diff mbox

[i386,AVX512,8/n] Extend substs for new patterns.

Message ID 20140813142335.GC57184@msticlxl57.ims.intel.com
State New
Headers show

Commit Message

Kirill Yukhin Aug. 13, 2014, 2:23 p.m. UTC 
On 13 Aug 15:21, Uros Bizjak wrote:
> On Wed, Aug 13, 2014 at 2:14 PM, Kirill Yukhin <kirill.yukhin@gmail.com> wrote:
> 
> >> In a couple of places, <MODE_SIZE> checks are changed to GET_MODE_SIZE
> >> (GET_MODE (operands[0])) and in a similar way, <MODE>mode checks were
> >> changed to GET_MODE (operands[0]). The previous checks are more
> >> efficient and are preferred. Is there a reason you need to check
> >> operands[0] dynamically?
> >
> > I agree, runtime tests are worse. But this subst attributes are used
> > in patterns which actually do not contain mode iterators at all.
> 
> IMO, two different subst patterns should be used in this case, one for
> patterns with and one for patterns w/o mode iterators.

That was my second thought after I hit `send'.

Updated patch in the bottom

Bootstrapped.

(Regtest w/ new tests is in progress).

Is it ok if pass?

gcc/
	* config/i386/sse.md: Allow V64QI, V32QI, V32HI, V4HI modes.
	* config/i386/subst.md
	(define_mode_iterator SUBST_V): Update.
	(define_mode_iterator SUBST_A): Ditto.
	(define_subst_attr "mask_operand7"): New.
	(define_subst_attr "mask_operand10"): New.
	(define_subst_attr "mask_operand_arg34") : New.
	(define_subst_attr "mask_expand_op3"): New.
	(define_subst_attr "mask_mode512bit_condition"): Handle TARGET_AVX512VL.
	(define_subst_attr "sd_mask_mode512bit_condition"): Ditto.
	(define_subst_attr "mask_avx512vl_condition"): New.
	(define_subst_attr "round_mask_operand4"): Ditto.
	(define_subst_attr "round_mask_scalar_op3"): Delete.
	(define_subst_attr "round_mask_op4"): New.
	(define_subst_attr "round_mode512bit_condition"): Allow V8DImode,
	V16SImode.
	(define_subst_attr "round_modev8sf_condition"): New.
	(define_subst_attr "round_modev4sf_condition"): GET_MODE instead of
	<MODE>mode.
	(define_subst_attr "round_saeonly_mask_operand4"): New.
	(define_subst_attr "round_saeonly_mask_op4"): New.
	(define_subst_attr "round_saeonly_mode512bit_condition"): Allow
	V8DImode, V16SImode.
	(define_subst_attr "round_saeonly_modev8sf_condition"): New.
	(define_subst_attr "mask_expand4_name" "mask_expand4"): New.
	(define_subst_attr "mask_expand4_args"): New.
	(define_subst "mask_expand4"): New.

--
Thanks, K

Comments

Uros Bizjak Aug. 13, 2014, 5:40 p.m. UTC | #1 
On Wed, Aug 13, 2014 at 4:23 PM, Kirill Yukhin <kirill.yukhin@gmail.com> wrote:

