target/arm: Vectorize SABD/UABD

Include 64-bit element size in preparation for SVE2.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20200513163245.17915-16-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

5 files changed, 176 insertions, 4 deletions

diff --git a/target/arm/helper.h b/target/arm/helper.h
index aed3050..4678d3a 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -731,6 +731,16 @@ DEF_HELPER_FLAGS_3(gvec_sli_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
 DEF_HELPER_FLAGS_3(gvec_sli_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
 DEF_HELPER_FLAGS_3(gvec_sli_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
 
+DEF_HELPER_FLAGS_4(gvec_sabd_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_sabd_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_sabd_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_sabd_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(gvec_uabd_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_uabd_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_uabd_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_uabd_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
 #ifdef TARGET_AARCH64
 #include "helper-a64.h"
 #include "helper-sve.h"
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index 4577df3..54b0655 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -11190,6 +11190,13 @@ static void disas_simd_3same_int(DisasContext *s, uint32_t insn)
             gen_gvec_fn3(s, is_q, rd, rn, rm, tcg_gen_gvec_smin, size);
         }
         return;
+    case 0xe: /* SABD, UABD */
+        if (u) {
+            gen_gvec_fn3(s, is_q, rd, rn, rm, gen_gvec_uabd, size);
+        } else {
+            gen_gvec_fn3(s, is_q, rd, rn, rm, gen_gvec_sabd, size);
+        }
+        return;
     case 0x10: /* ADD, SUB */
         if (u) {
             gen_gvec_fn3(s, is_q, rd, rn, rm, tcg_gen_gvec_sub, size);
@@ -11322,7 +11329,6 @@ static void disas_simd_3same_int(DisasContext *s, uint32_t insn)
                 genenvfn = fns[size][u];
                 break;
             }
-            case 0xe: /* SABD, UABD */
             case 0xf: /* SABA, UABA */
             {
                 static NeonGenTwoOpFn * const fns[3][2] = {
diff --git a/target/arm/translate.c b/target/arm/translate.c
index 21529a9..d288721 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -5102,6 +5102,126 @@ void gen_gvec_sqsub_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
                    rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
 }
 
+static void gen_sabd_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
+{
+    TCGv_i32 t = tcg_temp_new_i32();
+
+    tcg_gen_sub_i32(t, a, b);
+    tcg_gen_sub_i32(d, b, a);
+    tcg_gen_movcond_i32(TCG_COND_LT, d, a, b, d, t);
+    tcg_temp_free_i32(t);
+}
+
+static void gen_sabd_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
+{
+    TCGv_i64 t = tcg_temp_new_i64();
+
+    tcg_gen_sub_i64(t, a, b);
+    tcg_gen_sub_i64(d, b, a);
+    tcg_gen_movcond_i64(TCG_COND_LT, d, a, b, d, t);
+    tcg_temp_free_i64(t);
+}
+
+static void gen_sabd_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
+{
+    TCGv_vec t = tcg_temp_new_vec_matching(d);
+
+    tcg_gen_smin_vec(vece, t, a, b);
+    tcg_gen_smax_vec(vece, d, a, b);
+    tcg_gen_sub_vec(vece, d, d, t);
+    tcg_temp_free_vec(t);
+}
+
+void gen_gvec_sabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
+                   uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
+{
+    static const TCGOpcode vecop_list[] = {
+        INDEX_op_sub_vec, INDEX_op_smin_vec, INDEX_op_smax_vec, 0
+    };
+    static const GVecGen3 ops[4] = {
+        { .fniv = gen_sabd_vec,
+          .fno = gen_helper_gvec_sabd_b,
+          .opt_opc = vecop_list,
+          .vece = MO_8 },
+        { .fniv = gen_sabd_vec,
+          .fno = gen_helper_gvec_sabd_h,
+          .opt_opc = vecop_list,
+          .vece = MO_16 },
+        { .fni4 = gen_sabd_i32,
+          .fniv = gen_sabd_vec,
+          .fno = gen_helper_gvec_sabd_s,
+          .opt_opc = vecop_list,
+          .vece = MO_32 },
+        { .fni8 = gen_sabd_i64,
+          .fniv = gen_sabd_vec,
+          .fno = gen_helper_gvec_sabd_d,
+          .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+          .opt_opc = vecop_list,
+          .vece = MO_64 },
+    };
+    tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
+}
+
+static void gen_uabd_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
+{
+    TCGv_i32 t = tcg_temp_new_i32();
+
+    tcg_gen_sub_i32(t, a, b);
+    tcg_gen_sub_i32(d, b, a);
+    tcg_gen_movcond_i32(TCG_COND_LTU, d, a, b, d, t);
+    tcg_temp_free_i32(t);
+}
+
+static void gen_uabd_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
+{
+    TCGv_i64 t = tcg_temp_new_i64();
+
+    tcg_gen_sub_i64(t, a, b);
+    tcg_gen_sub_i64(d, b, a);
+    tcg_gen_movcond_i64(TCG_COND_LTU, d, a, b, d, t);
+    tcg_temp_free_i64(t);
+}
+
+static void gen_uabd_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
+{
+    TCGv_vec t = tcg_temp_new_vec_matching(d);
+
+    tcg_gen_umin_vec(vece, t, a, b);
+    tcg_gen_umax_vec(vece, d, a, b);
+    tcg_gen_sub_vec(vece, d, d, t);
+    tcg_temp_free_vec(t);
+}
+
+void gen_gvec_uabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
+                   uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
+{
+    static const TCGOpcode vecop_list[] = {
+        INDEX_op_sub_vec, INDEX_op_umin_vec, INDEX_op_umax_vec, 0
+    };
+    static const GVecGen3 ops[4] = {
+        { .fniv = gen_uabd_vec,
+          .fno = gen_helper_gvec_uabd_b,
+          .opt_opc = vecop_list,
+          .vece = MO_8 },
+        { .fniv = gen_uabd_vec,
+          .fno = gen_helper_gvec_uabd_h,
+          .opt_opc = vecop_list,
+          .vece = MO_16 },
+        { .fni4 = gen_uabd_i32,
+          .fniv = gen_uabd_vec,
+          .fno = gen_helper_gvec_uabd_s,
+          .opt_opc = vecop_list,
+          .vece = MO_32 },
+        { .fni8 = gen_uabd_i64,
+          .fniv = gen_uabd_vec,
+          .fno = gen_helper_gvec_uabd_d,
+          .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+          .opt_opc = vecop_list,
+          .vece = MO_64 },
+    };
+    tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
+}
+
 /* Translate a NEON data processing instruction.  Return nonzero if the
    instruction is invalid.
    We process data in a mixture of 32-bit and 64-bit chunks.
@@ -5236,6 +5356,16 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
             }
             return 1;
 
