target/arm: Convert Neon VPMIN/VPMAX/VPADD float 3-reg-same insns to decodetree

Convert the Neon float VPMIN, VPMAX and VPADD 3-reg-same insns to
decodetree. These are the only remaining 'pairwise' operations,
so we can delete the pairwise-specific bits of the old decoder's
for-each-element loop now.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20200512163904.10918-13-peter.maydell@linaro.org

3 files changed, 76 insertions, 55 deletions

diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode
index 0dd02f3..d66c67c 100644
--- a/target/arm/neon-dp.decode
+++ b/target/arm/neon-dp.decode
@@ -48,6 +48,8 @@
 # For FP insns the high bit of 'size' is used as part of opcode decode
 @3same_fp        .... ... . . . . size:1 .... .... .... . q:1 . . .... \
                  &3same vm=%vm_dp vn=%vn_dp vd=%vd_dp
+@3same_fp_q0     .... ... . . . . size:1 .... .... .... . 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
@@ -176,4 +178,7 @@ VQRDMLSH_3s      1111 001 1 0 . .. .... .... 1100 ... 1 .... @3same
 
 VADD_fp_3s       1111 001 0 0 . 0 . .... .... 1101 ... 0 .... @3same_fp
 VSUB_fp_3s       1111 001 0 0 . 1 . .... .... 1101 ... 0 .... @3same_fp
+VPADD_fp_3s      1111 001 1 0 . 0 . .... .... 1101 ... 0 .... @3same_fp_q0
 VABD_fp_3s       1111 001 1 0 . 1 . .... .... 1101 ... 0 .... @3same_fp
+VPMAX_fp_3s      1111 001 1 0 . 0 . .... .... 1111 ... 0 .... @3same_fp_q0
+VPMIN_fp_3s      1111 001 1 0 . 1 . .... .... 1111 ... 0 .... @3same_fp_q0
diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c
index 540720f..7bdf1e3 100644
--- a/target/arm/translate-neon.inc.c
+++ b/target/arm/translate-neon.inc.c
@@ -1049,3 +1049,66 @@ DO_3SAME_VQDMULH(VQRDMULH, qrdmulh)
 DO_3S_FP_GVEC(VADD, gen_helper_gvec_fadd_s)
 DO_3S_FP_GVEC(VSUB, gen_helper_gvec_fsub_s)
 DO_3S_FP_GVEC(VABD, gen_helper_gvec_fabd_s)
+
+static bool do_3same_fp_pair(DisasContext *s, arg_3same *a, VFPGen3OpSPFn *fn)
+{
+    /* FP operations handled pairwise 32 bits at a time */
+    TCGv_i32 tmp, tmp2, tmp3;
+    TCGv_ptr fpstatus;
+
+    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 (!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.
+     */
+    fpstatus = get_fpstatus_ptr(1);
+    tmp = neon_load_reg(a->vn, 0);
+    tmp2 = neon_load_reg(a->vn, 1);
+    fn(tmp, tmp, tmp2, fpstatus);
+    tcg_temp_free_i32(tmp2);
+
+    tmp3 = neon_load_reg(a->vm, 0);
+    tmp2 = neon_load_reg(a->vm, 1);
+    fn(tmp3, tmp3, tmp2, fpstatus);
+    tcg_temp_free_i32(tmp2);
+    tcg_temp_free_ptr(fpstatus);
+
+    neon_store_reg(a->vd, 0, tmp);
+    neon_store_reg(a->vd, 1, tmp3);
+    return true;
+}
+
+/*
+ * For all the functions using this macro, size == 1 means fp16,
+ * which is an architecture extension we don't implement yet.
+ */
+#define DO_3S_FP_PAIR(INSN,FUNC)                                    \
+    static bool trans_##INSN##_fp_3s(DisasContext *s, arg_3same *a) \
+    {                                                               \
+        if (a->size != 0) {                                         \
+            /* TODO fp16 support */                                 \
+            return false;                                           \
+        }                                                           \
+        return do_3same_fp_pair(s, a, FUNC);                        \
+    }
+
+DO_3S_FP_PAIR(VPADD, gen_helper_vfp_adds)
+DO_3S_FP_PAIR(VPMAX, gen_helper_vfp_maxs)
+DO_3S_FP_PAIR(VPMIN, gen_helper_vfp_mins)
diff --git a/target/arm/translate.c b/target/arm/translate.c
index 8a94856..ca6ed09 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -5348,7 +5348,6 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
     int shift;
     int pass;
     int count;
-    int pairwise;
     int u;
     int vec_size;
     uint32_t imm;
@@ -5433,6 +5432,7 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
         case NEON_3R_VPMIN:
         case NEON_3R_VPADD_VQRDMLAH:
         case NEON_3R_VQDMULH_VQRDMULH:
+        case NEON_3R_FLOAT_ARITH:
             /* Already handled by decodetree */
             return 1;
         }
@@ -5441,16 +5441,11 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
             /* 64-bit element instructions: handled by decodetree */
             return 1;
         }
-        pairwise = 0;
         switch (op) {
-        case NEON_3R_FLOAT_ARITH:
-            pairwise = (u && size < 2); /* if VPADD (float) */
-            if (!pairwise) {
-                return 1; /* handled by decodetree */
-            }
-            break;
         case NEON_3R_FLOAT_MINMAX:
-            pairwise = u; /* if VPMIN/VPMAX (float) */
+            if (u) {
+                return 1; /* VPMIN/VPMAX handled by decodetree */
+            }
             break;
         case NEON_3R_FLOAT_CMP:
             if (!u && size) {
@@ -5478,41 +5473,12 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
             break;
         }
 
