[Qemu-devel] [PATCH v2 29/42] target/arm: Convert VABS to decodetree

[Peter Maydell <peter.maydell@linaro.org>]

Convert the VFP VABS instruction to decodetree.

Unlike the 3-op versions, we don't pass fpst to the VFPGen2OpSPFn or
VFPGen2OpDPFn because none of the operations which use this format
and support short vectors will need it.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/translate-vfp.inc.c | 167 +++++++++++++++++++++++++++++++++
 target/arm/translate.c         |  12 ++-
 target/arm/vfp.decode          |   5 +
 3 files changed, 180 insertions(+), 4 deletions(-)

diff --git a/target/arm/translate-vfp.inc.c b/target/arm/translate-vfp.inc.c
index a2eeb6cb511..d0282f1f921 100644
--- a/target/arm/translate-vfp.inc.c
+++ b/target/arm/translate-vfp.inc.c
@@ -1111,6 +1111,14 @@ typedef void VFPGen3OpSPFn(TCGv_i32 vd,
 typedef void VFPGen3OpDPFn(TCGv_i64 vd,
                            TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst);
 
+/*
+ * Types for callbacks for do_vfp_2op_sp() and do_vfp_2op_dp().
+ * The callback should emit code to write a value to vd (which
+ * should be written to only).
+ */
+typedef void VFPGen2OpSPFn(TCGv_i32 vd, TCGv_i32 vm);
+typedef void VFPGen2OpDPFn(TCGv_i64 vd, TCGv_i64 vm);
+
 /*
  * Perform a 3-operand VFP data processing instruction. fn is the
  * callback to do the actual operation; this function deals with the
@@ -1274,6 +1282,155 @@ static bool do_vfp_3op_dp(DisasContext *s, VFPGen3OpDPFn *fn,
     return true;
 }
 
