On 10/24/2016 09:08 PM, Nikunj A Dadhania wrote:
Richard Henderson <[email protected]> writes:
On 10/24/2016 02:14 AM, Nikunj A Dadhania wrote:
+#define EXTRACT_BITS(size) \
+static inline uint##size##_t extract_bits_u##size(uint##size##_t reg, \
+ uint##size##_t start, \
+ uint##size##_t end) \
+{ \
+ uint##size##_t nr_mask_bits = end - start + 1; \
+ uint##size##_t val = 1; \
+ uint##size##_t mask = (val << nr_mask_bits) - 1; \
+ uint##size##_t shifted_reg = reg >> ((size - 1) - end); \
+ return shifted_reg & mask; \
+}
+
+EXTRACT_BITS(64);
+EXTRACT_BITS(32);
We already have extract32 and extract64, which you're (nearly) duplicating.
The bit position number notation is different, because of this using the
above routine, MSB=0 and LSB=63.
While the below assumes: MSB=63 and LSB=0
static inline uint64_t extract64(uint64_t value, int start, int length)
{
assert(start >= 0 && length > 0 && length <= 64 - start);
return (value >> start) & (~0ULL >> (64 - length));
}
Let me know if I am missing something here.
Since the arguments to extract_bits_uN are completely under your control, via
the arguments to VRLMI, this is a non-argument. Just change them to
little-endian position + length.
(And, after you do that conversion for vrldmi and vilwmi, you'll see why
big-endian bit numbering is the spawn of the devil. ;-)
r~