+        case NEON_3R_VABD:
+            if (u) {
+                gen_gvec_uabd(size, rd_ofs, rn_ofs, rm_ofs,
+                              vec_size, vec_size);
+            } else {
+                gen_gvec_sabd(size, rd_ofs, rn_ofs, rm_ofs,
+                              vec_size, vec_size);
+            }
+            return 0;
+
         case NEON_3R_VADD_VSUB:
         case NEON_3R_LOGIC:
         case NEON_3R_VMAX:
@@ -5380,9 +5510,6 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
         case NEON_3R_VQRSHL:
             GEN_NEON_INTEGER_OP_ENV(qrshl);
             break;
-        case NEON_3R_VABD:
-            GEN_NEON_INTEGER_OP(abd);
-            break;
         case NEON_3R_VABA:
             GEN_NEON_INTEGER_OP(abd);
             tcg_temp_free_i32(tmp2);
diff --git a/target/arm/translate.h b/target/arm/translate.h
index aea8a97..bbfe3d1 100644
--- a/target/arm/translate.h
+++ b/target/arm/translate.h
@@ -337,6 +337,11 @@ void gen_gvec_sqrdmlah_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
 void gen_gvec_sqrdmlsh_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
                           uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz);
 
+void gen_gvec_sabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
+                   uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_uabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
+                   uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz);
+
 /*
  * Forward to the isar_feature_* tests given a DisasContext pointer.
  */
diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c
index a483841..a4972d0 100644
--- a/target/arm/vec_helper.c
+++ b/target/arm/vec_helper.c
@@ -1407,3 +1407,27 @@ DO_CMP0(gvec_cgt0_h, int16_t, >)
 DO_CMP0(gvec_cge0_h, int16_t, >=)
 
 #undef DO_CMP0
+
+#define DO_ABD(NAME, TYPE)                                      \
+void HELPER(NAME)(void *vd, void *vn, void *vm, uint32_t desc)  \
+{                                                               \
+    intptr_t i, opr_sz = simd_oprsz(desc);                      \
+    TYPE *d = vd, *n = vn, *m = vm;                             \
+                                                                \
+    for (i = 0; i < opr_sz / sizeof(TYPE); ++i) {               \
+        d[i] = n[i] < m[i] ? m[i] - n[i] : n[i] - m[i];         \
+    }                                                           \
+    clear_tail(d, opr_sz, simd_maxsz(desc));                    \
+}
+
+DO_ABD(gvec_sabd_b, int8_t)
+DO_ABD(gvec_sabd_h, int16_t)
+DO_ABD(gvec_sabd_s, int32_t)
+DO_ABD(gvec_sabd_d, int64_t)
+
+DO_ABD(gvec_uabd_b, uint8_t)
+DO_ABD(gvec_uabd_h, uint16_t)
+DO_ABD(gvec_uabd_s, uint32_t)
+DO_ABD(gvec_uabd_d, uint64_t)
+
+#undef DO_ABD