From: Richard Sandiford <richard.sandiford@arm.com>
To: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
Cc: GCC Patches <gcc-patches@gcc.gnu.org>
Subject: Re: [PATCH][AArch64] Cleanup move immediate code
Date: Thu, 01 Dec 2022 10:59:23 +0000 [thread overview]
Message-ID: <mptzgc74cfo.fsf@arm.com> (raw)
In-Reply-To: <PAWPR08MB898241FDAA6707BBAF46722B83129@PAWPR08MB8982.eurprd08.prod.outlook.com> (Wilco Dijkstra's message of "Tue, 29 Nov 2022 13:57:45 +0000")
Wilco Dijkstra <Wilco.Dijkstra@arm.com> writes:
> Hi Richard,
>
>> Just to make sure I understand: isn't it really just MOVN? I would have
>> expected a 32-bit MOVZ to be equivalent to (and add no capabilities over)
>> a 64-bit MOVZ.
>
> The 32-bit MOVZ immediates are equivalent, MOVN never overlaps, and
> MOVI has some overlaps . Since we allow all 3 variants, the 2 alternatives
> in the movdi pattern are overlapping for MOVZ and MOVI immediates.
>
>> I agree the ctz trick is more elegant than (and an improvement over)
>> the current approach to testing for movz. But I think the overall logic
>> is harder to follow than it was in the original version. Initially
>> canonicalising val2 based on the sign bit seems unintuitive since we
>> still need to handle all four combinations of (top bit set, top bit clear)
>> x (low 48 bits set, low 48 bits clear). I preferred the original
>> approach of testing once with the original value (for MOVZ) and once
>> with the inverted value (for MOVN).
>
> Yes, the canonicalization on the sign ends up requiring 2 special cases.
> Handling the MOVZ case first and then MOVN does avoid that, and makes
> things simpler overall, so I've used that approach in v2.
>
>> Don't the new cases boil down to: if mode is DImode and the upper 32 bits
>> are clear, we can test based on SImode instead? In other words, couldn't
>> the "(val >> 32) == 0" part of the final test be done first, with the
>> effect of changing the mode to SImode? Something like:
>
> Yes that works. I used masking of the top bits to avoid repeatedly testing the
> same condition. The new version removes most special cases and ends up
> both smaller and simpler:
>
>
> v2: Simplify the special cases in aarch64_move_imm, use aarch64_is_movz.
>
> Simplify, refactor and improve various move immediate functions.
> Allow 32-bit MOVZ/I/N as a valid 64-bit immediate which removes special
> cases in aarch64_internal_mov_immediate. Add new constraint so the movdi
> pattern only needs a single alternative for move immediate.
>
> Passes bootstrap and regress, OK for commit?
>
> gcc/ChangeLog:
>
> * config/aarch64/aarch64.cc (aarch64_bitmask_imm): Use unsigned type.
> (aarch64_zeroextended_move_imm): New function.
> (aarch64_move_imm): Refactor, assert mode is SImode or DImode.
> (aarch64_internal_mov_immediate): Assert mode is SImode or DImode.
> Simplify special cases.
> (aarch64_uimm12_shift): Simplify code.
> (aarch64_clamp_to_uimm12_shift): Likewise.
> (aarch64_movw_imm): Rename to aarch64_is_movz.
> (aarch64_float_const_rtx_p): Pass either SImode or DImode to
> aarch64_internal_mov_immediate.
> (aarch64_rtx_costs): Likewise.
> * config/aarch64/aarch64.md (movdi_aarch64): Merge 'N' and 'M'
> constraints into single 'O'.
> (mov<mode>_aarch64): Likewise.
> * config/aarch64/aarch64-protos.h (aarch64_move_imm): Use unsigned.
> (aarch64_bitmask_imm): Likewise.
> (aarch64_uimm12_shift): Likewise.
> (aarch64_zeroextended_move_imm): New prototype.
> * config/aarch64/constraints.md: Add 'O' for 32/64-bit immediates,
> limit 'N' to 64-bit only moves.
