[Qemu-arm] [PATCH 34/42] target/arm: Convert the VCVT-from-f16 insns to decodetree

[Peter Maydell]

Convert the VCVTT, VCVTB instructions that deal with conversion
from half-precision floats to f32 or 64 to decodetree.

Since we're no longer constrained to the old decoder's style
using cpu_F0s and cpu_F0d we can perform a direct 16 bit
load of the right half of the input single-precision register
rather than loading the full 32 bits and then doing a
separate shift or sign-extension.

Signed-off-by: Peter Maydell <address@hidden>
---
 target/arm/translate-vfp.inc.c | 82 ++++++++++++++++++++++++++++++++++
 target/arm/translate.c         | 56 +----------------------
 target/arm/vfp.decode          |  6 +++
 3 files changed, 89 insertions(+), 55 deletions(-)

diff --git a/target/arm/translate-vfp.inc.c b/target/arm/translate-vfp.inc.c
index 872b01621cc..3ad93cc48f4 100644
--- a/target/arm/translate-vfp.inc.c
+++ b/target/arm/translate-vfp.inc.c
@@ -30,6 +30,26 @@
 #include "decode-vfp.inc.c"
 #include "decode-vfp-uncond.inc.c"
 
+/*
+ * Return the offset of a 16-bit half of the specified VFP single-precision
+ * register. If top is true, returns the top 16 bits; otherwise the bottom
+ * 16 bits.
+ */
+static inline long vfp_f16_offset(unsigned reg, bool top)
+{
+    long offs = vfp_reg_offset(false, reg);
+#ifdef HOST_WORDS_BIGENDIAN
+    if (!top) {
+        offs += 2;
+    }
+#else
+    if (top) {
+        offs += 2;
+    }
+#endif
+    return offs;
+}
+
 /*
  * Check that VFP access is enabled. If it is, do the necessary
  * M-profile lazy-FP handling and then return true.
@@ -2005,3 +2025,65 @@ static bool trans_VCMP_dp(DisasContext *s, arg_VCMP_dp 
*a)
     }
     return true;
 }
+
+static bool trans_VCVT_f32_f16(DisasContext *s, arg_VCVT_f32_f16 *a)
+{
+    TCGv_ptr fpst;
+    TCGv_i32 ahp_mode;
+    TCGv_i32 tmp;
+
+    if (!dc_isar_feature(aa32_fp16_spconv, s)) {
+        return false;
+    }
+
+    if (!vfp_access_check(s)) {
+        return true;
+    }
+
+    fpst = get_fpstatus_ptr(false);
+    ahp_mode = get_ahp_flag();
+    tmp = tcg_temp_new_i32();
+    /* The T bit tells us if we want the low or high 16 bits of Vm */
+    tcg_gen_ld16u_i32(tmp, cpu_env, vfp_f16_offset(a->vm, a->t));
+    gen_helper_vfp_fcvt_f16_to_f32(tmp, tmp, fpst, ahp_mode);
+    neon_store_reg32(tmp, a->vd);
+    tcg_temp_free_i32(ahp_mode);
+    tcg_temp_free_ptr(fpst);
+    tcg_temp_free_i32(tmp);
+    return true;
+}
+
+static bool trans_VCVT_f64_f16(DisasContext *s, arg_VCVT_f64_f16 *a)
+{
+    TCGv_ptr fpst;
+    TCGv_i32 ahp_mode;
+    TCGv_i32 tmp;
+    TCGv_i64 vd;
+
+    if (!dc_isar_feature(aa32_fp16_dpconv, s)) {
+        return false;
+    }
+
+    /* UNDEF accesses to D16-D31 if they don't exist. */
+    if (!dc_isar_feature(aa32_fp_d32, s) && (a->vd  & 0x10)) {
+        return false;
+    }
+
+    if (!vfp_access_check(s)) {
+        return true;
+    }
+
+    fpst = get_fpstatus_ptr(false);
+    ahp_mode = get_ahp_flag();
+    tmp = tcg_temp_new_i32();
+    /* The T bit tells us if we want the low or high 16 bits of Vm */
+    tcg_gen_ld16u_i32(tmp, cpu_env, vfp_f16_offset(a->vm, a->t));
+    vd = tcg_temp_new_i64();
+    gen_helper_vfp_fcvt_f16_to_f64(vd, tmp, fpst, ahp_mode);
+    neon_store_reg64(vd, a->vd);
+    tcg_temp_free_i32(ahp_mode);
+    tcg_temp_free_ptr(fpst);
+    tcg_temp_free_i32(tmp);
+    tcg_temp_free_i64(vd);
+    return true;
+}
diff --git a/target/arm/translate.c b/target/arm/translate.c
index 26994e0d427..65e69873654 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -3066,7 +3066,7 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                 return 1;
             case 15:
                 switch (rn) {
-                case 0 ... 3:
+                case 0 ... 5:
                 case 8 ... 11:
                     /* Already handled by decodetree */
                     return 1;
@@ -3080,24 +3080,6 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
