[PATCH v2 08/17] target/arm: Convert Neon VPMAX/VPMIN 3-reg-same insns to decodetree

[Peter Maydell]

Convert the Neon integer VPMAX and VPMIN 3-reg-same insns to
decodetree. These are 'pairwise' operations.

Signed-off-by: Peter Maydell <address@hidden>
Reviewed-by: Richard Henderson <address@hidden>
---
 target/arm/neon-dp.decode       |  9 +++++
 target/arm/translate-neon.inc.c | 71 +++++++++++++++++++++++++++++++++
 target/arm/translate.c          | 17 +-------
 3 files changed, 82 insertions(+), 15 deletions(-)

diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode
index e0da5a39fc2..03f39a0b44b 100644
--- a/target/arm/neon-dp.decode
+++ b/target/arm/neon-dp.decode
@@ -42,6 +42,9 @@
 @3same           .... ... . . . size:2 .... .... .... . q:1 . . .... \
                  &3same vm=%vm_dp vn=%vn_dp vd=%vd_dp
 
+@3same_q0        .... ... . . . size:2 .... .... .... . 0 . . .... \
+                 &3same vm=%vm_dp vn=%vn_dp vd=%vd_dp q=0
+
 VHADD_S_3s       1111 001 0 0 . .. .... .... 0000 . . . 0 .... @3same
 VHADD_U_3s       1111 001 1 0 . .. .... .... 0000 . . . 0 .... @3same
 VQADD_S_3s       1111 001 0 0 . .. .... .... 0000 . . . 1 .... @3same
@@ -144,6 +147,12 @@ VMLS_3s          1111 001 1 0 . .. .... .... 1001 . . . 0 
.... @3same
 VMUL_3s          1111 001 0 0 . .. .... .... 1001 . . . 1 .... @3same
 VMUL_p_3s        1111 001 1 0 . .. .... .... 1001 . . . 1 .... @3same
 
+VPMAX_S_3s       1111 001 0 0 . .. .... .... 1010 . . . 0 .... @3same_q0
+VPMAX_U_3s       1111 001 1 0 . .. .... .... 1010 . . . 0 .... @3same_q0
+
+VPMIN_S_3s       1111 001 0 0 . .. .... .... 1010 . . . 1 .... @3same_q0
+VPMIN_U_3s       1111 001 1 0 . .. .... .... 1010 . . . 1 .... @3same_q0
+
 VQRDMLAH_3s      1111 001 1 0 . .. .... .... 1011 ... 1 .... @3same
 
