diff mbox series

[06/12] target/ppc: Move VAVG[SU][BHW] to decodetree and use gvec

Message ID 20220923214754.217819-7-lucas.araujo@eldorado.org.br
State New
Headers show
Series VMX/VSX instructions with gvec | expand

Commit Message

Lucas Mateus Martins Araujo e Castro Sept. 23, 2022, 9:47 p.m. UTC 
From: "Lucas Mateus Castro (alqotel)" <lucas.araujo@eldorado.org.br>

Moved the instructions VAVGUB, VAVGUH, VAVGUW, VAVGSB, VAVGSH, VAVGSW,
to decodetree and use gvec with them. For these one the right shift
had to be made before the sum as to avoid an overflow, so add 1 at the
end if any of the entries had 1 in its LSB as to replicate the "+ 1"
before the shift described by the ISA.

vavgub:
rept    loop    master             patch
8       12500   0,02616600         0,00754200 (-71.2%)
25      4000    0,02530000         0,00637700 (-74.8%)
100     1000    0,02604600         0,00790100 (-69.7%)
500     200     0,03189300         0,01838400 (-42.4%)
2500    40      0,06006900         0,06851000 (+14.1%)
8000    12      0,13941000         0,20548500 (+47.4%)

vavguh:
rept    loop    master             patch
8       12500   0,01818200         0,00780600 (-57.1%)
25      4000    0,01789300         0,00641600 (-64.1%)
100     1000    0,01899100         0,00787200 (-58.5%)
500     200     0,02527200         0,01828400 (-27.7%)
2500    40      0,05361800         0,06773000 (+26.3%)
8000    12      0,12886600         0,20291400 (+57.5%)

vavguw:
rept    loop    master             patch
8       12500   0,01423100         0,00776600 (-45.4%)
25      4000    0,01780800         0,00638600 (-64.1%)
100     1000    0,02085500         0,00787000 (-62.3%)
500     200     0,02737100         0,01828800 (-33.2%)
2500    40      0,05572600         0,06774200 (+21.6%)
8000    12      0,13101700         0,20311600 (+55.0%)

vavgsb:
rept    loop    master             patch
8       12500   0,03006000         0,00788600 (-73.8%)
25      4000    0,02882200         0,00637800 (-77.9%)
100     1000    0,02958000         0,00791400 (-73.2%)
500     200     0,03548800         0,01860400 (-47.6%)
2500    40      0,06360000         0,06850800 (+7.7%)
8000    12      0,13816500         0,20550300 (+48.7%)

vavgsh:
rept    loop    master             patch
8       12500   0,01965900         0,00776600 (-60.5%)
25      4000    0,01875400         0,00638700 (-65.9%)
100     1000    0,01952200         0,00786900 (-59.7%)
500     200     0,02562000         0,01760300 (-31.3%)
2500    40      0,05384300         0,06742800 (+25.2%)
8000    12      0,13240800         0,20330000 (+53.5%)

vavgsw:
rept    loop    master             patch
8       12500   0,01407700         0,00775600 (-44.9%)
25      4000    0,01762300         0,00640000 (-63.7%)
100     1000    0,02046500         0,00788500 (-61.5%)
500     200     0,02745600         0,01843000 (-32.9%)
2500    40      0,05375500         0,06820500 (+26.9%)
8000    12      0,13068300         0,20304900 (+55.4%)

These results to me seems to indicate that with gvec the results have a
slower translation but faster execution.

Signed-off-by: Lucas Mateus Castro (alqotel) <lucas.araujo@eldorado.org.br>
---
 target/ppc/helper.h                 |  12 +--
 target/ppc/insn32.decode            |   9 +++
 target/ppc/int_helper.c             |  32 ++++----
 target/ppc/translate/vmx-impl.c.inc | 109 +++++++++++++++++++++++++---
 target/ppc/translate/vmx-ops.c.inc  |   9 +--
 5 files changed, 130 insertions(+), 41 deletions(-)

Comments

Richard Henderson Sept. 24, 2022, 2:35 p.m. UTC | #1 
On 9/23/22 21:47, Lucas Mateus Castro(alqotel) wrote:
> +    TCGv_vec one = tcg_temp_new_vec_matching(t);
> +    tcg_gen_dupi_vec(vece, one, 1);
> +    tcg_gen_or_vec(vece, tmp, a, b);
> +    tcg_gen_and_vec(vece, tmp, tmp, one);

tcg_constant_vec_matching.  With that,
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>


r~
diff mbox series

Patch

diff --git a/target/ppc/helper.h b/target/ppc/helper.h
index 6a43e32ad3..f88d9d3996 100644
--- a/target/ppc/helper.h
+++ b/target/ppc/helper.h
@@ -143,15 +143,15 @@  DEF_HELPER_FLAGS_1(ftsqrt, TCG_CALL_NO_RWG_SE, i32, i64)
 #define dh_ctype_acc ppc_acc_t *
 #define dh_typecode_acc dh_typecode_ptr
 
