Thanks for the feedback Kyrill.

On 28/07/2020, 10:16, "Kyrylo Tkachov" <kyrylo.tkac...@arm.com> wrote:

    Hi Joe,

    > -----Original Message-----
    > From: Gcc-patches <gcc-patches-boun...@gcc.gnu.org> On Behalf Of Joe
    > Ramsay
    > Sent: 27 July 2020 15:08
    > To: Jakub Jelinek via Gcc-patches <gcc-patches@gcc.gnu.org>
    > Subject: [PATCH] [PATCH][GCC] arm: Enable no-writeback vldr.16/vstr.16.
    > 
    > Hi,
    > 
    > There was previously no way to specify that a register operand cannot
    > have any writeback modifiers, and as a result the argument to vldr.16
    > and vstr.16 could be erroneously output with post-increment. This
    > change adds an operand specifier which forbids all writeback, and
    > selects it in the relevant case for vldr.16 and vstr.16
    > 
    > Bootstrapped on arm-linux, gcc and CMSIS-DSP testsuites are clean.
    > Is this patch OK for trunk? If yes, please commit on my behalf as I don't
    > have commit rights.
    > 
    > Thanks,
    > Joe
    > 
    > gcc/ChangeLog:
    > 
    > 2020-05-20  Joe Ramsay  <joe.ram...@arm.com>
    > 
    >         * config/arm/arm-protos.h (arm_coproc_mem_operand_no_writeback):
    > Declare prototype.
    >         (arm_mve_mode_and_operands_type_check): Declare prototype.
    >         * config/arm/arm.c (arm_coproc_mem_operand): Refactor to use
    > _arm_coproc_mem_operand.
    >         (arm_coproc_mem_operand_wb): New function to cover full, limited
    > and no writeback.
    >         (arm_coproc_mem_operand_no_writeback): New constraint for
    > memory operand with no writeback.
    >         (arm_print_operand): Implement 'j' specifier for memory operand 
that
    > does not support
    >         writeback.
    >         (arm_mve_mode_and_operands_type_check): New constraint check for
    > MVE memory operands.
    >         * config/arm/constraints.md: Add Uj constraint for VFP vldr.16 and
    > vstr.16.
    >         * config/arm/vfp.md (*mov_load_vfp_hf16): New pattern for vldr.16.
    >         (*mov_store_vfp_hf16): New pattern for vstr.16.
    >         (*mov<mode>_vfp_<mode>16): Remove MVE moves.
    > 
    > gcc/testsuite/ChangeLog:
    > 
    > 2020-05-20  Joe Ramsay  <joe.ram...@arm.com>
    > 
    >         * gcc.target/arm/mve/intrinsics/mve-vldstr16-no-writeback.c: New 
test.
    > 
    > ---
    >  gcc/config/arm/arm-protos.h                        |   3 +
    >  gcc/config/arm/arm.c                               | 100 
++++++++++++++++++---
    >  gcc/config/arm/constraints.md                      |   7 ++
    >  gcc/config/arm/vfp.md                              |  28 ++++--
    >  .../arm/mve/intrinsics/mve-vldstr16-no-writeback.c |  17 ++++
    >  5 files changed, 135 insertions(+), 20 deletions(-)
    >  create mode 100644 gcc/testsuite/gcc.target/arm/mve/intrinsics/mve-
    > vldstr16-no-writeback.c
    > 
    > diff --git a/gcc/config/arm/arm-protos.h b/gcc/config/arm/arm-protos.h
    > index 33d162c..e811da4 100644
    > --- a/gcc/config/arm/arm-protos.h
    > +++ b/gcc/config/arm/arm-protos.h
    > @@ -115,8 +115,11 @@ extern enum reg_class
    > coproc_secondary_reload_class (machine_mode, rtx,
    >  extern bool arm_tls_referenced_p (rtx);
    > 
    >  extern int arm_coproc_mem_operand (rtx, bool);
    > +extern int arm_coproc_mem_operand_no_writeback (rtx);
    > +extern int arm_coproc_mem_operand_wb (rtx, int);
    >  extern int neon_vector_mem_operand (rtx, int, bool);
    >  extern int mve_vector_mem_operand (machine_mode, rtx, bool);
    > +bool arm_mve_mode_and_operands_type_check (machine_mode, rtx, rtx);
    >  extern int neon_struct_mem_operand (rtx);
    > 
    >  extern rtx *neon_vcmla_lane_prepare_operands (rtx *);
    > diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
    > index 6b7ca82..ed080d2 100644
    > --- a/gcc/config/arm/arm.c
    > +++ b/gcc/config/arm/arm.c
    > @@ -13217,13 +13217,14 @@ neon_element_bits (machine_mode mode)
