Optimize floating-point minimum/maximum operations.

Combine a floating-point compare + select into min/max operations on
x86 using SSE2 instructions, where equivalent. Only handles the most
common cases for now (but for both scalars and vectors):

x > y ? x : y -> max
x < y ? x : y -> min

BUG=swiftshader:19

Change-Id: Ic87bfa90cefd8014af5624d85a2ecef1c891e328
Reviewed-on: https://chromium-review.googlesource.com/439814
Reviewed-by: Jim Stichnoth <stichnot@chromium.org>
2 files changed