(vsrd, 2, 27);
> -GEN_VXFORM(vsrab, 2, 12);
> -GEN_VXFORM(vsrah, 2, 13);
> -GEN_VXFORM(vsraw, 2, 14);
> -GEN_VXFORM(vsrad, 2, 15);
> +GEN_VXFORM_V(vsld, MO_64, tcg_gen_gvec_shlv, 2, 23);
> +GEN_VXFORM_V(vsrb, MO_8, tcg_gen_gvec_shrv, 2, 8);
> +GEN_VXFORM_V(vsrh, MO_16, tcg_gen_gvec_shrv, 2, 9);
> +GEN_VXFORM_V(vsrw, MO_32, tcg_gen_gvec_shrv, 2, 10);
> +GEN_VXFORM_V(vsrd, MO_64, tcg_gen_gvec_shrv, 2, 27);
> +GEN_VXFORM_V(vsrab, MO_8, tcg_gen_gvec_sarv, 2, 12);
> +GEN_VXFORM_V(vsrah, MO_16, tcg_gen_gvec_sarv, 2, 13);
> +GEN_VXFORM_V(vsraw, MO_32, tcg_gen_gvec_sarv, 2, 14);
> +GEN_VXFORM_V(vsrad, MO_64, tcg_gen_gvec_sarv, 2, 15);
>  GEN_VXFORM(vsrv, 2, 28);
>  GEN_VXFORM(vslv, 2, 29);
>  GEN_VXFORM(vslo, 6, 16);
> --
> 2.17.1
>
>

Re: [Qemu-devel] [PATCH 1/2] target/ppc: Use vector variable shifts for VSL, VSR, VSRA

[Richard Henderson]

On 5/20/19 2:49 AM, Aleksandar Markovic wrote:
> 
> On May 18, 2019 9:21 PM, "Richard Henderson" <richard.henderson@linaro.org
> <mailto:richard.henderson@linaro.org>> wrote:
>>
>> The gvec expanders take care of masking the shift amount
>> against the element width.
>>
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org
> <mailto:richard.henderson@linaro.org>>
>> ---
>>  target/ppc/helper.h                 | 12 ----------
>>  target/ppc/int_helper.c             | 37 -----------------------------
>>  target/ppc/translate/vmx-impl.inc.c | 24 +++++++++----------
> 
> You changed the line
> 
> -GEN_VXFORM(vslw, 2, 6);
> 
> to be:
> 
> +GEN_VXFORM_V(vslw, MO_32, tcg_gen_gvec_shlv, 2, 6);
> 
> and left this line unchanged (even though it deals with the same vslw instruction):
> 
> GEN_VXFORM_DUAL(vslw, PPC_ALTIVEC, PPC_NONE, \                 vrlwnm,
> PPC_NONE, PPC2_ISA300)
> 
> I just want to doublecheck - is this really what you wanted to do?

Yes, the macros do two different things.

The first defines a function using tcg_gen_gvec_shlv as the implementation.

The second defines a function that chooses between two overloaded encodings,
depending on whether PPC_ALTIVEC or PPC2_ISA300 is enabled.  If PPC_ALTIVEC, it
will forward the implementation to the function defined with the first macro.


r~

[Qemu-devel] [PATCH 2/2] target/ppc: Use tcg_gen_gvec_bitsel

[Richard Henderson]

Replace the target-specific implementation of XXSEL.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/ppc/translate/vsx-impl.inc.c | 24 ++----------------------
 1 file changed, 2 insertions(+), 22 deletions(-)

diff --git a/target/ppc/translate/vsx-impl.inc.c b/target/ppc/translate/vsx-impl.inc.c
index 11d9b75d01..7a5d0e1f46 100644
--- a/target/ppc/translate/vsx-impl.inc.c
+++ b/target/ppc/translate/vsx-impl.inc.c
@@ -1290,28 +1290,8 @@ static void glue(gen_, name)(DisasContext *ctx)             \
 VSX_XXMRG(xxmrghw, 1)
 VSX_XXMRG(xxmrglw, 0)
 