    >  /* Predicates for `match_operand' and `match_operator'.  */
    > 
    >  /* Return TRUE if OP is a valid coprocessor memory address pattern.
    > -   WB is true if full writeback address modes are allowed and is false
    > +   WB level is 2 if full writeback address modes are allowed, 1
    >     if limited writeback address modes (POST_INC and PRE_DEC) are
    > -   allowed.  */
    > +   allowed and 0 if no writeback at all is supported.  */
    > 
    >  int
    > -arm_coproc_mem_operand (rtx op, bool wb)
    > +arm_coproc_mem_operand_wb (rtx op, int wb_level)
    >  {
    > +  gcc_assert (wb_level == 0 || wb_level == 1 || wb_level == 2);
    >    rtx ind;
    > 
    >    /* Reject eliminable registers.  */
    > @@ -13256,16 +13257,18 @@ arm_coproc_mem_operand (rtx op, bool wb)
    > 
    >    /* Autoincremment addressing modes.  POST_INC and PRE_DEC are
    >       acceptable in any case (subject to verification by
    > -     arm_address_register_rtx_p).  We need WB to be true to accept
    > +     arm_address_register_rtx_p).  We need full writeback to accept
    > +     PRE_INC and POST_DEC, and at least restricted writeback for
    >       PRE_INC and POST_DEC.  */
    > -  if (GET_CODE (ind) == POST_INC
    > -      || GET_CODE (ind) == PRE_DEC
    > -      || (wb
    > -   && (GET_CODE (ind) == PRE_INC
    > -       || GET_CODE (ind) == POST_DEC)))
    > +  if (wb_level > 0
    > +      && (GET_CODE (ind) == POST_INC
    > +   || GET_CODE (ind) == PRE_DEC
    > +   || (wb_level > 1
    > +       && (GET_CODE (ind) == PRE_INC
    > +           || GET_CODE (ind) == POST_DEC))))
    >      return arm_address_register_rtx_p (XEXP (ind, 0), 0);
    > 
    > -  if (wb
    > +  if (wb_level > 1
    >        && (GET_CODE (ind) == POST_MODIFY || GET_CODE (ind) ==
    > PRE_MODIFY)
    >        && arm_address_register_rtx_p (XEXP (ind, 0), 0)
    >        && GET_CODE (XEXP (ind, 1)) == PLUS
    > @@ -13287,6 +13290,25 @@ arm_coproc_mem_operand (rtx op, bool wb)
    >    return FALSE;
    >  }
    > 
    > +/* Return TRUE if OP is a valid coprocessor memory address pattern.
    > +   WB is true if full writeback address modes are allowed and is false
    > +   if limited writeback address modes (POST_INC and PRE_DEC) are
    > +   allowed.  */
    > +
    > +int arm_coproc_mem_operand (rtx op, bool wb)
    > +{
    > +  return arm_coproc_mem_operand_wb (op, wb ? 2 : 1);
    > +}
    > +
    > +/* Return TRUE if OP is a valid coprocessor memory address pattern in a
    > +   context in which no writeback address modes are allowed.  */
    > +
    > +int
    > +arm_coproc_mem_operand_no_writeback (rtx op)
    > +{
    > +  return arm_coproc_mem_operand_wb (op, 0);
    > +}
    > +
    >  /* This function returns TRUE on matching mode and op.
    >  1. For given modes, check for [Rn], return TRUE for Rn <= LO_REGS.
    >  2. For other modes, check for [Rn], return TRUE for Rn < R15 (expect 
R13).
    > */
    > @@ -23549,8 +23571,8 @@ arm_print_condition (FILE *stream)
    > 
    >  /* Globally reserved letters: acln
    >     Puncutation letters currently used: @_|?().!#
    > -   Lower case letters currently used: bcdefhimpqtvwxyz
    > -   Upper case letters currently used: ABCDFGHJKLMNOPQRSTU
    > +   Lower case letters currently used: bcdefhijmpqtvwxyz
    > +   Upper case letters currently used: ABCDEFGHIJKLMNOPQRSTU
    >     Letters previously used, but now deprecated/obsolete: sVWXYZ.
    > 
    >     Note that the global reservation for 'c' is only for 
CONSTANT_ADDRESS_P.
    > @@ -24163,6 +24185,41 @@ arm_print_operand (FILE *stream, rtx x, int
    > code)
    >        }
    >        return;
    > 
    > +    /* To print memory operand in the case that the instruction does
    > +       not support writeback.  i.e. the output will look like either of
    > +       the following:
    > +       1. [Rn]
    > +       2. [Rn, #+/-<imm>] */

