target/arm: Convert the VSEL instructions to decodetree

Convert the VSEL instructions to decodetree.
We leave trans_VSEL() in translate.c for now as this allows
the patch to show just the changes from the old handle_vsel().

In the old code the check for "do D16-D31 exist" was hidden in
the VFP_DREG macro, and assumed that VFPv3 always implied that
D16-D31 exist. In the new code we do the correct ID register test.
This gives identical behaviour for most of our CPUs, and fixes
previously incorrect handling for Cortex-R5F, Cortex-M4 and
Cortex-M33, which all implement VFPv3 or better with only 16
double-precision registers.

Backports commit b3ff4b87b4ae08120a51fe12592725e1dca8a085 from qemu

qemu/target/arm/cpu.h

@@ -3328,6 +3328,12 @@ static inline bool isar_feature_aa32_fp16_arith(const ARMISARegisters *id)
     return FIELD_EX64(id->id_aa64pfr0, ID_AA64PFR0, FP) == 1;
 }
 
+static inline bool isar_feature_aa32_fp_d32(const ARMISARegisters *id)
+{
+    /* Return true if D16-D31 are implemented */
+    return FIELD_EX64(id->mvfr0, MVFR0, SIMDREG) >= 2;
+}
+
 /*
  * We always set the FP and SIMD FP16 fields to indicate identical
  * levels of support (assuming SIMD is implemented at all), so

qemu/target/arm/translate-vfp.inc.c

@@ -131,3 +131,12 @@ static bool full_vfp_access_check(DisasContext *s, bool ignore_vfp_enabled)
 
     return true;
 }
+
+/*
+ * The most usual kind of VFP access check, for everything except
+ * FMXR/FMRX to the always-available special registers.
+ */
+static bool vfp_access_check(DisasContext *s)
+{
+    return full_vfp_access_check(s, false);
+}

qemu/target/arm/translate.c

@@ -3177,11 +3177,28 @@ static void gen_neon_dup_high16(DisasContext *s, TCGv_i32 var)
     tcg_temp_free_i32(tcg_ctx, tmp);
 }
 
-static int handle_vsel(DisasContext *s, uint32_t insn, uint32_t rd, uint32_t rn, uint32_t rm,
-                       uint32_t dp)
+static bool trans_VSEL(DisasContext *s, arg_VSEL *a)
 {
     TCGContext *tcg_ctx = s->uc->tcg_ctx;
-    uint32_t cc = extract32(insn, 20, 2);
+    uint32_t rd, rn, rm;
+    bool dp = a->dp;
+
+    if (!dc_isar_feature(aa32_vsel, s)) {
+        return false;
+    }
+
+    /* UNDEF accesses to D16-D31 if they don't exist */
+    if (dp && !dc_isar_feature(aa32_fp_d32, s) &&
+        ((a->vm | a->vn | a->vd) & 0x10)) {
+        return false;
+    }
+    rd = a->vd;
+    rn = a->vn;
+    rm = a->vm;
+
+    if (!vfp_access_check(s)) {
+        return true;
+    }
 
     if (dp) {
         TCGv_i64 frn, frm, dest;
@@ -3203,7 +3220,7 @@ static int handle_vsel(DisasContext *s, uint32_t insn, uint32_t rd, uint32_t rn,
 
         tcg_gen_ld_f64(tcg_ctx, frn, tcg_ctx->cpu_env, vfp_reg_offset(dp, rn));
         tcg_gen_ld_f64(tcg_ctx, frm, tcg_ctx->cpu_env, vfp_reg_offset(dp, rm));
-        switch (cc) {
+        switch (a->cc) {
         case 0: /* eq: Z */
             tcg_gen_movcond_i64(tcg_ctx, TCG_COND_EQ, dest, zf, zero,
                                 frn, frm);
@@ -3250,7 +3267,7 @@ static int handle_vsel(DisasContext *s, uint32_t insn, uint32_t rd, uint32_t rn,
         dest = tcg_temp_new_i32(tcg_ctx);
         tcg_gen_ld_f32(tcg_ctx, frn, tcg_ctx->cpu_env, vfp_reg_offset(dp, rn));
         tcg_gen_ld_f32(tcg_ctx, frm, tcg_ctx->cpu_env, vfp_reg_offset(dp, rm));
-        switch (cc) {
+        switch (a->cc) {
         case 0: /* eq: Z */
             tcg_gen_movcond_i32(tcg_ctx, TCG_COND_EQ, dest, tcg_ctx->cpu_ZF, zero,
                                 frn, frm);
@@ -3284,7 +3301,7 @@ static int handle_vsel(DisasContext *s, uint32_t insn, uint32_t rd, uint32_t rn,
         tcg_temp_free_i32(tcg_ctx, zero);
     }
 
-    return 0;
+    return true;
 }
 
 static int handle_vminmaxnm(DisasContext *s, uint32_t insn, uint32_t rd, uint32_t rn,
@@ -3459,10 +3476,8 @@ static int disas_vfp_misc_insn(DisasContext *s, uint32_t insn)
         rm = VFP_SREG_M(insn);
     }
 
-    if ((insn & 0x0f800e50) == 0x0e000a00 && dc_isar_feature(aa32_vsel, s)) {
-        return handle_vsel(s, insn, rd, rn, rm, dp);
-    } else if ((insn & 0x0fb00e10) == 0x0e800a00 &&
-               dc_isar_feature(aa32_vminmaxnm, s)) {
+    if ((insn & 0x0fb00e10) == 0x0e800a00 &&
+        dc_isar_feature(aa32_vminmaxnm, s)) {
         return handle_vminmaxnm(s, insn, rd, rn, rm, dp);
     } else if ((insn & 0x0fbc0ed0) == 0x0eb80a40 &&
                dc_isar_feature(aa32_vrint, s)) {

qemu/target/arm/vfp-uncond.decode

@@ -26,3 +26,22 @@
 #  1111 1110 .... .... .... 101. .... ....
 # (but those patterns might also cover some Neon instructions,
 # which do not live in this file.)
+
+# VFP registers have an odd encoding with a four-bit field
+# and a one-bit field which are assembled in different orders
+# depending on whether the register is double or single precision.
+# Each individual instruction function must do the checks for
+# "double register selected but CPU does not have double support"
+# and "double register number has bit 4 set but CPU does not
+# support D16-D31" (which should UNDEF).
+%vm_dp  5:1 0:4
+%vm_sp  0:4 5:1
+%vn_dp  7:1 16:4
+%vn_sp  16:4 7:1
+%vd_dp  22:1 12:4
+%vd_sp  12:4 22:1
+
+VSEL        1111 1110 0. cc:2 .... .... 1010 .0.0 .... \
+            vm=%vm_sp vn=%vn_sp vd=%vd_sp dp=0
+VSEL        1111 1110 0. cc:2 .... .... 1011 .0.0 .... \
+            vm=%vm_dp vn=%vn_dp vd=%vd_dp dp=1