Kyrill,

Indeed, the correct log would be:

    The type assigned to some insn definitions was not correct.

    gcc/
        * config/aarch64/aarch64.md
(*movhf_aarch64): Change the type of "mov %0.h[0], %1.h[0] to "neon_move".
        (*movtf_aarch64): Change the type of "fmov %s0, wzr" to "f_mcr".
(*cmov<mode>_insn): Change the types of "mov %<w>0, {-1,1}" to "mov_imm".
        (*cmovsi_insn_uxtw): Likewise.

Thank you,

--
Evandro Menezes

On 10/20/2015 05:59 AM, Kyrill Tkachov wrote:
Hi Evandro,

On 19/10/15 22:05, Evandro Menezes wrote:
The type assigned to some insn definitions was seemingly not correct:

 * "movi %d0, %1" was of type "fmov"
 * "fmov %s0, wzr" was of type "fconstd"
 * "mov %<w>0, {-1,1}" were of type "csel"

This patch changes their types to:

 * "movi %d0, %1" to type "neon_move"
 * "fmov %s0, wzr" to type "f_mcr"
 * "mov %<w>0, {-1,1}" to type "mov_imm"

Please, commit if it's alright.

Thank you,


Looking at your ChangeLog...

gcc/
    * config/aarch64/aarch64.md
    (*movdi_aarch64): Change the type of "movi %d0, %1" to "neon_move".
    (*movtf_aarch64): Change the type of "fmov %s0, wzr" to "f_mcr".
(*cmov<mode>_insn): Change the types of "mov %<w>0, {-1,1}" to "mov_imm".
    (*cmovsi_insn_uxtw): Idem

The preferred form is "Likewise" rather than "Idem" AFAIK. Also, full stop at the end.

--- a/gcc/config/aarch64/aarch64.md
+++ b/gcc/config/aarch64/aarch64.md
@@ -1130,7 +1130,7 @@
    ldrh\\t%w0, %1
    strh\\t%w1, %0
    mov\\t%w0, %w1"
-  [(set_attr "type" "neon_from_gp,neon_to_gp,fmov,\
+  [(set_attr "type" "neon_from_gp,neon_to_gp,neon_move,\
                      f_loads,f_stores,load1,store1,mov_reg")
    (set_attr "simd" "yes,yes,yes,*,*,*,*,*")
    (set_attr "fp"   "*,*,*,yes,yes,*,*,*")]


I don't think this matches up with your changelog entry.
This isn't the *movdi_aarch64 pattern. From what I can see the *movdi_aarch64 pattern already has the type neon_move on the movi\\t%d0, %1 alternative (the last one).

In fact, if I apply your patch using "patch -p1" I see it being applied to the *movhf_aarch64 pattern.
Is that what you intended?

Thanks,
Kyrill



Reply via email to