[Bug target/96401] [nvptx] Take advantage of subword ld/st/cvt

[vries at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96401

--- Comment #3 from Tom de Vries <vries at gcc dot gnu.org> ---
Note that with the proposed TARGET_TRULY_NOOP_TRUNCATION -> false change (
https://gcc.gnu.org/pipermail/gcc-patches/2020-July/549896.html ), we start out
with the same ptx insns, but with the cvt.u16.u32 a truncate instead of a
subreg move:
...
//(insn 5 2 6 2
//    (set (reg:SI 22 [ v$0_1 ])
//         (mem/v/c:SI (reg/f:DI 2 %frame) [1 v+0 S4 A128]))
//     "test.c":7:6 6 {*movsi_insn}
//     (nil))
                ld.u32  %r22, [%frame]; // 5    [c=4]  *movsi_insn/1

//(insn 6 5 9 2
//    (set (reg:HI 24 [ v$0_1 ])
//         (truncate:HI (reg:SI 22 [ v$0_1 ])))
       "test.c":7:6 30 {truncsihi2}
//     (expr_list:REG_DEAD (reg:SI 22 [ v$0_1 ])
//     (nil)))
                cvt.u16.u32     %r24, %r22;     // 6    [c=4]  truncsihi2/0

//(insn 9 6 12 2
//    (set (mem/v/c:HI (plus:DI (reg/f:DI 2 %frame)
//                              (const_int 4 [0x4])) [2 v2+0 S2 A32])
//         (reg:HI 24 [ v$0_1 ])) "test.c":7:6 5 {*movhi_insn}
//     (expr_list:REG_DEAD (reg:HI 24 [ v$0_1 ])
//        (nil)))
                st.u16  [%frame+4], %r24;       // 9    [c=4]  *movhi_insn/2
...

Still, with the changes in comment 1 enabled we end up with the desired two
insns, though a bit later, at cse2 (265r), and not using movhi_insn:
...
(insn 9 5 0 2 (set (mem/v/c:HI (plus:DI (reg/f:DI 2 %frame)
                (const_int 4 [0x4])) [2 v2+0 S2 A32])
        (truncate:HI (reg:SI 22 [ v$0_1 ]))) "test.c":7:6 30 {truncsihi2}
     (expr_list:REG_DEAD (reg:HI 24 [ v$0_1 ])
        (nil)))
...
so we might get this just with the nvptx_modes_tieable_p change.