diff mbox series

[20/34] target/ppc: convert vsplt[bhw] to use vector operations

Message ID 20181218063911.2112-21-richard.henderson@linaro.org
State New
Headers show
Series tcg, target/ppc vector improvements | expand

Commit Message

Richard Henderson Dec. 18, 2018, 6:38 a.m. UTC 
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

---
 target/ppc/helper.h                 |  3 --
 target/ppc/int_helper.c             | 24 ---------------
 target/ppc/translate/vmx-impl.inc.c | 45 +++++++++++++++++------------
 3 files changed, 26 insertions(+), 46 deletions(-)

-- 
2.17.2

Comments

David Gibson Dec. 19, 2018, 6:32 a.m. UTC | #1 
On Mon, Dec 17, 2018 at 10:38:57PM -0800, Richard Henderson wrote:
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>


Acked-by: David Gibson <david@gibson.dropbear.id.au>


> ---

>  target/ppc/helper.h                 |  3 --

>  target/ppc/int_helper.c             | 24 ---------------

>  target/ppc/translate/vmx-impl.inc.c | 45 +++++++++++++++++------------

>  3 files changed, 26 insertions(+), 46 deletions(-)

> 

> diff --git a/target/ppc/helper.h b/target/ppc/helper.h

> index 2aa60e5d36..069daa9883 100644

> --- a/target/ppc/helper.h

> +++ b/target/ppc/helper.h

> @@ -246,9 +246,6 @@ DEF_HELPER_3(vrld, void, avr, avr, avr)

>  DEF_HELPER_3(vsl, void, avr, avr, avr)

>  DEF_HELPER_3(vsr, void, avr, avr, avr)

>  DEF_HELPER_4(vsldoi, void, avr, avr, avr, i32)

> -DEF_HELPER_3(vspltb, void, avr, avr, i32)

> -DEF_HELPER_3(vsplth, void, avr, avr, i32)

> -DEF_HELPER_3(vspltw, void, avr, avr, i32)

>  DEF_HELPER_3(vextractub, void, avr, avr, i32)

>  DEF_HELPER_3(vextractuh, void, avr, avr, i32)

>  DEF_HELPER_3(vextractuw, void, avr, avr, i32)

> diff --git a/target/ppc/int_helper.c b/target/ppc/int_helper.c

> index e44c0d90ee..3bf0fdb6c5 100644

> --- a/target/ppc/int_helper.c

> +++ b/target/ppc/int_helper.c

> @@ -1918,30 +1918,6 @@ void helper_vslo(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)

>  #endif

>  }

>  

> -/* Experimental testing shows that hardware masks the immediate.  */

> -#define _SPLAT_MASKED(element) (splat & (ARRAY_SIZE(r->element) - 1))

> -#if defined(HOST_WORDS_BIGENDIAN)

> -#define SPLAT_ELEMENT(element) _SPLAT_MASKED(element)

> -#else

> -#define SPLAT_ELEMENT(element)                                  \

> -    (ARRAY_SIZE(r->element) - 1 - _SPLAT_MASKED(element))

> -#endif

> -#define VSPLT(suffix, element)                                          \

> -    void helper_vsplt##suffix(ppc_avr_t *r, ppc_avr_t *b, uint32_t splat) \

> -    {                                                                   \

> -        uint32_t s = b->element[SPLAT_ELEMENT(element)];                \

> -        int i;                                                          \

> -                                                                        \

> -        for (i = 0; i < ARRAY_SIZE(r->element); i++) {                  \

> -            r->element[i] = s;                                          \

> -        }                                                               \

> -    }

> -VSPLT(b, u8)

> -VSPLT(h, u16)

> -VSPLT(w, u32)

> -#undef VSPLT

> -#undef SPLAT_ELEMENT

> -#undef _SPLAT_MASKED

>  #if defined(HOST_WORDS_BIGENDIAN)

>  #define VINSERT(suffix, element)                                            \

>      void helper_vinsert##suffix(ppc_avr_t *r, ppc_avr_t *b, uint32_t index) \

> diff --git a/target/ppc/translate/vmx-impl.inc.c b/target/ppc/translate/vmx-impl.inc.c

> index be638cdb1a..529ae0e5f5 100644

> --- a/target/ppc/translate/vmx-impl.inc.c

> +++ b/target/ppc/translate/vmx-impl.inc.c

> @@ -814,24 +814,31 @@ GEN_VXFORM_NOA(vprtybw, 1, 24);

>  GEN_VXFORM_NOA(vprtybd, 1, 24);

>  GEN_VXFORM_NOA(vprtybq, 1, 24);

>  

> -#define GEN_VXFORM_UIMM(name, opc2, opc3)                               \

> -static void glue(gen_, name)(DisasContext *ctx)                                 \