>> >> In a couple of places, <MODE_SIZE> checks are changed to GET_MODE_SIZE
>> >> (GET_MODE (operands[0])) and in a similar way, <MODE>mode checks were
>> >> changed to GET_MODE (operands[0]). The previous checks are more
>> >> efficient and are preferred. Is there a reason you need to check
>> >> operands[0] dynamically?
>> >
>> > I agree, runtime tests are worse. But this subst attributes are used
>> > in patterns which actually do not contain mode iterators at all.
>>
>> IMO, two different subst patterns should be used in this case, one for
>> patterns with and one for patterns w/o mode iterators.
>
> That was my second thought after I hit `send'.
>
> Updated patch in the bottom
>
> Bootstrapped.
>
> (Regtest w/ new tests is in progress).
>
> Is it ok if pass?
>
> gcc/
>         * config/i386/sse.md: Allow V64QI, V32QI, V32HI, V4HI modes.
>         * config/i386/subst.md
>         (define_mode_iterator SUBST_V): Update.
>         (define_mode_iterator SUBST_A): Ditto.
>         (define_subst_attr "mask_operand7"): New.
>         (define_subst_attr "mask_operand10"): New.
>         (define_subst_attr "mask_operand_arg34") : New.
>         (define_subst_attr "mask_expand_op3"): New.
>         (define_subst_attr "mask_mode512bit_condition"): Handle TARGET_AVX512VL.
>         (define_subst_attr "sd_mask_mode512bit_condition"): Ditto.
>         (define_subst_attr "mask_avx512vl_condition"): New.
>         (define_subst_attr "round_mask_operand4"): Ditto.
>         (define_subst_attr "round_mask_scalar_op3"): Delete.
>         (define_subst_attr "round_mask_op4"): New.
>         (define_subst_attr "round_mode512bit_condition"): Allow V8DImode,
>         V16SImode.
>         (define_subst_attr "round_modev8sf_condition"): New.
>         (define_subst_attr "round_modev4sf_condition"): GET_MODE instead of
>         <MODE>mode.
>         (define_subst_attr "round_saeonly_mask_operand4"): New.
>         (define_subst_attr "round_saeonly_mask_op4"): New.
>         (define_subst_attr "round_saeonly_mode512bit_condition"): Allow
>         V8DImode, V16SImode.
>         (define_subst_attr "round_saeonly_modev8sf_condition"): New.
>         (define_subst_attr "mask_expand4_name" "mask_expand4"): New.
>         (define_subst_attr "mask_expand4_args"): New.
>         (define_subst "mask_expand4"): New.

This is OK.

Thanks,
Uros.
diff mbox

Patch

diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index 3337104..ebe38f3 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -471,8 +471,8 @@ 
 
 ;; Mapping of vector modes to corresponding mask size
 (define_mode_attr avx512fmaskmode
-  [(V16QI "HI")
-   (V16HI "HI") (V8HI  "QI")
+  [(V64QI "DI") (V32QI "SI") (V16QI "HI")
+   (V32HI "SI") (V16HI "HI") (V8HI  "QI") (V4HI "QI")
    (V16SI "HI") (V8SI  "QI") (V4SI  "QI")
    (V8DI  "QI") (V4DI  "QI") (V2DI  "QI")
    (V16SF "HI") (V8SF  "QI") (V4SF  "QI")
diff --git a/gcc/config/i386/subst.md b/gcc/config/i386/subst.md
index 1654cba..edb0965 100644
--- a/gcc/config/i386/subst.md
+++ b/gcc/config/i386/subst.md
@@ -20,8 +20,8 @@ 
 ;; Some iterators for extending subst as much as possible
 ;; All vectors (Use it for destination)
 (define_mode_iterator SUBST_V
-  [V16QI
-   V16HI V8HI
+  [V64QI V32QI V16QI
+   V32HI V16HI V8HI
    V16SI V8SI  V4SI
    V8DI  V4DI  V2DI
    V16SF V8SF  V4SF
@@ -31,8 +31,8 @@ 
   [QI HI SI DI])
 
 (define_mode_iterator SUBST_A
-  [V16QI
-   V16HI V8HI
+  [V64QI V32QI V16QI
+   V32HI V16HI V8HI
    V16SI V8SI  V4SI
    V8DI  V4DI  V2DI
    V16SF V8SF  V4SF
@@ -47,16 +47,21 @@ 
 (define_subst_attr "mask_operand3_1" "mask" "" "%%{%%4%%}%%N3") ;; for sprintf
 (define_subst_attr "mask_operand4" "mask" "" "%{%5%}%N4")
 (define_subst_attr "mask_operand6" "mask" "" "%{%7%}%N6")
+(define_subst_attr "mask_operand7" "mask" "" "%{%8%}%N7")
+(define_subst_attr "mask_operand10" "mask" "" "%{%11%}%N10")
 (define_subst_attr "mask_operand11" "mask" "" "%{%12%}%N11")
 (define_subst_attr "mask_operand18" "mask" "" "%{%19%}%N18")
 (define_subst_attr "mask_operand19" "mask" "" "%{%20%}%N19")
 (define_subst_attr "mask_codefor" "mask" "*" "")
-(define_subst_attr "mask_mode512bit_condition" "mask" "1" "(<MODE_SIZE> == 64)")
+(define_subst_attr "mask_operand_arg34" "mask" "" ", operands[3], operands[4]")
+(define_subst_attr "mask_mode512bit_condition" "mask" "1" "(<MODE_SIZE> == 64 || TARGET_AVX512VL)")
+(define_subst_attr "mask_avx512vl_condition" "mask" "1" "TARGET_AVX512VL")
 (define_subst_attr "store_mask_constraint" "mask" "vm" "v")
 (define_subst_attr "store_mask_predicate" "mask" "nonimmediate_operand" "register_operand")
 (define_subst_attr "mask_prefix" "mask" "vex" "evex")
 (define_subst_attr "mask_prefix2" "mask" "maybe_vex" "evex")
 (define_subst_attr "mask_prefix3" "mask" "orig,vex" "evex")
+(define_subst_attr "mask_expand_op3" "mask" "3" "5")
 
