Use native crc32 on x86 if available.

[Teselka]

This adds support for the native _mm_crc32_u8 on x86 provided by the SSE4.2 instruction set (please note that this one actually is crc32c, which uses a different polynomial that is used internally by the Dear ImGui currently).

Because it's a native instruction, it doesn't require a 1 kb lookup table.

[Teselka]

Oh, i forgot to see if anyone did this before #4933

[ocornut]

Hello, thanks for the PR! This looks simpler than #4933.

which uses a different polynomial that is used internally by the Dear ImGui currently).

I would want to tackle this first.
What would it take to change the regular C++ implementation to have output matching the use of _mm_crc32_u8() ? (and potentially of wider versions _mm_crc32_u32 if we decide it is worth using them).

[Teselka]

Hello, thanks for the PR! This looks simpler than #4933.

which uses a different polynomial that is used internally by the Dear ImGui currently).

I would want to tackle this first. What would it take to change the regular C++ implementation to have output matching the use of _mm_crc32_u8() ? (and potentially of wider versions _mm_crc32_u32 if we decide it is worth using them).

That would require to change only current crc lookup table

[Teselka]

Changed it, now it produces same results with or without native instruction.

[ocornut]

This is merged (in reverse order) as e6dd8f6 + 326dc95
I also updated test engine to match. Thank you!

[ocornut]

This is causing problem for people crafting/storing persistent .ini data, which is semi frequently done as a workaround to docking api limitations. I have added a #define IMGUI_USE_LEGACY_CRC32_ADLER compile-time option to revert to old tables:
c3ffd4c