target-arm: A64: Add saturating accumulate ops (USQADD/SUQADD)

Add the saturating accumulate operations USQADD and SUQADD
to the A64 instruction set. This completes coverage of A64 Neon.
These operations (which are unsigned + signed -> signed and
signed + unsigned -> unsigned) don't exist in the A32/T32
instruction set, so require a complete new set of helper functions.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>

1 files changed, 165 insertions, 0 deletions

diff --git a/target-arm/neon_helper.c b/target-arm/neon_helper.c
index e23f224..8d6f9a9 100644
--- a/target-arm/neon_helper.c
+++ b/target-arm/neon_helper.c
@@ -236,6 +236,171 @@ uint64_t HELPER(neon_qadd_s64)(CPUARMState *env, uint64_t src1, uint64_t src2)
     return res;
 }
 
+/* Unsigned saturating accumulate of signed value
+ *
+ * Op1/Rn is treated as signed
+ * Op2/Rd is treated as unsigned
+ *
+ * Explicit casting is used to ensure the correct sign extension of
+ * inputs. The result is treated as a unsigned value and saturated as such.
+ *
+ * We use a macro for the 8/16 bit cases which expects signed integers of va,
+ * vb, and vr for interim calculation and an unsigned 32 bit result value r.
+ */
+
+#define USATACC(bits, shift) \
+    do { \
+        va = sextract32(a, shift, bits);                                \
+        vb = extract32(b, shift, bits);                                 \
+        vr = va + vb;                                                   \
+        if (vr > UINT##bits##_MAX) {                                    \
+            SET_QC();                                                   \
+            vr = UINT##bits##_MAX;                                      \
+        } else if (vr < 0) {                                            \
+            SET_QC();                                                   \
+            vr = 0;                                                     \
+        }                                                               \
+        r = deposit32(r, shift, bits, vr);                              \
+   } while (0)
+
+uint32_t HELPER(neon_uqadd_s8)(CPUARMState *env, uint32_t a, uint32_t b)
+{
+    int16_t va, vb, vr;
+    uint32_t r = 0;
+
+    USATACC(8, 0);
+    USATACC(8, 8);
+    USATACC(8, 16);
+    USATACC(8, 24);
+    return r;
+}
+
+uint32_t HELPER(neon_uqadd_s16)(CPUARMState *env, uint32_t a, uint32_t b)
+{
+    int32_t va, vb, vr;
+    uint64_t r = 0;
+
+    USATACC(16, 0);
+    USATACC(16, 16);
+    return r;
+}
+
+#undef USATACC
+
+uint32_t HELPER(neon_uqadd_s32)(CPUARMState *env, uint32_t a, uint32_t b)
+{
+    int64_t va = (int32_t)a;
+    int64_t vb = (uint32_t)b;
+    int64_t vr = va + vb;
+    if (vr > UINT32_MAX) {
+        SET_QC();
+        vr = UINT32_MAX;
+    } else if (vr < 0) {
+        SET_QC();
+        vr = 0;
+    }
+    return vr;
+}
+
+uint64_t HELPER(neon_uqadd_s64)(CPUARMState *env, uint64_t a, uint64_t b)
+{
+    uint64_t res;
+    res = a + b;
+    /* We only need to look at the pattern of SIGN bits to detect
+     * +ve/-ve saturation
+     */
+    if (~a & b & ~res & SIGNBIT64) {
+        SET_QC();
+        res = UINT64_MAX;
+    } else if (a & ~b & res & SIGNBIT64) {
+        SET_QC();
+        res = 0;
+    }
+    return res;
+}
+
+/* Signed saturating accumulate of unsigned value
+ *
+ * Op1/Rn is treated as unsigned
+ * Op2/Rd is treated as signed
+ *
+ * The result is treated as a signed value and saturated as such
+ *
+ * We use a macro for the 8/16 bit cases which expects signed integers of va,
+ * vb, and vr for interim calculation and an unsigned 32 bit result value r.
+ */
+
+#define SSATACC(bits, shift) \
+    do { \
+        va = extract32(a, shift, bits);                                 \
+        vb = sextract32(b, shift, bits);                                \
+        vr = va + vb;                                                   \
+        if (vr > INT##bits##_MAX) {                                     \
+            SET_QC();                                                   \
+            vr = INT##bits##_MAX;                                       \
+        } else if (vr < INT##bits##_MIN) {                              \
+            SET_QC();                                                   \
+            vr = INT##bits##_MIN;                                       \
+        }                                                               \
+        r = deposit32(r, shift, bits, vr);                              \
+    } while (0)
+
+uint32_t HELPER(neon_sqadd_u8)(CPUARMState *env, uint32_t a, uint32_t b)
+{
+    int16_t va, vb, vr;
+    uint32_t r = 0;
+
+    SSATACC(8, 0);
+    SSATACC(8, 8);
+    SSATACC(8, 16);
+    SSATACC(8, 24);
+    return r;
+}
+
+uint32_t HELPER(neon_sqadd_u16)(CPUARMState *env, uint32_t a, uint32_t b)
+{
+    int32_t va, vb, vr;
+    uint32_t r = 0;
+
+    SSATACC(16, 0);
+    SSATACC(16, 16);
+
+    return r;
+}
+
+#undef SSATACC
+
+uint32_t HELPER(neon_sqadd_u32)(CPUARMState *env, uint32_t a, uint32_t b)
+{
+    int64_t res;
+    int64_t op1 = (uint32_t)a;
+    int64_t op2 = (int32_t)b;
+    res = op1 + op2;
+    if (res > INT32_MAX) {
+        SET_QC();
+        res = INT32_MAX;
+    } else if (res < INT32_MIN) {
+        SET_QC();
+        res = INT32_MIN;
+    }
+    return res;
+}
+
+uint64_t HELPER(neon_sqadd_u64)(CPUARMState *env, uint64_t a, uint64_t b)
+{
+    uint64_t res;
+    res = a + b;
+    /* We only need to look at the pattern of SIGN bits to detect an overflow */
+    if (((a & res)
+         | (~b & res)
+         | (a & ~b)) & SIGNBIT64) {
+        SET_QC();
+        res = INT64_MAX;
+    }
+    return res;
+}
+
+
 #define NEON_USAT(dest, src1, src2, type) do { \
     uint32_t tmp = (uint32_t)src1 - (uint32_t)src2; \
     if (tmp != (type)tmp) { \