target/arm: Implement PSEL

This is an SVE instruction that operates using the SVE vector
length but that it is present only if SME is implemented.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20220708151540.18136-29-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

diff --git a/target/arm/sve.decode b/target/arm/sve.decode
index 95af08c..966803c 100644 (file)
--- a/target/arm/sve.decode
+++ b/target/arm/sve.decode
@@ -1674,3 +1674,23 @@ BFMLALT_zzxw    01100100 11 1 ..... 0100.1 ..... .....     @rrxr_3a esz=2
 
 ### SVE2 floating-point bfloat16 dot-product (indexed)
 BFDOT_zzxz      01100100 01 1 ..... 010000 ..... .....     @rrxr_2 esz=2
+
+### SVE broadcast predicate element
+
+&psel           esz pd pn pm rv imm
+%psel_rv        16:2 !function=plus_12
+%psel_imm_b     22:2 19:2
+%psel_imm_h     22:2 20:1
+%psel_imm_s     22:2
+%psel_imm_d     23:1
+@psel           ........ .. . ... .. .. pn:4 . pm:4 . pd:4  \
+                &psel rv=%psel_rv
+
+PSEL            00100101 .. 1 ..1 .. 01 .... 0 .... 0 ....  \
+                @psel esz=0 imm=%psel_imm_b
+PSEL            00100101 .. 1 .10 .. 01 .... 0 .... 0 ....  \
+                @psel esz=1 imm=%psel_imm_h
+PSEL            00100101 .. 1 100 .. 01 .... 0 .... 0 ....  \
+                @psel esz=2 imm=%psel_imm_s
+PSEL            00100101 .1 1 000 .. 01 .... 0 .... 0 ....  \
+                @psel esz=3 imm=%psel_imm_d

diff --git a/target/arm/translate-sve.c b/target/arm/translate-sve.c
index fd1a173..24ffb69 100644 (file)
--- a/target/arm/translate-sve.c
+++ b/target/arm/translate-sve.c
@@ -7419,3 +7419,60 @@ static bool do_BFMLAL_zzxw(DisasContext *s, arg_rrxr_esz *a, bool sel)
 
 TRANS_FEAT(BFMLALB_zzxw, aa64_sve_bf16, do_BFMLAL_zzxw, a, false)
 TRANS_FEAT(BFMLALT_zzxw, aa64_sve_bf16, do_BFMLAL_zzxw, a, true)
+
+static bool trans_PSEL(DisasContext *s, arg_psel *a)
+{
+    int vl = vec_full_reg_size(s);
+    int pl = pred_gvec_reg_size(s);
+    int elements = vl >> a->esz;
+    TCGv_i64 tmp, didx, dbit;
+    TCGv_ptr ptr;
+
+    if (!dc_isar_feature(aa64_sme, s)) {
+        return false;
+    }
+    if (!sve_access_check(s)) {
+        return true;
+    }
+
+    tmp = tcg_temp_new_i64();
+    dbit = tcg_temp_new_i64();
+    didx = tcg_temp_new_i64();
+    ptr = tcg_temp_new_ptr();
+
+    /* Compute the predicate element. */
+    tcg_gen_addi_i64(tmp, cpu_reg(s, a->rv), a->imm);
+    if (is_power_of_2(elements)) {
+        tcg_gen_andi_i64(tmp, tmp, elements - 1);
+    } else {
+        tcg_gen_remu_i64(tmp, tmp, tcg_constant_i64(elements));
+    }
+
+    /* Extract the predicate byte and bit indices. */
+    tcg_gen_shli_i64(tmp, tmp, a->esz);
+    tcg_gen_andi_i64(dbit, tmp, 7);
+    tcg_gen_shri_i64(didx, tmp, 3);
+    if (HOST_BIG_ENDIAN) {
+        tcg_gen_xori_i64(didx, didx, 7);
+    }
+
+    /* Load the predicate word. */
+    tcg_gen_trunc_i64_ptr(ptr, didx);
+    tcg_gen_add_ptr(ptr, ptr, cpu_env);
+    tcg_gen_ld8u_i64(tmp, ptr, pred_full_reg_offset(s, a->pm));
+
+    /* Extract the predicate bit and replicate to MO_64. */
+    tcg_gen_shr_i64(tmp, tmp, dbit);
+    tcg_gen_andi_i64(tmp, tmp, 1);
+    tcg_gen_neg_i64(tmp, tmp);
+
+    /* Apply to either copy the source, or write zeros. */
+    tcg_gen_gvec_ands(MO_64, pred_full_reg_offset(s, a->pd),
+                      pred_full_reg_offset(s, a->pn), tmp, pl, pl);
+
+    tcg_temp_free_i64(tmp);
+    tcg_temp_free_i64(dbit);
+    tcg_temp_free_i64(didx);
+    tcg_temp_free_ptr(ptr);
+    return true;
+}