-DEF_HELPER_FLAGS_3(vavgub, TCG_CALL_NO_RWG, void, avr, avr, avr)
-DEF_HELPER_FLAGS_3(vavguh, TCG_CALL_NO_RWG, void, avr, avr, avr)
-DEF_HELPER_FLAGS_3(vavguw, TCG_CALL_NO_RWG, void, avr, avr, avr)
+DEF_HELPER_FLAGS_4(VAVGUB, TCG_CALL_NO_RWG, void, avr, avr, avr, i32)
+DEF_HELPER_FLAGS_4(VAVGUH, TCG_CALL_NO_RWG, void, avr, avr, avr, i32)
+DEF_HELPER_FLAGS_4(VAVGUW, TCG_CALL_NO_RWG, void, avr, avr, avr, i32)
 DEF_HELPER_FLAGS_3(vabsdub, TCG_CALL_NO_RWG, void, avr, avr, avr)
 DEF_HELPER_FLAGS_3(vabsduh, TCG_CALL_NO_RWG, void, avr, avr, avr)
 DEF_HELPER_FLAGS_3(vabsduw, TCG_CALL_NO_RWG, void, avr, avr, avr)
-DEF_HELPER_FLAGS_3(vavgsb, TCG_CALL_NO_RWG, void, avr, avr, avr)
-DEF_HELPER_FLAGS_3(vavgsh, TCG_CALL_NO_RWG, void, avr, avr, avr)
-DEF_HELPER_FLAGS_3(vavgsw, TCG_CALL_NO_RWG, void, avr, avr, avr)
+DEF_HELPER_FLAGS_4(VAVGSB, TCG_CALL_NO_RWG, void, avr, avr, avr, i32)
+DEF_HELPER_FLAGS_4(VAVGSH, TCG_CALL_NO_RWG, void, avr, avr, avr, i32)
+DEF_HELPER_FLAGS_4(VAVGSW, TCG_CALL_NO_RWG, void, avr, avr, avr, i32)
 DEF_HELPER_4(vcmpeqfp, void, env, avr, avr, avr)
 DEF_HELPER_4(vcmpgefp, void, env, avr, avr, avr)
 DEF_HELPER_4(vcmpgtfp, void, env, avr, avr, avr)
diff --git a/target/ppc/insn32.decode b/target/ppc/insn32.decode
index aa4968e6b9..38458c01de 100644
--- a/target/ppc/insn32.decode
+++ b/target/ppc/insn32.decode
@@ -519,6 +519,15 @@  VCMPNEZW        000100 ..... ..... ..... . 0110000111   @VC
 VCMPSQ          000100 ... -- ..... ..... 00101000001   @VX_bf
 VCMPUQ          000100 ... -- ..... ..... 00100000001   @VX_bf
 
+## Vector Integer Average Instructions
+
+VAVGSB          000100 ..... ..... ..... 10100000010    @VX
+VAVGSH          000100 ..... ..... ..... 10101000010    @VX
+VAVGSW          000100 ..... ..... ..... 10110000010    @VX
+VAVGUB          000100 ..... ..... ..... 10000000010    @VX
+VAVGUH          000100 ..... ..... ..... 10001000010    @VX
+VAVGUW          000100 ..... ..... ..... 10010000010    @VX
+
 ## Vector Bit Manipulation Instruction
 
 VGNB            000100 ..... -- ... ..... 10011001100   @VX_n
diff --git a/target/ppc/int_helper.c b/target/ppc/int_helper.c
index 64b2d44a66..0a142441e5 100644
--- a/target/ppc/int_helper.c
+++ b/target/ppc/int_helper.c
@@ -602,25 +602,23 @@  VARITHSAT_UNSIGNED(w, u32, uint64_t, cvtsduw)
 #undef VARITHSAT_SIGNED
 #undef VARITHSAT_UNSIGNED
 