> -    {                                                                   \

> -        TCGv_ptr rb, rd;                                                \

> -        TCGv_i32 uimm;                                                  \

> -        if (unlikely(!ctx->altivec_enabled)) {                          \

> -            gen_exception(ctx, POWERPC_EXCP_VPU);                       \

> -            return;                                                     \

> -        }                                                               \

> -        uimm = tcg_const_i32(UIMM5(ctx->opcode));                       \

> -        rb = gen_avr_ptr(rB(ctx->opcode));                              \

> -        rd = gen_avr_ptr(rD(ctx->opcode));                              \

> -        gen_helper_##name (rd, rb, uimm);                               \

> -        tcg_temp_free_i32(uimm);                                        \

> -        tcg_temp_free_ptr(rb);                                          \

> -        tcg_temp_free_ptr(rd);                                          \

> +static void gen_vsplt(DisasContext *ctx, int vece)

> +{

> +    int uimm, dofs, bofs;

> +

> +    if (unlikely(!ctx->altivec_enabled)) {

> +        gen_exception(ctx, POWERPC_EXCP_VPU);

> +        return;

>      }

>  

> +    uimm = UIMM5(ctx->opcode);

> +    bofs = avr64_offset(rB(ctx->opcode), true);

> +    dofs = avr64_offset(rD(ctx->opcode), true);

> +

> +    /* Experimental testing shows that hardware masks the immediate.  */

> +    bofs += (uimm << vece) & 15;

> +#ifndef HOST_WORDS_BIGENDIAN

> +    bofs ^= 15;

> +#endif

> +

> +    tcg_gen_gvec_dup_mem(vece, dofs, bofs, 16, 16);

> +}

> +

> +#define GEN_VXFORM_VSPLT(name, vece, opc2, opc3) \

> +static void glue(gen_, name)(DisasContext *ctx) { gen_vsplt(ctx, vece); }

> +

>  #define GEN_VXFORM_UIMM_ENV(name, opc2, opc3)                           \

>  static void glue(gen_, name)(DisasContext *ctx)                         \

>      {                                                                   \

> @@ -873,9 +880,9 @@ static void glue(gen_, name)(DisasContext *ctx)                         \

>          tcg_temp_free_ptr(rd);                                          \

>      }

>  

> -GEN_VXFORM_UIMM(vspltb, 6, 8);

> -GEN_VXFORM_UIMM(vsplth, 6, 9);

> -GEN_VXFORM_UIMM(vspltw, 6, 10);

> +GEN_VXFORM_VSPLT(vspltb, MO_8, 6, 8);

> +GEN_VXFORM_VSPLT(vsplth, MO_16, 6, 9);

> +GEN_VXFORM_VSPLT(vspltw, MO_32, 6, 10);

>  GEN_VXFORM_UIMM_SPLAT(vextractub, 6, 8, 15);

>  GEN_VXFORM_UIMM_SPLAT(vextractuh, 6, 9, 14);

>  GEN_VXFORM_UIMM_SPLAT(vextractuw, 6, 10, 12);


-- 
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
diff mbox series

Patch

diff --git a/target/ppc/helper.h b/target/ppc/helper.h
index 2aa60e5d36..069daa9883 100644
--- a/target/ppc/helper.h
+++ b/target/ppc/helper.h
@@ -246,9 +246,6 @@  DEF_HELPER_3(vrld, void, avr, avr, avr)
 DEF_HELPER_3(vsl, void, avr, avr, avr)
 DEF_HELPER_3(vsr, void, avr, avr, avr)
 DEF_HELPER_4(vsldoi, void, avr, avr, avr, i32)
-DEF_HELPER_3(vspltb, void, avr, avr, i32)
-DEF_HELPER_3(vsplth, void, avr, avr, i32)
-DEF_HELPER_3(vspltw, void, avr, avr, i32)
 DEF_HELPER_3(vextractub, void, avr, avr, i32)
 DEF_HELPER_3(vextractuh, void, avr, avr, i32)
 DEF_HELPER_3(vextractuw, void, avr, avr, i32)
diff --git a/target/ppc/int_helper.c b/target/ppc/int_helper.c
index e44c0d90ee..3bf0fdb6c5 100644
--- a/target/ppc/int_helper.c
+++ b/target/ppc/int_helper.c
@@ -1918,30 +1918,6 @@  void helper_vslo(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
 #endif
 }
 
