This hasn't been true in 6+ years, if it was even true then. Before
we rewrote the compiler and introduced GLSL IR in 2010-2011, i965 used
to have two compiler backends for WM programs, based on Mesa IR. One
handled flow control and was SIMD8-only, while the other was SIMD16
only and didn't handle flow control. Or something like that.
Even then, this certainly didn't handle vertex shaders, so "all ...
code generation" is a bit strong.
}
}
-/**
- * All Mesa program -> GPU code generation goes through this function.
- * Depending on the instructions used (i.e. flow control instructions)
- * we'll use one of two code generators.
- */
static bool
brw_codegen_wm_prog(struct brw_context *brw,
struct brw_program *fp,