-#define VAVG_DO(name, element, etype)                                   \
-    void helper_v##name(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)       \
-    {                                                                   \
-        int i;                                                          \
-                                                                        \
-        for (i = 0; i < ARRAY_SIZE(r->element); i++) {                  \
-            etype x = (etype)a->element[i] + (etype)b->element[i] + 1;  \
-            r->element[i] = x >> 1;                                     \
-        }                                                               \
+#define VAVG(name, element, etype)                                          \
+    void helper_##name(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b, uint32_t v)\
+    {                                                                       \
+        int i;                                                              \
+                                                                            \
+        for (i = 0; i < ARRAY_SIZE(r->element); i++) {                      \
+            etype x = (etype)a->element[i] + (etype)b->element[i] + 1;      \
+            r->element[i] = x >> 1;                                         \
+        }                                                                   \
     }
 
-#define VAVG(type, signed_element, signed_type, unsigned_element,       \
-             unsigned_type)                                             \
-    VAVG_DO(avgs##type, signed_element, signed_type)                    \
-    VAVG_DO(avgu##type, unsigned_element, unsigned_type)
-VAVG(b, s8, int16_t, u8, uint16_t)
-VAVG(h, s16, int32_t, u16, uint32_t)
-VAVG(w, s32, int64_t, u32, uint64_t)
-#undef VAVG_DO
+VAVG(VAVGSB, s8, int16_t)
+VAVG(VAVGUB, u8, uint16_t)
+VAVG(VAVGSH, s16, int32_t)
+VAVG(VAVGUH, u16, uint32_t)
+VAVG(VAVGSW, s32, int64_t)
+VAVG(VAVGUW, u32, uint64_t)
 #undef VAVG
 
 #define VABSDU_DO(name, element)                                        \
diff --git a/target/ppc/translate/vmx-impl.c.inc b/target/ppc/translate/vmx-impl.c.inc
index 06d91d1304..8ff7c6ff3a 100644
--- a/target/ppc/translate/vmx-impl.c.inc
+++ b/target/ppc/translate/vmx-impl.c.inc
@@ -431,21 +431,9 @@  GEN_VXFORM_V(vminsb, MO_8, tcg_gen_gvec_smin, 1, 12);
 GEN_VXFORM_V(vminsh, MO_16, tcg_gen_gvec_smin, 1, 13);
 GEN_VXFORM_V(vminsw, MO_32, tcg_gen_gvec_smin, 1, 14);
 GEN_VXFORM_V(vminsd, MO_64, tcg_gen_gvec_smin, 1, 15);
-GEN_VXFORM(vavgub, 1, 16);
 GEN_VXFORM(vabsdub, 1, 16);
-GEN_VXFORM_DUAL(vavgub, PPC_ALTIVEC, PPC_NONE, \
-                vabsdub, PPC_NONE, PPC2_ISA300)
-GEN_VXFORM(vavguh, 1, 17);
 GEN_VXFORM(vabsduh, 1, 17);
-GEN_VXFORM_DUAL(vavguh, PPC_ALTIVEC, PPC_NONE, \
-                vabsduh, PPC_NONE, PPC2_ISA300)
-GEN_VXFORM(vavguw, 1, 18);
 GEN_VXFORM(vabsduw, 1, 18);
-GEN_VXFORM_DUAL(vavguw, PPC_ALTIVEC, PPC_NONE, \
-                vabsduw, PPC_NONE, PPC2_ISA300)
-GEN_VXFORM(vavgsb, 1, 20);
-GEN_VXFORM(vavgsh, 1, 21);
-GEN_VXFORM(vavgsw, 1, 22);
 GEN_VXFORM(vmrghb, 6, 0);
 GEN_VXFORM(vmrghh, 6, 1);
 GEN_VXFORM(vmrghw, 6, 2);
@@ -3349,6 +3337,103 @@  TRANS(VMULHSD, do_vx_mulh, true , do_vx_vmulhd_i64)
 TRANS(VMULHUW, do_vx_mulh, false, do_vx_vmulhw_i64)
 TRANS(VMULHUD, do_vx_mulh, false, do_vx_vmulhd_i64)
 
