target-arm: A64: Implement PMULL instruction

Implement the PMULL instruction; this is the last unimplemented insn
in the three-reg-diff group.

Note that PMULL with size 3 is considered part of the AES part
of the crypto extensions (see the ID_AA64ISAR0_EL1 register definition
in the v8 ARM ARM), so it isn't necessary to burn an extra feature
bit on it, even though we're using more feature bits than a single
"crypto extension present/not present" toggle.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Message-id: 1394822294-14837-2-git-send-email-peter.maydell@linaro.org

target-arm/helper-a64.c

@@ -180,6 +180,36 @@ uint64_t HELPER(simd_tbl)(CPUARMState *env, uint64_t result, uint64_t indices,
     return result;
 }
 
+/* Helper function for 64 bit polynomial multiply case:
+ * perform PolynomialMult(op1, op2) and return either the top or
+ * bottom half of the 128 bit result.
+ */
+uint64_t HELPER(neon_pmull_64_lo)(uint64_t op1, uint64_t op2)
+{
+    int bitnum;
+    uint64_t res = 0;
+
+    for (bitnum = 0; bitnum < 64; bitnum++) {
+        if (op1 & (1ULL << bitnum)) {
+            res ^= op2 << bitnum;
+        }
+    }
+    return res;
+}
+uint64_t HELPER(neon_pmull_64_hi)(uint64_t op1, uint64_t op2)
+{
+    int bitnum;
+    uint64_t res = 0;
+
+    /* bit 0 of op1 can't influence the high 64 bits at all */
+    for (bitnum = 1; bitnum < 64; bitnum++) {
+        if (op1 & (1ULL << bitnum)) {
+            res ^= op2 >> (64 - bitnum);
+        }
+    }
+    return res;
+}
+
 /* 64bit/double versions of the neon float compare functions */
 uint64_t HELPER(neon_ceq_f64)(float64 a, float64 b, void *fpstp)
 {

target-arm/helper-a64.h

@@ -27,6 +27,8 @@ DEF_HELPER_3(vfp_cmpes_a64, i64, f32, f32, ptr)
 DEF_HELPER_3(vfp_cmpd_a64, i64, f64, f64, ptr)
 DEF_HELPER_3(vfp_cmped_a64, i64, f64, f64, ptr)
 DEF_HELPER_FLAGS_5(simd_tbl, TCG_CALL_NO_RWG_SE, i64, env, i64, i64, i32, i32)
+DEF_HELPER_FLAGS_2(neon_pmull_64_lo, TCG_CALL_NO_RWG_SE, i64, i64, i64)
+DEF_HELPER_FLAGS_2(neon_pmull_64_hi, TCG_CALL_NO_RWG_SE, i64, i64, i64)
 DEF_HELPER_FLAGS_3(vfp_mulxs, TCG_CALL_NO_RWG, f32, f32, f32, ptr)
 DEF_HELPER_FLAGS_3(vfp_mulxd, TCG_CALL_NO_RWG, f64, f64, f64, ptr)
 DEF_HELPER_FLAGS_3(neon_ceq_f64, TCG_CALL_NO_RWG, i64, i64, i64, ptr)

target-arm/translate-a64.c

@@ -7124,6 +7124,10 @@ static void handle_3rd_widening(DisasContext *s, int is_q, int is_u, int size,
                 gen_helper_neon_addl_saturate_s32(tcg_passres, cpu_env,
                                                   tcg_passres, tcg_passres);
                 break;
+            case 14: /* PMULL */
+                assert(size == 0);
+                gen_helper_neon_mull_p8(tcg_passres, tcg_op1, tcg_op2);
+                break;
             default:
                 g_assert_not_reached();
             }
@@ -7243,6 +7247,30 @@ static void handle_3rd_narrowing(DisasContext *s, int is_q, int is_u, int size,
     }
 }
 