>
> ---
>
> diff --git a/gcc/config/aarch64/aarch64-protos.h b/gcc/config/aarch64/aarch64-protos.h
> index 4be93c93c26e091f878bc8e4cf06e90888405fb2..8bce6ec7599edcc2e6a1d8006450f35c0ce7f61f 100644
> --- a/gcc/config/aarch64/aarch64-protos.h
> +++ b/gcc/config/aarch64/aarch64-protos.h
> @@ -756,7 +756,7 @@ void aarch64_post_cfi_startproc (void);
> poly_int64 aarch64_initial_elimination_offset (unsigned, unsigned);
> int aarch64_get_condition_code (rtx);
> bool aarch64_address_valid_for_prefetch_p (rtx, bool);
> -bool aarch64_bitmask_imm (HOST_WIDE_INT val, machine_mode);
> +bool aarch64_bitmask_imm (unsigned HOST_WIDE_INT val, machine_mode);
> unsigned HOST_WIDE_INT aarch64_and_split_imm1 (HOST_WIDE_INT val_in);
> unsigned HOST_WIDE_INT aarch64_and_split_imm2 (HOST_WIDE_INT val_in);
> bool aarch64_and_bitmask_imm (unsigned HOST_WIDE_INT val_in, machine_mode mode);
> @@ -793,7 +793,7 @@ bool aarch64_masks_and_shift_for_bfi_p (scalar_int_mode, unsigned HOST_WIDE_INT,
> unsigned HOST_WIDE_INT,
> unsigned HOST_WIDE_INT);
> bool aarch64_zero_extend_const_eq (machine_mode, rtx, machine_mode, rtx);
> -bool aarch64_move_imm (HOST_WIDE_INT, machine_mode);
> +bool aarch64_move_imm (unsigned HOST_WIDE_INT, machine_mode);
> machine_mode aarch64_sve_int_mode (machine_mode);
> opt_machine_mode aarch64_sve_pred_mode (unsigned int);
> machine_mode aarch64_sve_pred_mode (machine_mode);
> @@ -843,8 +843,9 @@ bool aarch64_sve_float_arith_immediate_p (rtx, bool);
> bool aarch64_sve_float_mul_immediate_p (rtx);
> bool aarch64_split_dimode_const_store (rtx, rtx);
> bool aarch64_symbolic_address_p (rtx);
> -bool aarch64_uimm12_shift (HOST_WIDE_INT);
> +bool aarch64_uimm12_shift (unsigned HOST_WIDE_INT);
> int aarch64_movk_shift (const wide_int_ref &, const wide_int_ref &);
> +bool aarch64_zeroextended_move_imm (unsigned HOST_WIDE_INT);
> bool aarch64_use_return_insn_p (void);
> const char *aarch64_output_casesi (rtx *);
>
> diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
> index e97f3b32f7c7f43564d6a4207eae5a34b9e9bfe7..fab478ada8d72d163c8daa3713e5ff6945de409f 100644
> --- a/gcc/config/aarch64/aarch64.cc
> +++ b/gcc/config/aarch64/aarch64.cc
> @@ -5625,12 +5625,10 @@ aarch64_bitmask_imm (unsigned HOST_WIDE_INT val)
>
> /* Return true if VAL is a valid bitmask immediate for MODE. */
> bool
> -aarch64_bitmask_imm (HOST_WIDE_INT val_in, machine_mode mode)
> +aarch64_bitmask_imm (unsigned HOST_WIDE_INT val, machine_mode mode)
> {
> if (mode == DImode)
> - return aarch64_bitmask_imm (val_in);
> -
> - unsigned HOST_WIDE_INT val = val_in;
> + return aarch64_bitmask_imm (val);
>
> if (mode == SImode)
> return aarch64_bitmask_imm ((val & 0xffffffff) | (val << 32));
> @@ -5669,51 +5667,57 @@ aarch64_check_bitmask (unsigned HOST_WIDE_INT val,
> }
>
>
> -/* Return true if val is an immediate that can be loaded into a
> - register by a MOVZ instruction. */
> -static bool
> -aarch64_movw_imm (HOST_WIDE_INT val, scalar_int_mode mode)
> +/* Return true if immediate VAL can only be created by using a 32-bit
> + zero-extended move immediate, not by a 64-bit move. */
> +bool
> +aarch64_zeroextended_move_imm (unsigned HOST_WIDE_INT val)
> {
> - if (GET_MODE_SIZE (mode) > 4)
> - {
> - if ((val & (((HOST_WIDE_INT) 0xffff) << 32)) == val