+static void do_vavg(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b,
+                    void (*gen_shr_vec)(unsigned, TCGv_vec, TCGv_vec, int64_t))
+{
+    TCGv_vec tmp = tcg_temp_new_vec_matching(t);
+    TCGv_vec one = tcg_temp_new_vec_matching(t);
+    tcg_gen_dupi_vec(vece, one, 1);
+    tcg_gen_or_vec(vece, tmp, a, b);
+    tcg_gen_and_vec(vece, tmp, tmp, one);
+    gen_shr_vec(vece, a, a, 1);
+    gen_shr_vec(vece, b, b, 1);
+    tcg_gen_add_vec(vece, t, a, b);
+    tcg_gen_add_vec(vece, t, t, tmp);
+    tcg_temp_free_vec(tmp);
+    tcg_temp_free_vec(one);
+}
+
+QEMU_FLATTEN
+static void gen_vavgu(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b)
+{
+    do_vavg(vece, t, a, b, tcg_gen_shri_vec);
+}
+
+QEMU_FLATTEN
+static void gen_vavgs(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b)
+{
+    do_vavg(vece, t, a, b, tcg_gen_sari_vec);
+}
+
+static bool do_vx_vavg(DisasContext *ctx, arg_VX *a, int sign, int vece)
+{
+    static const TCGOpcode vecop_list_s[] = {
+        INDEX_op_add_vec, INDEX_op_sari_vec, 0
+    };
+    static const TCGOpcode vecop_list_u[] = {
+        INDEX_op_add_vec, INDEX_op_shri_vec, 0
+    };
+
+    static const GVecGen3 op[2][3] = {
+        {
+            {
+                .fniv = gen_vavgu,
+                .fno = gen_helper_VAVGUB,
+                .opt_opc = vecop_list_u,
+                .vece = MO_8
+            },
+            {
+                .fniv = gen_vavgu,
+                .fno = gen_helper_VAVGUH,
+                .opt_opc = vecop_list_u,
+                .vece = MO_16
+            },
+            {
+                .fniv = gen_vavgu,
+                .fno = gen_helper_VAVGUW,
+                .opt_opc = vecop_list_u,
+                .vece = MO_32
+            },
+        },
+        {
+            {
+                .fniv = gen_vavgs,
+                .fno = gen_helper_VAVGSB,
+                .opt_opc = vecop_list_s,
+                .vece = MO_8
+            },
+            {
+                .fniv = gen_vavgs,
+                .fno = gen_helper_VAVGSH,
+                .opt_opc = vecop_list_s,
+                .vece = MO_16
+            },
+            {
+                .fniv = gen_vavgs,
+                .fno = gen_helper_VAVGSW,
+                .opt_opc = vecop_list_s,
+                .vece = MO_32
+            },
+        },
+    };
+
+    REQUIRE_VECTOR(ctx);
+
+    tcg_gen_gvec_3(avr_full_offset(a->vrt), avr_full_offset(a->vra),
+                   avr_full_offset(a->vrb), 16, 16, &op[sign][vece]);
+
+
+    return true;
+}
+
+
+TRANS_FLAGS(ALTIVEC, VAVGSB, do_vx_vavg, 1, MO_8)
+TRANS_FLAGS(ALTIVEC, VAVGSH, do_vx_vavg, 1, MO_16)
+TRANS_FLAGS(ALTIVEC, VAVGSW, do_vx_vavg, 1, MO_32)
+TRANS_FLAGS(ALTIVEC, VAVGUB, do_vx_vavg, 0, MO_8)
+TRANS_FLAGS(ALTIVEC, VAVGUH, do_vx_vavg, 0, MO_16)
+TRANS_FLAGS(ALTIVEC, VAVGUW, do_vx_vavg, 0, MO_32)
+
 static bool do_vdiv_vmod(DisasContext *ctx, arg_VX *a, const int vece,
                          void (*func_32)(TCGv_i32 t, TCGv_i32 a, TCGv_i32 b),
                          void (*func_64)(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b))
diff --git a/target/ppc/translate/vmx-ops.c.inc b/target/ppc/translate/vmx-ops.c.inc
index 46a620a232..02db51def0 100644
--- a/target/ppc/translate/vmx-ops.c.inc
+++ b/target/ppc/translate/vmx-ops.c.inc
@@ -83,12 +83,9 @@  GEN_VXFORM(vminsb, 1, 12),
 GEN_VXFORM(vminsh, 1, 13),
 GEN_VXFORM(vminsw, 1, 14),
 GEN_VXFORM_207(vminsd, 1, 15),
-GEN_VXFORM_DUAL(vavgub, vabsdub, 1, 16, PPC_ALTIVEC, PPC_NONE),
-GEN_VXFORM_DUAL(vavguh, vabsduh, 1, 17, PPC_ALTIVEC, PPC_NONE),
-GEN_VXFORM_DUAL(vavguw, vabsduw, 1, 18, PPC_ALTIVEC, PPC_NONE),
-GEN_VXFORM(vavgsb, 1, 20),
-GEN_VXFORM(vavgsh, 1, 21),
-GEN_VXFORM(vavgsw, 1, 22),
+GEN_VXFORM(vabsdub, 1, 16),
+GEN_VXFORM(vabsduh, 1, 17),
+GEN_VXFORM(vabsduw, 1, 18),
 GEN_VXFORM(vmrghb, 6, 0),
 GEN_VXFORM(vmrghh, 6, 1),
 GEN_VXFORM(vmrghw, 6, 2),