    I'm unsure why we need a new output modifier here ('j').
    Shouldn't the constraints/predicates on the patterns prevent the formation 
of writeback addresses that should be handled by the normal address printing 
code?

    Thanks,
    Kyrill

I don't think the existing specifiers are sufficient for this address mode. 
Previously we were using A specifier here, which has no provision for const int 
offset. We also considered E specifier, but E only supports offset with 
writeback (not without), so as far as I can tell we need a new specifier to 
handle the new addressing mode.

    > +    case 'j':
    > +      {
    > + gcc_assert (MEM_P (x));
    > + rtx addr = XEXP (x, 0);
    > +
    > + switch (GET_CODE (addr))
    > +   {
    > +   case REG:
    > +     asm_fprintf (stream, "[%r]", REGNO (addr));
    > +     return;
    > +
    > +   case PLUS:
    > +     {
    > +       rtx base = XEXP (addr, 0);
    > +       rtx index = XEXP (addr, 1);
    > +
    > +       if (!REG_P (base) || !CONST_INT_P (index))
    > +         gcc_unreachable ();
    > +
    > +       HOST_WIDE_INT offset = INTVAL (index);
    > +       asm_fprintf (stream, "[%r, #%wd]", REGNO (base), offset);
    > +       return;
    > +     }
    > +
    > +   default:
    > +     gcc_unreachable ();
    > +   }
    > + return;
    > +      }
    > +
    >      case 'C':
    >        {
    >   rtx addr;
    > @@ -33496,4 +33553,23 @@ arm_mode_base_reg_class (machine_mode
    > mode)
    > 
    >  struct gcc_target targetm = TARGET_INITIALIZER;
    > 
    > +bool
    > +arm_mve_mode_and_operands_type_check (machine_mode mode, rtx op0,
    > rtx op1)
    > +{
    > +  if (!(TARGET_HAVE_MVE || TARGET_HAVE_MVE_FLOAT))
    > +    {
    > +       return true;
    > +    }
    > +  else if (mode == E_BFmode)
    > +    {
    > +      return false;
    > +    }
    > +  else if ((s_register_operand (op0, mode) && MEM_P (op1))
    > +    || (s_register_operand (op1, mode) && MEM_P (op0)))
    > +    {
    > +      return false;
    > +    }
    > +  return true;
    > +}
    > +
    >  #include "gt-arm.h"
    > diff --git a/gcc/config/arm/constraints.md b/gcc/config/arm/constraints.md
    > index 011badc..ff229aa 100644
    > --- a/gcc/config/arm/constraints.md
    > +++ b/gcc/config/arm/constraints.md
    > @@ -452,6 +452,13 @@
    >   (and (match_code "mem")
    >        (match_test "TARGET_32BIT && arm_coproc_mem_operand (op,
    > FALSE)")))
    > 
    > +(define_memory_constraint "Uj"
    > + "@internal
    > +  In ARM/Thumb-2 state an VFP load/store address which does not support
    > +  writeback at all (eg vldr.16)."
    > + (and (match_code "mem")
    > +      (match_test "TARGET_32BIT &&
    > arm_coproc_mem_operand_no_writeback (op)")))
    > +
    >  (define_memory_constraint "Uy"
    >   "@internal
    >    In ARM/Thumb-2 state a valid iWMMX load/store address."
    > diff --git a/gcc/config/arm/vfp.md b/gcc/config/arm/vfp.md