> - || (val & (((HOST_WIDE_INT) 0xffff) << 48)) == val)
> - return 1;
> - }
> - else
> - {
> - /* Ignore sign extension. */
> - val &= (HOST_WIDE_INT) 0xffffffff;
> - }
> - return ((val & (((HOST_WIDE_INT) 0xffff) << 0)) == val
> - || (val & (((HOST_WIDE_INT) 0xffff) << 16)) == val);
> + if (val < 65536 || (val >> 32) != 0 || (val & 0xffff) == 0)
> + return false;
> + return !aarch64_bitmask_imm (val);
> }
>
>
> -/* Return true if VAL is an immediate that can be loaded into a
> - register in a single instruction. */
> +/* Return true if VAL is a valid MOVZ immediate. */
> +static inline bool
> +aarch64_is_movz (unsigned HOST_WIDE_INT val)
> +{
> + return (val >> (ctz_hwi (val) & 48)) < 65536;
> +}
> +
> +
> +/* Return true if VAL is an immediate that can be created by a single
> + MOV instruction. */
> bool
> -aarch64_move_imm (HOST_WIDE_INT val, machine_mode mode)
> +aarch64_move_imm (unsigned HOST_WIDE_INT val, machine_mode mode)
> {
> - scalar_int_mode int_mode;
> - if (!is_a <scalar_int_mode> (mode, &int_mode))
> - return false;
> + gcc_assert (mode == SImode || mode == DImode);
>
> - if (aarch64_movw_imm (val, int_mode) || aarch64_movw_imm (~val, int_mode))
> - return 1;
> - return aarch64_bitmask_imm (val, int_mode);
> + if (val < 65536)
> + return true;
> +
> + unsigned HOST_WIDE_INT mask =
> + (val >> 32) == 0 || mode == SImode ? 0xffffffff : HOST_WIDE_INT_M1U;
> +
> + if (aarch64_is_movz (val & mask) || aarch64_is_movz (~val & mask))
> + return true;
> +
> + val = (val & mask) | ((val << 32) & ~mask);
> + return aarch64_bitmask_imm (val);
> }
>
>
> static int
> aarch64_internal_mov_immediate (rtx dest, rtx imm, bool generate,
> - scalar_int_mode mode)
> + machine_mode mode)
> {
> int i;
> unsigned HOST_WIDE_INT val, val2, mask;
> int one_match, zero_match;
> int num_insns;
>
> + gcc_assert (mode == SImode || mode == DImode);
> +
> val = INTVAL (imm);
>
> if (aarch64_move_imm (val, mode))
> @@ -5723,31 +5727,6 @@ aarch64_internal_mov_immediate (rtx dest, rtx imm, bool generate,
> return 1;
> }
>
> - /* Check to see if the low 32 bits are either 0xffffXXXX or 0xXXXXffff
> - (with XXXX non-zero). In that case check to see if the move can be done in
> - a smaller mode. */
> - val2 = val & 0xffffffff;
> - if (mode == DImode
> - && aarch64_move_imm (val2, SImode)
> - && (((val >> 32) & 0xffff) == 0 || (val >> 48) == 0))
> - {
> - if (generate)
> - emit_insn (gen_rtx_SET (dest, GEN_INT (val2)));
> -
> - /* Check if we have to emit a second instruction by checking to see
> - if any of the upper 32 bits of the original DI mode value is set. */
> - if (val == val2)
> - return 1;
> -
> - i = (val >> 48) ? 48 : 32;
> -
> - if (generate)
> - emit_insn (gen_insv_immdi (dest, GEN_INT (i),
> - GEN_INT ((val >> i) & 0xffff)));
> -
> - return 2;
> - }
> -
> if ((val >> 32) == 0 || mode == SImode)
> {
> if (generate)
> @@ -5771,24 +5750,31 @@ aarch64_internal_mov_immediate (rtx dest, rtx imm, bool generate,
> one_match = ((~val & mask) == 0) + ((~val & (mask << 16)) == 0) +
> ((~val & (mask << 32)) == 0) + ((~val & (mask << 48)) == 0);
>
> + /* Try a bitmask immediate and a movk to generate the immediate
> + in 2 instructions. */
> +
> if (zero_match < 2 && one_match < 2)
> {
> - /* Try emitting a bitmask immediate with a movk replacing 16 bits.