 SHA1_3s          1111 001 0 0 . optype:2 .... .... 1100 . 1 . 0 .... \
diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c
index 7097c18f334..7db6b856598 100644
--- a/target/arm/translate-neon.inc.c
+++ b/target/arm/translate-neon.inc.c
@@ -924,3 +924,74 @@ DO_3SAME_32_ENV(VQSHL_S, qshl_s)
 DO_3SAME_32_ENV(VQSHL_U, qshl_u)
 DO_3SAME_32_ENV(VQRSHL_S, qrshl_s)
 DO_3SAME_32_ENV(VQRSHL_U, qrshl_u)
+
+static bool do_3same_pair(DisasContext *s, arg_3same *a, NeonGenTwoOpFn *fn)
+{
+    /* Operations handled pairwise 32 bits at a time */
+    TCGv_i32 tmp, tmp2, tmp3;
+
+    if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
+        return false;
+    }
+
+    /* UNDEF accesses to D16-D31 if they don't exist. */
+    if (!dc_isar_feature(aa32_simd_r32, s) &&
+        ((a->vd | a->vn | a->vm) & 0x10)) {
+        return false;
+    }
+
+    if (a->size == 3) {
+        return false;
+    }
+
+    if (!vfp_access_check(s)) {
+        return true;
+    }
+
+    assert(a->q == 0); /* enforced by decode patterns */
+
+    /*
+     * Note that we have to be careful not to clobber the source operands
+     * in the "vm == vd" case by storing the result of the first pass too
+     * early. Since Q is 0 there are always just two passes, so instead
+     * of a complicated loop over each pass we just unroll.
+     */
+    tmp = neon_load_reg(a->vn, 0);
+    tmp2 = neon_load_reg(a->vn, 1);
+    fn(tmp, tmp, tmp2);
+    tcg_temp_free_i32(tmp2);
+
+    tmp3 = neon_load_reg(a->vm, 0);
+    tmp2 = neon_load_reg(a->vm, 1);
+    fn(tmp3, tmp3, tmp2);
+    tcg_temp_free_i32(tmp2);
+
+    neon_store_reg(a->vd, 0, tmp);
+    neon_store_reg(a->vd, 1, tmp3);
+    return true;
+}
+
+#define DO_3SAME_PAIR(INSN, func)                                       \
+    static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a)        \
+    {                                                                   \
+        static NeonGenTwoOpFn * const fns[] = {                         \
+            gen_helper_neon_##func##8,                                  \
+            gen_helper_neon_##func##16,                                 \
+            gen_helper_neon_##func##32,                                 \
+        };                                                              \
+        if (a->size > 2) {                                              \
+            return false;                                               \
+        }                                                               \
+        return do_3same_pair(s, a, fns[a->size]);                       \
+    }
+
+/* 32-bit pairwise ops end up the same as the elementwise versions.  */
+#define gen_helper_neon_pmax_s32  tcg_gen_smax_i32
+#define gen_helper_neon_pmax_u32  tcg_gen_umax_i32
+#define gen_helper_neon_pmin_s32  tcg_gen_smin_i32
+#define gen_helper_neon_pmin_u32  tcg_gen_umin_i32
+
+DO_3SAME_PAIR(VPMAX_S, pmax_s)
+DO_3SAME_PAIR(VPMIN_S, pmin_s)
+DO_3SAME_PAIR(VPMAX_U, pmax_u)
+DO_3SAME_PAIR(VPMIN_U, pmin_u)
diff --git a/target/arm/translate.c b/target/arm/translate.c
index 3aabb18720b..82be4d40282 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -3011,12 +3011,6 @@ static inline void gen_neon_rsb(int size, TCGv_i32 t0, 
TCGv_i32 t1)
     }
 }
 
-/* 32-bit pairwise ops end up the same as the elementwise versions.  */
-#define gen_helper_neon_pmax_s32  tcg_gen_smax_i32
-#define gen_helper_neon_pmax_u32  tcg_gen_umax_i32
-#define gen_helper_neon_pmin_s32  tcg_gen_smin_i32
-#define gen_helper_neon_pmin_u32  tcg_gen_umin_i32
-
 #define GEN_NEON_INTEGER_OP_ENV(name) do { \
     switch ((size << 1) | u) { \
     case 0: \
@@ -5442,6 +5436,8 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t 
insn)
         case NEON_3R_VQSHL:
         case NEON_3R_VRSHL:
         case NEON_3R_VQRSHL:
+        case NEON_3R_VPMAX:
+        case NEON_3R_VPMIN:
             /* Already handled by decodetree */
             return 1;
         }
@@ -5453,8 +5449,6 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t 
insn)
         pairwise = 0;
         switch (op) {
         case NEON_3R_VPADD_VQRDMLAH:
-        case NEON_3R_VPMAX:
-        case NEON_3R_VPMIN:
             pairwise = 1;
             break;
         case NEON_3R_FLOAT_ARITH:
@@ -5511,13 +5505,6 @@ static int disas_neon_data_insn(DisasContext *s, 
uint32_t insn)
             tmp2 = neon_load_reg(rm, pass);
         }
         switch (op) {
-            break;
-        case NEON_3R_VPMAX:
-            GEN_NEON_INTEGER_OP(pmax);
-            break;
-        case NEON_3R_VPMIN:
-            GEN_NEON_INTEGER_OP(pmin);
-            break;
         case NEON_3R_VQDMULH_VQRDMULH: /* Multiply high.  */
             if (!u) { /* VQDMULH */
                 switch (size) {
-- 
2.20.1