target/arm: Speed up aarch64 TBL/TBX

Always perform one call instead of two for 16-byte operands.
Use byte loads/stores directly into the vector register file
instead of extractions and deposits to a 64-bit local variable.

In order to easily receive pointers into the vector register file,
convert the helper to the gvec out-of-line signature.  Move the
helper into vec_helper.c, where it can make use of H1 and clear_tail.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Message-id: 20210224230532.276878-1-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

1 files changed, 7 insertions, 51 deletions

diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index 3aebdb4..b591f09 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -7532,10 +7532,8 @@ static void disas_simd_tb(DisasContext *s, uint32_t insn)
     int rm = extract32(insn, 16, 5);
     int rn = extract32(insn, 5, 5);
     int rd = extract32(insn, 0, 5);
-    int is_tblx = extract32(insn, 12, 1);
-    int len = extract32(insn, 13, 2);
-    TCGv_i64 tcg_resl, tcg_resh, tcg_idx;
-    TCGv_i32 tcg_regno, tcg_numregs;
+    int is_tbx = extract32(insn, 12, 1);
+    int len = (extract32(insn, 13, 2) + 1) * 16;
 
     if (op2 != 0) {
         unallocated_encoding(s);
@@ -7546,53 +7544,11 @@ static void disas_simd_tb(DisasContext *s, uint32_t insn)
         return;
     }
 
-    /* This does a table lookup: for every byte element in the input
-     * we index into a table formed from up to four vector registers,
-     * and then the output is the result of the lookups. Our helper
-     * function does the lookup operation for a single 64 bit part of
-     * the input.
-     */
-    tcg_resl = tcg_temp_new_i64();
-    tcg_resh = NULL;
-
-    if (is_tblx) {
-        read_vec_element(s, tcg_resl, rd, 0, MO_64);
-    } else {
-        tcg_gen_movi_i64(tcg_resl, 0);
-    }
-
-    if (is_q) {
-        tcg_resh = tcg_temp_new_i64();
-        if (is_tblx) {
-            read_vec_element(s, tcg_resh, rd, 1, MO_64);
-        } else {
-            tcg_gen_movi_i64(tcg_resh, 0);
-        }
-    }
-
-    tcg_idx = tcg_temp_new_i64();
-    tcg_regno = tcg_const_i32(rn);
-    tcg_numregs = tcg_const_i32(len + 1);
-    read_vec_element(s, tcg_idx, rm, 0, MO_64);
-    gen_helper_simd_tbl(tcg_resl, cpu_env, tcg_resl, tcg_idx,
-                        tcg_regno, tcg_numregs);
-    if (is_q) {
-        read_vec_element(s, tcg_idx, rm, 1, MO_64);
-        gen_helper_simd_tbl(tcg_resh, cpu_env, tcg_resh, tcg_idx,
-                            tcg_regno, tcg_numregs);
-    }
-    tcg_temp_free_i64(tcg_idx);
-    tcg_temp_free_i32(tcg_regno);
-    tcg_temp_free_i32(tcg_numregs);
-
-    write_vec_element(s, tcg_resl, rd, 0, MO_64);
-    tcg_temp_free_i64(tcg_resl);
-
-    if (is_q) {
-        write_vec_element(s, tcg_resh, rd, 1, MO_64);
-        tcg_temp_free_i64(tcg_resh);
-    }
-    clear_vec_high(s, is_q, rd);
+    tcg_gen_gvec_2_ptr(vec_full_reg_offset(s, rd),
+                       vec_full_reg_offset(s, rm), cpu_env,
+                       is_q ? 16 : 8, vec_full_reg_size(s),
+                       (len << 6) | (is_tbx << 5) | rn,
+                       gen_helper_simd_tblx);
 }
 
 /* ZIP/UZP/TRN