> - For a 64-bit bitmask try whether changing 16 bits to all ones or
> - zeroes creates a valid bitmask. To check any repeated bitmask,
> - try using 16 bits from the other 32-bit half of val. */
> -
> for (i = 0; i < 64; i += 16)
> - if (aarch64_check_bitmask (val, val2, mask << i))
> - {
> - if (generate)
> - {
> - emit_insn (gen_rtx_SET (dest, GEN_INT (val2)));
> - emit_insn (gen_insv_immdi (dest, GEN_INT (i),
> - GEN_INT ((val >> i) & 0xffff)));
> - }
> - return 2;
> - }
> + {
> + if (aarch64_check_bitmask (val, val2, mask << i))
> + break;
> +
> + val2 = val & ~(mask << i);
> + if ((val2 >> 32) == 0 && aarch64_move_imm (val2, DImode))
> + break;
> + }
> +
> + if (i != 64)
> + {
> + if (generate)
> + {
> + emit_insn (gen_rtx_SET (dest, GEN_INT (val2)));
> + emit_insn (gen_insv_immdi (dest, GEN_INT (i),
> + GEN_INT ((val >> i) & 0xffff)));
> + }
> + return 2;
> + }
> }
>
> /* Try a bitmask plus 2 movk to generate the immediate in 3 instructions. */
> @@ -5857,26 +5843,24 @@ aarch64_mov128_immediate (rtx imm)
> /* Return true if val can be encoded as a 12-bit unsigned immediate with
> a left shift of 0 or 12 bits. */
> bool
> -aarch64_uimm12_shift (HOST_WIDE_INT val)
> +aarch64_uimm12_shift (unsigned HOST_WIDE_INT val)
> {
> - return ((val & (((HOST_WIDE_INT) 0xfff) << 0)) == val
> - || (val & (((HOST_WIDE_INT) 0xfff) << 12)) == val
> - );
> + return val < 4096 || (val & 0xfff000) == val;
> }
>
> /* Returns the nearest value to VAL that will fit as a 12-bit unsigned immediate
> that can be created with a left shift of 0 or 12. */
> static HOST_WIDE_INT
> -aarch64_clamp_to_uimm12_shift (HOST_WIDE_INT val)
> +aarch64_clamp_to_uimm12_shift (unsigned HOST_WIDE_INT val)
> {
> /* Check to see if the value fits in 24 bits, as that is the maximum we can
> handle correctly. */
> - gcc_assert ((val & 0xffffff) == val);
> + gcc_assert (val < 0x1000000);
>
> - if (((val & 0xfff) << 0) == val)
> + if (val < 4096)
> return val;
>
> - return val & (0xfff << 12);
> + return val & 0xfff000;
> }
>
>
> @@ -7024,8 +7008,7 @@ aarch64_expand_mov_immediate (rtx dest, rtx imm)
> return;
> }
>
> - aarch64_internal_mov_immediate (dest, imm, true,
> - as_a <scalar_int_mode> (mode));
> + aarch64_internal_mov_immediate (dest, imm, true, mode);
> }
>
> /* Return the MEM rtx that provides the canary value that should be used
> @@ -11197,9 +11180,7 @@ aarch64_float_const_rtx_p (rtx x)
> && SCALAR_FLOAT_MODE_P (mode)
> && aarch64_reinterpret_float_as_int (x, &ival))
> {
> - scalar_int_mode imode = (mode == HFmode
> - ? SImode
> - : int_mode_for_mode (mode).require ());
> + machine_mode imode = (mode == DFmode) ? DImode : SImode;
It looks like this might mishandle DDmode, if not now then in the future.