-        if (pairwise && q) {
-            /* All the pairwise insns UNDEF if Q is set */
-            return 1;
-        }
-
         for (pass = 0; pass < (q ? 4 : 2); pass++) {
 
-        if (pairwise) {
-            /* Pairwise.  */
-            if (pass < 1) {
-                tmp = neon_load_reg(rn, 0);
-                tmp2 = neon_load_reg(rn, 1);
-            } else {
-                tmp = neon_load_reg(rm, 0);
-                tmp2 = neon_load_reg(rm, 1);
-            }
-        } else {
-            /* Elementwise.  */
-            tmp = neon_load_reg(rn, pass);
-            tmp2 = neon_load_reg(rm, pass);
-        }
+        /* Elementwise.  */
+        tmp = neon_load_reg(rn, pass);
+        tmp2 = neon_load_reg(rm, pass);
         switch (op) {
-        case NEON_3R_FLOAT_ARITH: /* Floating point arithmetic. */
-        {
-            TCGv_ptr fpstatus = get_fpstatus_ptr(1);
-            switch ((u << 2) | size) {
-            case 4: /* VPADD */
-                gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
-                break;
-            default:
-                abort();
-            }
-            tcg_temp_free_ptr(fpstatus);
-            break;
-        }
         case NEON_3R_FLOAT_MULTIPLY:
         {
             TCGv_ptr fpstatus = get_fpstatus_ptr(1);
@@ -5603,22 +5569,9 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
         }
         tcg_temp_free_i32(tmp2);
 
-        /* Save the result.  For elementwise operations we can put it
-           straight into the destination register.  For pairwise operations
-           we have to be careful to avoid clobbering the source operands.  */
-        if (pairwise && rd == rm) {
-            neon_store_scratch(pass, tmp);
-        } else {
-            neon_store_reg(rd, pass, tmp);
-        }
+        neon_store_reg(rd, pass, tmp);
 
         } /* for pass */
-        if (pairwise && rd == rm) {
-            for (pass = 0; pass < (q ? 4 : 2); pass++) {
-                tmp = neon_load_scratch(pass);
-                neon_store_reg(rd, pass, tmp);
-            }
-        }
         /* End of 3 register same size operations.  */
     } else if (insn & (1 << 4)) {
         if ((insn & 0x00380080) != 0) {