-/* Experimental testing shows that hardware masks the immediate.  */
-#define _SPLAT_MASKED(element) (splat & (ARRAY_SIZE(r->element) - 1))
-#if defined(HOST_WORDS_BIGENDIAN)
-#define SPLAT_ELEMENT(element) _SPLAT_MASKED(element)
-#else
-#define SPLAT_ELEMENT(element)                                  \
-    (ARRAY_SIZE(r->element) - 1 - _SPLAT_MASKED(element))
-#endif
-#define VSPLT(suffix, element)                                          \
-    void helper_vsplt##suffix(ppc_avr_t *r, ppc_avr_t *b, uint32_t splat) \
-    {                                                                   \
-        uint32_t s = b->element[SPLAT_ELEMENT(element)];                \
-        int i;                                                          \
-                                                                        \
-        for (i = 0; i < ARRAY_SIZE(r->element); i++) {                  \
-            r->element[i] = s;                                          \
-        }                                                               \
-    }
-VSPLT(b, u8)
-VSPLT(h, u16)
-VSPLT(w, u32)
-#undef VSPLT
-#undef SPLAT_ELEMENT
-#undef _SPLAT_MASKED
 #if defined(HOST_WORDS_BIGENDIAN)
 #define VINSERT(suffix, element)                                            \
     void helper_vinsert##suffix(ppc_avr_t *r, ppc_avr_t *b, uint32_t index) \
diff --git a/target/ppc/translate/vmx-impl.inc.c b/target/ppc/translate/vmx-impl.inc.c
index be638cdb1a..529ae0e5f5 100644
--- a/target/ppc/translate/vmx-impl.inc.c
+++ b/target/ppc/translate/vmx-impl.inc.c
@@ -814,24 +814,31 @@  GEN_VXFORM_NOA(vprtybw, 1, 24);
 GEN_VXFORM_NOA(vprtybd, 1, 24);
 GEN_VXFORM_NOA(vprtybq, 1, 24);
 
-#define GEN_VXFORM_UIMM(name, opc2, opc3)                               \
-static void glue(gen_, name)(DisasContext *ctx)                                 \
-    {                                                                   \
-        TCGv_ptr rb, rd;                                                \
-        TCGv_i32 uimm;                                                  \
-        if (unlikely(!ctx->altivec_enabled)) {                          \
-            gen_exception(ctx, POWERPC_EXCP_VPU);                       \
-            return;                                                     \
-        }                                                               \
-        uimm = tcg_const_i32(UIMM5(ctx->opcode));                       \
-        rb = gen_avr_ptr(rB(ctx->opcode));                              \
-        rd = gen_avr_ptr(rD(ctx->opcode));                              \
-        gen_helper_##name (rd, rb, uimm);                               \
-        tcg_temp_free_i32(uimm);                                        \
-        tcg_temp_free_ptr(rb);                                          \
-        tcg_temp_free_ptr(rd);                                          \
+static void gen_vsplt(DisasContext *ctx, int vece)
+{
+    int uimm, dofs, bofs;
+
+    if (unlikely(!ctx->altivec_enabled)) {
+        gen_exception(ctx, POWERPC_EXCP_VPU);
+        return;
     }
 
+    uimm = UIMM5(ctx->opcode);
+    bofs = avr64_offset(rB(ctx->opcode), true);
+    dofs = avr64_offset(rD(ctx->opcode), true);
+
+    /* Experimental testing shows that hardware masks the immediate.  */
+    bofs += (uimm << vece) & 15;
+#ifndef HOST_WORDS_BIGENDIAN
+    bofs ^= 15;
+#endif
+
+    tcg_gen_gvec_dup_mem(vece, dofs, bofs, 16, 16);
+}
+
+#define GEN_VXFORM_VSPLT(name, vece, opc2, opc3) \
+static void glue(gen_, name)(DisasContext *ctx) { gen_vsplt(ctx, vece); }
+
 #define GEN_VXFORM_UIMM_ENV(name, opc2, opc3)                           \
 static void glue(gen_, name)(DisasContext *ctx)                         \
     {                                                                   \
@@ -873,9 +880,9 @@  static void glue(gen_, name)(DisasContext *ctx)                         \
         tcg_temp_free_ptr(rd);                                          \
     }
 
-GEN_VXFORM_UIMM(vspltb, 6, 8);
-GEN_VXFORM_UIMM(vsplth, 6, 9);
-GEN_VXFORM_UIMM(vspltw, 6, 10);
+GEN_VXFORM_VSPLT(vspltb, MO_8, 6, 8);
+GEN_VXFORM_VSPLT(vsplth, MO_16, 6, 9);
+GEN_VXFORM_VSPLT(vspltw, MO_32, 6, 10);
 GEN_VXFORM_UIMM_SPLAT(vextractub, 6, 8, 15);
 GEN_VXFORM_UIMM_SPLAT(vextractuh, 6, 9, 14);
 GEN_VXFORM_UIMM_SPLAT(vextractuw, 6, 10, 12);