+static bool do_vfp_2op_sp(DisasContext *s, VFPGen2OpSPFn *fn, int vd, int vm)
+{
+    uint32_t delta_m = 0;
+    uint32_t delta_d = 0;
+    uint32_t bank_mask = 0;
+    int veclen = s->vec_len;
+    TCGv_i32 f0, fd;
+
+    if (!dc_isar_feature(aa32_fpshvec, s) &&
+        (veclen != 0 || s->vec_stride != 0)) {
+        return false;
+    }
+
+    if (!vfp_access_check(s)) {
+        return true;
+    }
+
+    if (veclen > 0) {
+        bank_mask = 0x18;
+
+        /* Figure out what type of vector operation this is.  */
+        if ((vd & bank_mask) == 0) {
+            /* scalar */
+            veclen = 0;
+        } else {
+            delta_d = s->vec_stride + 1;
+
+            if ((vm & bank_mask) == 0) {
+                /* mixed scalar/vector */
+                delta_m = 0;
+            } else {
+                /* vector */
+                delta_m = delta_d;
+            }
+        }
+    }
+
+    f0 = tcg_temp_new_i32();
+    fd = tcg_temp_new_i32();
+
+    neon_load_reg32(f0, vm);
+
+    for (;;) {
+        fn(fd, f0);
+        neon_store_reg32(fd, vd);
+
+        if (veclen == 0) {
+            break;
+        }
+
+        if (delta_m == 0) {
+            /* single source one-many */
+            while (veclen--) {
+                vd = ((vd + delta_d) & (bank_mask - 1)) | (vd & bank_mask);
+                neon_store_reg32(fd, vd);
+            }
+            break;
+        }
+
+        /* Set up the operands for the next iteration */
+        veclen--;
+        vd = ((vd + delta_d) & (bank_mask - 1)) | (vd & bank_mask);
+        vm = ((vm + delta_m) & (bank_mask - 1)) | (vm & bank_mask);
+        neon_load_reg32(f0, vm);
+    }
+
+    tcg_temp_free_i32(f0);
+    tcg_temp_free_i32(fd);
+
+    return true;
+}
+
+static bool do_vfp_2op_dp(DisasContext *s, VFPGen2OpDPFn *fn, int vd, int vm)
+{
+    uint32_t delta_m = 0;
+    uint32_t delta_d = 0;
+    uint32_t bank_mask = 0;
+    int veclen = s->vec_len;
+    TCGv_i64 f0, fd;
+
+    /* UNDEF accesses to D16-D31 if they don't exist */
+    if (!dc_isar_feature(aa32_fp_d32, s) && ((vd | vm) & 0x10)) {
+        return false;
+    }
+
+    if (!dc_isar_feature(aa32_fpshvec, s) &&
+        (veclen != 0 || s->vec_stride != 0)) {
+        return false;
+    }
+
+    if (!vfp_access_check(s)) {
+        return true;
+    }
+
+    if (veclen > 0) {
+        bank_mask = 0xc;
+
+        /* Figure out what type of vector operation this is.  */
+        if ((vd & bank_mask) == 0) {
+            /* scalar */
+            veclen = 0;
+        } else {
+            delta_d = (s->vec_stride >> 1) + 1;
+
+            if ((vm & bank_mask) == 0) {
+                /* mixed scalar/vector */
+                delta_m = 0;
+            } else {
+                /* vector */
+                delta_m = delta_d;
+            }
+        }
+    }
+
+    f0 = tcg_temp_new_i64();
+    fd = tcg_temp_new_i64();
+
+    neon_load_reg64(f0, vm);
+
+    for (;;) {
+        fn(fd, f0);
+        neon_store_reg64(fd, vd);
+
+        if (veclen == 0) {
+            break;
+        }
+
+        if (delta_m == 0) {
+            /* single source one-many */
+            while (veclen--) {
+                vd = ((vd + delta_d) & (bank_mask - 1)) | (vd & bank_mask);
+                neon_store_reg64(fd, vd);
+            }
+            break;
+        }
+
+        /* Set up the operands for the next iteration */
+        veclen--;
+        vd = ((vd + delta_d) & (bank_mask - 1)) | (vd & bank_mask);
+        vm = ((vm + delta_m) & (bank_mask - 1)) | (vm & bank_mask);
+        neon_load_reg64(f0, vm);
+    }
+
+    tcg_temp_free_i64(f0);
+    tcg_temp_free_i64(fd);
+
+    return true;
+}
+
 static void gen_VMLA_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
 {
     /* Note that order of inputs to the add matters for NaNs */
@@ -1731,3 +1888,13 @@ static bool trans_VMOV_imm_dp(DisasContext *s, arg_VMOV_imm_dp *a)
     tcg_temp_free_i64(fd);
     return true;
 }
+
+static bool trans_VABS_sp(DisasContext *s, arg_VABS_sp *a)
+{
+    return do_vfp_2op_sp(s, gen_helper_vfp_abss, a->vd, a->vm);
+}
+
+static bool trans_VABS_dp(DisasContext *s, arg_VABS_dp *a)
+{
+    return do_vfp_2op_dp(s, gen_helper_vfp_absd, a->vd, a->vm);
+}
diff --git a/target/arm/translate.c b/target/arm/translate.c
index b0a12991131..dc9076a60a3 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -3096,6 +3096,14 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
             case 0 ... 14:
                 /* Already handled by decodetree */
                 return 1;
+            case 15:
+                switch (rn) {
+                case 1:
+                    /* Already handled by decodetree */
+                    return 1;
+                default:
+                    break;
+                }
             default:
                 break;
             }
@@ -3104,7 +3112,6 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                 /* rn is opcode, encoded as per VFP_SREG_N. */
                 switch (rn) {
                 case 0x00: /* vmov */
-                case 0x01: /* vabs */
                 case 0x02: /* vneg */
                 case 0x03: /* vsqrt */
                     break;
@@ -3284,9 +3291,6 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                     case 0: /* cpy */
                         /* no-op */
                         break;
-                    case 1: /* abs */
-                        gen_vfp_abs(dp);
-                        break;
                     case 2: /* neg */
                         gen_vfp_neg(dp);
                         break;
diff --git a/target/arm/vfp.decode b/target/arm/vfp.decode
index 1818d4f71e1..7035861c270 100644
--- a/target/arm/vfp.decode
+++ b/target/arm/vfp.decode
@@ -156,3 +156,8 @@ VMOV_imm_sp  ---- 1110 1.11 imm4h:4 .... 1010 0000 imm4l:4 \
              vd=%vd_sp
 VMOV_imm_dp  ---- 1110 1.11 imm4h:4 .... 1011 0000 imm4l:4 \
              vd=%vd_dp
+
+VABS_sp      ---- 1110 1.11 0000 .... 1010 11.0 .... \
+             vd=%vd_sp vm=%vm_sp
+VABS_dp      ---- 1110 1.11 0000 .... 1011 11.0 .... \
+             vd=%vd_dp vm=%vm_dp
-- 
2.20.1