Seems safer to use known_eq (GET_MODE_SIZE (mode), 8)
> int num_instr = aarch64_internal_mov_immediate
> (NULL_RTX, gen_int_mode (ival, imode), false, imode);
> return num_instr < 3;
> @@ -13857,10 +13838,9 @@ aarch64_rtx_costs (rtx x, machine_mode mode, int outer ATTRIBUTE_UNUSED,
> proportionally expensive to the number of instructions
> required to build that constant. This is true whether we
> are compiling for SPEED or otherwise. */
> - if (!is_a <scalar_int_mode> (mode, &int_mode))
> - int_mode = word_mode;
> + machine_mode imode = (mode == SImode) ? SImode : DImode;
And here known_le (GET_MODE_SIZE (mode), 4), so that we handle QImode
and HImode as SImode rather than DImode.
> *cost = COSTS_N_INSNS (aarch64_internal_mov_immediate
> - (NULL_RTX, x, false, int_mode));
> + (NULL_RTX, x, false, imode));
> }
> return true;
>
> @@ -13876,9 +13856,7 @@ aarch64_rtx_costs (rtx x, machine_mode mode, int outer ATTRIBUTE_UNUSED,
> bool succeed = aarch64_reinterpret_float_as_int (x, &ival);
> gcc_assert (succeed);
>
> - scalar_int_mode imode = (mode == HFmode
> - ? SImode
> - : int_mode_for_mode (mode).require ());
> + machine_mode imode = (mode == DFmode) ? DImode : SImode;
Same GET_MODE_SIZE test here.
> int ncost = aarch64_internal_mov_immediate
> (NULL_RTX, gen_int_mode (ival, imode), false, imode);
> *cost += COSTS_N_INSNS (ncost);
> diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
> index 76b6898ca048c559ff7f7fba78119161b3d382f6..5694a1efea1df8e532520475140c8748b54e3d89 100644
> --- a/gcc/config/aarch64/aarch64.md
> +++ b/gcc/config/aarch64/aarch64.md
> @@ -1309,16 +1309,15 @@ (define_insn_and_split "*movsi_aarch64"
> )
>
> (define_insn_and_split "*movdi_aarch64"
> - [(set (match_operand:DI 0 "nonimmediate_operand" "=r,k,r,r,r,r,r, r,w, m,m, r, r, r, w,r,w, w")
> - (match_operand:DI 1 "aarch64_mov_operand" " r,r,k,N,M,n,Usv,m,m,rZ,w,Usw,Usa,Ush,rZ,w,w,Dd"))]
> + [(set (match_operand:DI 0 "nonimmediate_operand" "=r,k,r,r,r,r, r,w, m,m, r, r, r, w,r,w, w")
> + (match_operand:DI 1 "aarch64_mov_operand" " r,r,k,O,n,Usv,m,m,rZ,w,Usw,Usa,Ush,rZ,w,w,Dd"))]
> "(register_operand (operands[0], DImode)
> || aarch64_reg_or_zero (operands[1], DImode))"
> "@
> mov\\t%x0, %x1
> mov\\t%0, %x1
> mov\\t%x0, %1
> - mov\\t%x0, %1
> - mov\\t%w0, %1
> + * return aarch64_zeroextended_move_imm (INTVAL (operands[1])) ? \"mov\\t%w0, %1\" : \"mov\\t%x0, %1\";
> #
> * return aarch64_output_sve_cnt_immediate (\"cnt\", \"%x0\", operands[1]);
> ldr\\t%x0, %1
> @@ -1340,11 +1339,11 @@ (define_insn_and_split "*movdi_aarch64"
> DONE;
> }"
> ;; The "mov_imm" type for CNTD is just a placeholder.