    > index 3470679..13174bb 100644
    > --- a/gcc/config/arm/vfp.md
    > +++ b/gcc/config/arm/vfp.md
    > @@ -387,6 +387,22 @@
    >     (set_attr "arch"           
"t2,any,any,any,a,t2,any,any,any,any,any,any")]
    >  )
    > 
    > +(define_insn "*mov_load_vfp_hf16"
    > +  [(set (match_operand:HF 0 "s_register_operand" "=t")
    > + (match_operand:HF 1 "memory_operand" "Uj"))]
    > +  "TARGET_HAVE_MVE_FLOAT"
    > +  "vldr.16\\t%0, %j1"
    > +  [(set_attr "length" "4")]
    > +)
    > +
    > +(define_insn "*mov_store_vfp_hf16"
    > +  [(set (match_operand:HF 0 "memory_operand" "=Uj")
    > + (match_operand:HF 1 "s_register_operand"   "t"))]
    > +  "TARGET_HAVE_MVE_FLOAT"
    > +  "vstr.16\\t%1, %j0"
    > +  [(set_attr "length" "4")]
    > +)
    > +
    >  ;; HFmode and BFmode moves
    > 
    >  (define_insn "*mov<mode>_vfp_<mode>16"
    > @@ -396,6 +412,8 @@
    >                     "  m,r,t,r,r,t,Dv,Um,t, F"))]
    >    "TARGET_32BIT
    >     && TARGET_VFP_FP16INST
    > +   && arm_mve_mode_and_operands_type_check (<MODE>mode,
    > operands[0],
    > +                                     operands[1])
    >     && (s_register_operand (operands[0], <MODE>mode)
    >         || s_register_operand (operands[1], <MODE>mode))"
    >   {
    > @@ -414,15 +432,9 @@
    >      case 6: /* S register from immediate.  */
    >        return \"vmov.f16\\t%0, %1\t%@ __<fporbf>\";
    >      case 7: /* S register from memory.  */
    > -      if (TARGET_HAVE_MVE)
    > - return \"vldr.16\\t%0, %A1\";
    > -      else
    > - return \"vld1.16\\t{%z0}, %A1\";
    > +      return \"vld1.16\\t{%z0}, %A1\";
    >      case 8: /* Memory from S register.  */
    > -      if (TARGET_HAVE_MVE)
    > - return \"vstr.16\\t%1, %A0\";
    > -      else
    > - return \"vst1.16\\t{%z1}, %A0\";
    > +      return \"vst1.16\\t{%z1}, %A0\";
    >      case 9: /* ARM register from constant.  */
    >        {
    >   long bits;
    > diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/mve-vldstr16-no-
    > writeback.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/mve-vldstr16-no-
    > writeback.c
    > new file mode 100644
    > index 0000000..0a69ace
    > --- /dev/null
    > +++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/mve-vldstr16-no-
    > writeback.c
    > @@ -0,0 +1,17 @@
    > +/* { dg-require-effective-target arm_v8_1m_mve_fp_ok } */
    > +/* { dg-add-options arm_v8_1m_mve_fp } */
    > +/* { dg-additional-options "-O2" } */
    > +
    > +void
    > +fn1 (__fp16 *pSrc)
    > +{
    > +  __fp16 high;
    > +  __fp16 *pDst = 0;
    > +  unsigned i;
    > +  for (i = 0;; i++)
    > +    if (pSrc[i])
    > +      pDst[i] = high;
    > +}
    > +
    > +/* { dg-final { scan-assembler {vldr\.16\ts[0-9]+, \[r[0-9]+\]\n} } } */
    > +/* { dg-final { scan-assembler {vstr\.16\ts[0-9]+, \[r[0-9]+\]\n} } } */
    > --
    > 2.7.4
    > 


Reply via email to