target/arm: Implement MVE VADD (floating-point)

Implement the MVE VADD (floating-point) insn.  Handling of this is
similar to the 2-operand integer insns, except that we must take care
to only update the floating point exception status if the least
significant bit of the predicate mask for each element is active.

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

target/arm/helper-mve.h

@@ -410,6 +410,9 @@ DEF_HELPER_FLAGS_4(mve_vhcadd270b, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
 DEF_HELPER_FLAGS_4(mve_vhcadd270h, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
 DEF_HELPER_FLAGS_4(mve_vhcadd270w, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
 
+DEF_HELPER_FLAGS_4(mve_vfaddh, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
+DEF_HELPER_FLAGS_4(mve_vfadds, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
+
 DEF_HELPER_FLAGS_4(mve_vadd_scalarb, TCG_CALL_NO_WG, void, env, ptr, ptr, i32)
 DEF_HELPER_FLAGS_4(mve_vadd_scalarh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32)
 DEF_HELPER_FLAGS_4(mve_vadd_scalarw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32)

target/arm/mve.decode

@@ -26,6 +26,10 @@
 # VQDMULL has size in bit 28: 0 for 16 bit, 1 for 32 bit
 %size_28 28:1 !function=plus_1
 
+# 2 operand fp insns have size in bit 20: 1 for 16 bit, 0 for 32 bit,
+# like Neon FP insns.
+%2op_fp_size 20:1 !function=neon_3same_fp_size
+
 # 1imm format immediate
 %imm_28_16_0 28:1 16:3 0:4
 
@@ -118,6 +122,9 @@
 
 @vmaxv .... .... .... size:2 .. rda:4 .... .... .... &vmaxv qm=%qm
 
+@2op_fp .... .... .... .... .... .... .... .... &2op \
+        qd=%qd qn=%qn qm=%qm size=%2op_fp_size
+
 # Vector loads and stores
 
 # Widening loads and narrowing stores:
@@ -615,3 +622,6 @@ VCMPGE_scalar     1111 1110 0 . .. ... 1 ... 1 1111 0 1 0 0 .... @vcmp_scalar
 VCMPLT_scalar     1111 1110 0 . .. ... 1 ... 1 1111 1 1 0 0 .... @vcmp_scalar
 VCMPGT_scalar     1111 1110 0 . .. ... 1 ... 1 1111 0 1 1 0 .... @vcmp_scalar
 VCMPLE_scalar     1111 1110 0 . .. ... 1 ... 1 1111 1 1 1 0 .... @vcmp_scalar
+
+# 2-operand FP
+VADD_fp           1110 1111 0 . 0 . ... 0 ... 0 1101 . 1 . 0 ... 0 @2op_fp

target/arm/mve_helper.c

@@ -25,6 +25,7 @@
 #include "exec/cpu_ldst.h"
 #include "exec/exec-all.h"
 #include "tcg/tcg.h"
+#include "fpu/softfloat.h"
 
 static uint16_t mve_eci_mask(CPUARMState *env)
 {
@@ -2798,3 +2799,42 @@ DO_VMAXMINA(vmaxaw, 4, int32_t, uint32_t, DO_MAX)
 DO_VMAXMINA(vminab, 1, int8_t, uint8_t, DO_MIN)
 DO_VMAXMINA(vminah, 2, int16_t, uint16_t, DO_MIN)
 DO_VMAXMINA(vminaw, 4, int32_t, uint32_t, DO_MIN)
+
+/*
+ * 2-operand floating point. Note that if an element is partially
+ * predicated we must do the FP operation to update the non-predicated
+ * bytes, but we must be careful to avoid updating the FP exception
+ * state unless byte 0 of the element was unpredicated.
+ */
+#define DO_2OP_FP(OP, ESIZE, TYPE, FN)                                  \
+    void HELPER(glue(mve_, OP))(CPUARMState *env,                       \
+                                void *vd, void *vn, void *vm)           \
+    {                                                                   \
+        TYPE *d = vd, *n = vn, *m = vm;                                 \
+        TYPE r;                                                         \
+        uint16_t mask = mve_element_mask(env);                          \
+        unsigned e;                                                     \
+        float_status *fpst;                                             \
+        float_status scratch_fpst;                                      \
+        for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) {              \
+            if ((mask & MAKE_64BIT_MASK(0, ESIZE)) == 0) {              \
+                continue;                                               \
+            }                                                           \
+            fpst = (ESIZE == 2) ? &env->vfp.standard_fp_status_f16 :    \
+                &env->vfp.standard_fp_status;                           \
+            if (!(mask & 1)) {                                          \
+                /* We need the result but without updating flags */     \
+                scratch_fpst = *fpst;                                   \
+                fpst = &scratch_fpst;                                   \
+            }                                                           \
+            r = FN(n[H##ESIZE(e)], m[H##ESIZE(e)], fpst);               \
+            mergemask(&d[H##ESIZE(e)], r, mask);                        \
+        }                                                               \
+        mve_advance_vpt(env);                                           \
+    }
+
+#define DO_2OP_FP_ALL(OP, FN)                  \
+    DO_2OP_FP(OP##h, 2, float16, float16_##FN) \
+    DO_2OP_FP(OP##s, 4, float32, float32_##FN)
+
+DO_2OP_FP_ALL(vfadd, add)

target/arm/translate-mve.c

@@ -831,6 +831,23 @@ static bool trans_VSBCI(DisasContext *s, arg_2op *a)
     return do_2op(s, a, gen_helper_mve_vsbci);
 }
 
+#define DO_2OP_FP(INSN, FN)                                     \
+    static bool trans_##INSN(DisasContext *s, arg_2op *a)       \
+    {                                                           \
+        static MVEGenTwoOpFn * const fns[] = {                  \
+            NULL,                                               \
+            gen_helper_mve_##FN##h,                             \
+            gen_helper_mve_##FN##s,                             \
+            NULL,                                               \
+        };                                                      \
+        if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
+            return false;                                       \
+        }                                                       \
+        return do_2op(s, a, fns[a->size]);                      \
+    }
+
+DO_2OP_FP(VADD_fp, vfadd)
+
 static bool do_2op_scalar(DisasContext *s, arg_2scalar *a,
                           MVEGenTwoOpScalarFn fn)
 {

target/arm/translate-neon.c

@@ -28,12 +28,6 @@
 #include "translate.h"
 #include "translate-a32.h"
 
-static inline int neon_3same_fp_size(DisasContext *s, int x)
-{
-    /* Convert 0==fp32, 1==fp16 into a MO_* value */
-    return MO_32 - x;
-}
-
 /* Include the generated Neon decoder */
 #include "decode-neon-dp.c.inc"
 #include "decode-neon-ls.c.inc"

target/arm/translate.h

@@ -181,6 +181,12 @@ static inline int rsub_8(DisasContext *s, int x)
     return 8 - x;
 }
 
+static inline int neon_3same_fp_size(DisasContext *s, int x)
+{
+    /* Convert 0==fp32, 1==fp16 into a MO_* value */
+    return MO_32 - x;
+}
+
 static inline int arm_dc_feature(DisasContext *dc, int feature)
 {
     return (dc->features & (1ULL << feature)) != 0;