-static void xxsel_i64(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b, TCGv_i64 c)
-{
-    tcg_gen_and_i64(b, b, c);
-    tcg_gen_andc_i64(a, a, c);
-    tcg_gen_or_i64(t, a, b);
-}
-
-static void xxsel_vec(unsigned vece, TCGv_vec t, TCGv_vec a,
-                      TCGv_vec b, TCGv_vec c)
-{
-    tcg_gen_and_vec(vece, b, b, c);
-    tcg_gen_andc_vec(vece, a, a, c);
-    tcg_gen_or_vec(vece, t, a, b);
-}
-
 static void gen_xxsel(DisasContext *ctx)
 {
-    static const GVecGen4 g = {
-        .fni8 = xxsel_i64,
-        .fniv = xxsel_vec,
-        .vece = MO_64,
-    };
     int rt = xT(ctx->opcode);
     int ra = xA(ctx->opcode);
     int rb = xB(ctx->opcode);
@@ -1321,8 +1301,8 @@ static void gen_xxsel(DisasContext *ctx)
         gen_exception(ctx, POWERPC_EXCP_VSXU);
         return;
     }
-    tcg_gen_gvec_4(vsr_full_offset(rt), vsr_full_offset(ra),
-                   vsr_full_offset(rb), vsr_full_offset(rc), 16, 16, &g);
+    tcg_gen_gvec_bitsel(MO_64, vsr_full_offset(rt), vsr_full_offset(rc),
+                        vsr_full_offset(rb), vsr_full_offset(ra), 16, 16);
 }
 
 static void gen_xxspltw(DisasContext *ctx)
-- 
2.17.1

Re: [Qemu-devel] [PATCH 0/2] target/ppc: make use of new gvec expanders

[David Gibson]

[-- Attachment #1: Type: text/plain, Size: 990 bytes --]

On Sat, May 18, 2019 at 12:14:28PM -0700, Richard Henderson wrote:
> Based-on: <20190518190157.21255-1-richard.henderson@linaro.org>
> Aka "tcg: misc gvec improvements".
> 
> Since Mark's initial patches, we've added (or are adding)
> generic support for variable vector shifts and bitsel.

Applied, thanks.

> 
> 
> r~
> 
> 
> Richard Henderson (2):
>   target/ppc: Use vector variable shifts for VSL, VSR, VSRA
>   target/ppc: Use tcg_gen_gvec_bitsel
> 
>  target/ppc/helper.h                 | 12 ----------
>  target/ppc/int_helper.c             | 37 -----------------------------
>  target/ppc/translate/vmx-impl.inc.c | 24 +++++++++----------
>  target/ppc/translate/vsx-impl.inc.c | 24 ++-----------------
>  4 files changed, 14 insertions(+), 83 deletions(-)
> 

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [Qemu-devel] [PATCH 0/2] target/ppc: make use of new gvec expanders

[David Gibson]

[-- Attachment #1: Type: text/plain, Size: 1888 bytes --]

On Mon, May 20, 2019 at 02:01:42PM +1000, David Gibson wrote:
> On Sat, May 18, 2019 at 12:14:28PM -0700, Richard Henderson wrote:
> > Based-on: <20190518190157.21255-1-richard.henderson@linaro.org>
> > Aka "tcg: misc gvec improvements".
> > 
> > Since Mark's initial patches, we've added (or are adding)
> > generic support for variable vector shifts and bitsel.
> 
> Applied, thanks.

Urgh.. actually, I've removed 2/2 again.  I get this error:

In file included from /home/dwg/src/qemu/target/ppc/translate.c:6819:
/home/dwg/src/qemu/target/ppc/translate/vsx-impl.inc.c: In function ‘gen_xxsel’:
/home/dwg/src/qemu/target/ppc/translate/vsx-impl.inc.c:1352:5: error: implicit declaration of function ‘tcg_gen_gvec_bitsel’; did you mean ‘tcg_gen_gvec_2i_ool’? [-Werror=implicit-function-declaration]
 1352 |     tcg_gen_gvec_bitsel(MO_64, vsr_full_offset(rt), vsr_full_offset(rc),
      |     ^~~~~~~~~~~~~~~~~~~
      |     tcg_gen_gvec_2i_ool
/home/dwg/src/qemu/target/ppc/translate/vsx-impl.inc.c:1352:5: error: nested extern declaration of ‘tcg_gen_gvec_bitsel’ [-Werror=nested-externs]
cc1: all warnings being treated as errors


> 
> > 
> > 
> > r~
> > 
> > 
> > Richard Henderson (2):
> >   target/ppc: Use vector variable shifts for VSL, VSR, VSRA
> >   target/ppc: Use tcg_gen_gvec_bitsel
> > 
> >  target/ppc/helper.h                 | 12 ----------
> >  target/ppc/int_helper.c             | 37 -----------------------------
> >  target/ppc/translate/vmx-impl.inc.c | 24 +++++++++----------
> >  target/ppc/translate/vsx-impl.inc.c | 24 ++-----------------
> >  4 files changed, 14 insertions(+), 83 deletions(-)
> > 
> 



-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]