             if (op == 15) {
                 /* rn is opcode, encoded as per VFP_SREG_N. */
                 switch (rn) {
-                case 0x04: /* vcvtb.f64.f16, vcvtb.f32.f16 */
-                case 0x05: /* vcvtt.f64.f16, vcvtt.f32.f16 */
-                    /*
-                     * VCVTB, VCVTT: only present with the halfprec extension
-                     * UNPREDICTABLE if bit 8 is set prior to ARMv8
-                     * (we choose to UNDEF)
-                     */
-                    if (dp) {
-                        if (!dc_isar_feature(aa32_fp16_dpconv, s)) {
-                            return 1;
-                        }
-                    } else {
-                        if (!dc_isar_feature(aa32_fp16_spconv, s)) {
-                            return 1;
-                        }
-                    }
-                    rm_is_dp = false;
-                    break;
                 case 0x06: /* vcvtb.f16.f32, vcvtb.f16.f64 */
                 case 0x07: /* vcvtt.f16.f32, vcvtt.f16.f64 */
                     if (dp) {
@@ -3239,42 +3221,6 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                 switch (op) {
                 case 15: /* extension space */
                     switch (rn) {
-                    case 4: /* vcvtb.f32.f16, vcvtb.f64.f16 */
-                    {
-                        TCGv_ptr fpst = get_fpstatus_ptr(false);
-                        TCGv_i32 ahp_mode = get_ahp_flag();
-                        tmp = gen_vfp_mrs();
-                        tcg_gen_ext16u_i32(tmp, tmp);
-                        if (dp) {
-                            gen_helper_vfp_fcvt_f16_to_f64(cpu_F0d, tmp,
-                                                           fpst, ahp_mode);
-                        } else {
-                            gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp,
-                                                           fpst, ahp_mode);
-                        }
-                        tcg_temp_free_i32(ahp_mode);
-                        tcg_temp_free_ptr(fpst);
-                        tcg_temp_free_i32(tmp);
-                        break;
-                    }
-                    case 5: /* vcvtt.f32.f16, vcvtt.f64.f16 */
-                    {
-                        TCGv_ptr fpst = get_fpstatus_ptr(false);
-                        TCGv_i32 ahp = get_ahp_flag();
-                        tmp = gen_vfp_mrs();
-                        tcg_gen_shri_i32(tmp, tmp, 16);
-                        if (dp) {
-                            gen_helper_vfp_fcvt_f16_to_f64(cpu_F0d, tmp,
-                                                           fpst, ahp);
-                        } else {
-                            gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp,
-                                                           fpst, ahp);
-                        }
-                        tcg_temp_free_i32(tmp);
-                        tcg_temp_free_i32(ahp);
-                        tcg_temp_free_ptr(fpst);
-                        break;
-                    }
                     case 6: /* vcvtb.f16.f32, vcvtb.f16.f64 */
                     {
                         TCGv_ptr fpst = get_fpstatus_ptr(false);
diff --git a/target/arm/vfp.decode b/target/arm/vfp.decode
index 9db7aa7021a..53d9544f7cd 100644
--- a/target/arm/vfp.decode
+++ b/target/arm/vfp.decode
@@ -181,3 +181,9 @@ VCMP_sp      ---- 1110 1.11 010 z:1 .... 1010 e:1 1.0 .... \
              vd=%vd_sp vm=%vm_sp
 VCMP_dp      ---- 1110 1.11 010 z:1 .... 1011 e:1 1.0 .... \
              vd=%vd_dp vm=%vm_dp
+
+# VCVTT and VCVTB from f16: Vd format depends on size bit; Vm is always vm_sp
+VCVT_f32_f16 ---- 1110 1.11 0010 .... 1010 t:1 1.0 .... \
+             vd=%vd_sp vm=%vm_sp
+VCVT_f64_f16 ---- 1110 1.11 0010 .... 1011 t:1 1.0 .... \
+             vd=%vd_dp vm=%vm_sp
-- 
2.20.1