> - [(set_attr "type" "mov_reg,mov_reg,mov_reg,mov_imm,mov_imm,mov_imm,mov_imm,
> + [(set_attr "type" "mov_reg,mov_reg,mov_reg,mov_imm,mov_imm,mov_imm,
> load_8,load_8,store_8,store_8,load_8,adr,adr,f_mcr,f_mrc,
> fmov,neon_move")
> - (set_attr "arch" "*,*,*,*,*,*,sve,*,fp,*,fp,*,*,*,fp,fp,fp,simd")
> - (set_attr "length" "4,4,4,4,4,*, 4,4, 4,4, 4,8,4,4, 4, 4, 4, 4")]
> + (set_attr "arch" "*,*,*,*,*,sve,*,fp,*,fp,*,*,*,fp,fp,fp,simd")
> + (set_attr "length" "4,4,4,4,*, 4,4, 4,4, 4,8,4,4, 4, 4, 4, 4")]
> )
>
> (define_insn "insv_imm<mode>"
> @@ -1508,7 +1507,7 @@ (define_insn "*mov<mode>_aarch64"
>
> (define_insn "*mov<mode>_aarch64"
> [(set (match_operand:DFD 0 "nonimmediate_operand" "=w, w ,?r,w,w ,w ,w,m,r,m ,r,r")
> - (match_operand:DFD 1 "general_operand" "Y , ?rY, w,w,Ufc,Uvi,m,w,m,rY,r,N"))]
> + (match_operand:DFD 1 "general_operand" "Y , ?rY, w,w,Ufc,Uvi,m,w,m,rY,r,O"))]
> "TARGET_FLOAT && (register_operand (operands[0], <MODE>mode)
> || aarch64_reg_or_fp_zero (operands[1], <MODE>mode))"
> "@
> @@ -1523,7 +1522,7 @@ (define_insn "*mov<mode>_aarch64"
> ldr\\t%x0, %1
> str\\t%x1, %0
> mov\\t%x0, %x1
> - mov\\t%x0, %1"
> + * return aarch64_zeroextended_move_imm (INTVAL (operands[1])) ? \"mov\\t%w0, %1\" : \"mov\\t%x0, %1\";"
> [(set_attr "type" "neon_move,f_mcr,f_mrc,fmov,fconstd,neon_move,\
> f_loadd,f_stored,load_8,store_8,mov_reg,\
> fconstd")
> diff --git a/gcc/config/aarch64/constraints.md b/gcc/config/aarch64/constraints.md
> index 29efb6c0cff7574c9b239ef358acaca96dd75d03..6168295814c3ce4b6f65030cddfd08172b6febd8 100644
> --- a/gcc/config/aarch64/constraints.md
> +++ b/gcc/config/aarch64/constraints.md
> @@ -106,6 +106,12 @@ (define_constraint "M"
>
> (define_constraint "N"
> "A constant that can be used with a 64-bit MOV immediate operation."
> + (and (match_code "const_int")
> + (match_test "aarch64_move_imm (ival, DImode)")
> + (match_test "!aarch64_zeroextended_move_imm (ival)")))
Sorry for not noticing last time, but: rather than have
aarch64_zeroextended_move_imm (which is quite a complicated test),
could we just add an extra (default off) parameter to aarch64_move_imm
that suppresses the (val >> 32) == 0 test?
Thanks,
Richard
> +
> +(define_constraint "O"
> + "A constant that can be used with a 32 or 64-bit MOV immediate operation."
> (and (match_code "const_int")
> (match_test "aarch64_move_imm (ival, DImode)")))
>
next prev parent reply other threads:[~2022-12-01 10:59 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-11-01 13:08 Wilco Dijkstra
2022-11-24 12:20 ` Richard Sandiford
2022-11-29 13:57 ` Wilco Dijkstra
2022-12-01 10:59 ` Richard Sandiford [this message]
2022-12-05 12:12 ` Wilco Dijkstra
2022-12-05 13:40 ` Richard Sandiford
2022-12-07 14:55 ` Andreas Schwab
2022-12-07 17:18 ` Wilco Dijkstra
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=mptzgc74cfo.fsf@arm.com \
--to=richard.sandiford@arm.com \
--cc=Wilco.Dijkstra@arm.com \
--cc=gcc-patches@gcc.gnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).