Memory offset calculations in 32-bit SIMD elements limit us to
addressing at most 4 GiB. Signed arithmetic further restricts it to
2 GiB.

The legacy OpenGL ES implementation limits image allocations to 1 GiB,
to discourage excessive memory usage which wouldn't typically succeed
on other (mobile) implementations anyway. So until we have a strong use
case which requires more, let's impose the same limit.

Bug: b/146515574
Change-Id: Iaa7bd22789fd20b4e7975d3d2ffee2a8f90b006a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39828
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>