+static void handle_pmull_64(DisasContext *s, int is_q, int rd, int rn, int rm)
+{
+    /* PMULL of 64 x 64 -> 128 is an odd special case because it
+     * is the only three-reg-diff instruction which produces a
+     * 128-bit wide result from a single operation. However since
+     * it's possible to calculate the two halves more or less
+     * separately we just use two helper calls.
+     */
+    TCGv_i64 tcg_op1 = tcg_temp_new_i64();
+    TCGv_i64 tcg_op2 = tcg_temp_new_i64();
+    TCGv_i64 tcg_res = tcg_temp_new_i64();
+
+    read_vec_element(s, tcg_op1, rn, is_q, MO_64);
+    read_vec_element(s, tcg_op2, rm, is_q, MO_64);
+    gen_helper_neon_pmull_64_lo(tcg_res, tcg_op1, tcg_op2);
+    write_vec_element(s, tcg_res, rd, 0, MO_64);
+    gen_helper_neon_pmull_64_hi(tcg_res, tcg_op1, tcg_op2);
+    write_vec_element(s, tcg_res, rd, 1, MO_64);
+
+    tcg_temp_free_i64(tcg_op1);
+    tcg_temp_free_i64(tcg_op2);
+    tcg_temp_free_i64(tcg_res);
+}
+
 /* C3.6.15 AdvSIMD three different
  *   31  30  29 28       24 23  22  21 20  16 15    12 11 10 9    5 4    0
  * +---+---+---+-----------+------+---+------+--------+-----+------+------+
@@ -7293,8 +7321,15 @@ static void disas_simd_three_reg_diff(DisasContext *s, uint32_t insn)
             unallocated_encoding(s);
             return;
         }
-        unsupported_encoding(s, insn);
+        if (size == 3) {
-        break;
+            if (!arm_dc_feature(s, ARM_FEATURE_V8_AES)) {
+                unallocated_encoding(s);
+                return;
+            }
+            handle_pmull_64(s, is_q, rd, rn, rm);
+            return;
+        }
+        goto is_widening;
     case 9: /* SQDMLAL, SQDMLAL2 */
     case 11: /* SQDMLSL, SQDMLSL2 */
     case 13: /* SQDMULL, SQDMULL2 */
@@ -7315,6 +7350,7 @@ static void disas_simd_three_reg_diff(DisasContext *s, uint32_t insn)
             unallocated_encoding(s);
             return;
         }
+    is_widening:
         handle_3rd_widening(s, is_q, is_u, size, opcode, rd, rn, rm);
         break;
     default:
@@ -9045,6 +9081,7 @@ void gen_intermediate_code_internal_a64(ARMCPU *cpu,
     dc->vec_stride = 0;
     dc->cp_regs = cpu->cp_regs;
     dc->current_pl = arm_current_pl(env);
+    dc->features = env->features;
 
     init_tmp_a64_array(dc);
 

target-arm/translate.c

@@ -10654,6 +10654,7 @@ static inline void gen_intermediate_code_internal(ARMCPU *cpu,
     dc->vec_stride = ARM_TBFLAG_VECSTRIDE(tb->flags);
     dc->cp_regs = cpu->cp_regs;
     dc->current_pl = arm_current_pl(env);
+    dc->features = env->features;
 
     cpu_F0s = tcg_temp_new_i32();
     cpu_F1s = tcg_temp_new_i32();

target-arm/translate.h

@@ -26,6 +26,7 @@ typedef struct DisasContext {
     int aarch64;
     int current_pl;
     GHashTable *cp_regs;
+    uint64_t features; /* CPU features bits */
 #define TMP_A64_MAX 16
     int tmp_a64_count;
     TCGv_i64 tmp_a64[TMP_A64_MAX];
@@ -33,6 +34,11 @@ typedef struct DisasContext {
 
 extern TCGv_ptr cpu_env;
 
+static inline int arm_dc_feature(DisasContext *dc, int feature)
+{
+    return (dc->features & (1ULL << feature)) != 0;
+}
+
 /* target-specific extra values for is_jmp */
 /* These instructions trap after executing, so the A32/T32 decoder must
  * defer them until after the conditional execution state has been updated.