 (define_subst "mask"
   [(set (match_operand:SUBST_V 0)
@@ -85,7 +90,7 @@ 
 (define_subst_attr "sd_mask_op4" "sd" "" "%{%5%}%N4")
 (define_subst_attr "sd_mask_op5" "sd" "" "%{%6%}%N5")
 (define_subst_attr "sd_mask_codefor" "sd" "*" "")
-(define_subst_attr "sd_mask_mode512bit_condition" "sd" "1" "(<MODE_SIZE> == 64)")
+(define_subst_attr "sd_mask_mode512bit_condition" "sd" "1" "(<MODE_SIZE> == 64 || TARGET_AVX512VL)")
 
 (define_subst "sd"
  [(set (match_operand:SUBST_V 0)
@@ -101,6 +106,7 @@ 
 (define_subst_attr "round_name" "round" "" "_round")
 (define_subst_attr "round_mask_operand2" "mask" "%R2" "%R4")
 (define_subst_attr "round_mask_operand3" "mask" "%R3" "%R5")
+(define_subst_attr "round_mask_operand4" "mask" "%R4" "%R6")
 (define_subst_attr "round_sd_mask_operand4" "sd" "%R4" "%R6")
 (define_subst_attr "round_op2" "round" "" "%R2")
 (define_subst_attr "round_op3" "round" "" "%R3")
@@ -109,14 +115,18 @@ 
 (define_subst_attr "round_op6" "round" "" "%R6")
 (define_subst_attr "round_mask_op2" "round" "" "<round_mask_operand2>")
 (define_subst_attr "round_mask_op3" "round" "" "<round_mask_operand3>")
-(define_subst_attr "round_mask_scalar_op3" "round" "" "<round_mask_scalar_operand3>")
+(define_subst_attr "round_mask_op4" "round" "" "<round_mask_operand4>")
 (define_subst_attr "round_sd_mask_op4" "round" "" "<round_sd_mask_operand4>")
 (define_subst_attr "round_constraint" "round" "vm" "v")
 (define_subst_attr "round_constraint2" "round" "m" "v")
 (define_subst_attr "round_constraint3" "round" "rm" "r")
 (define_subst_attr "round_nimm_predicate" "round" "nonimmediate_operand" "register_operand")
 (define_subst_attr "round_prefix" "round" "vex" "evex")
-(define_subst_attr "round_mode512bit_condition" "round" "1" "(<MODE>mode == V16SFmode || <MODE>mode == V8DFmode)")
+(define_subst_attr "round_mode512bit_condition" "round" "1" "(<MODE>mode == V16SFmode
+							      || <MODE>mode == V8DFmode
+							      || <MODE>mode == V8DImode
+							      || <MODE>mode == V16SImode)")
+(define_subst_attr "round_modev8sf_condition" "round" "1" "(<MODE>mode == V8SFmode)")
 (define_subst_attr "round_modev4sf_condition" "round" "1" "(<MODE>mode == V4SFmode)")
 (define_subst_attr "round_codefor" "round" "*" "")
 (define_subst_attr "round_opnum" "round" "5" "6")
@@ -133,6 +143,7 @@ 
 (define_subst_attr "round_saeonly_name" "round_saeonly" "" "_round")
 (define_subst_attr "round_saeonly_mask_operand2" "mask" "%r2" "%r4")
 (define_subst_attr "round_saeonly_mask_operand3" "mask" "%r3" "%r5")
+(define_subst_attr "round_saeonly_mask_operand4" "mask" "%r4" "%r6")
 (define_subst_attr "round_saeonly_mask_scalar_merge_operand4" "mask_scalar_merge" "%r4" "%r5")
 (define_subst_attr "round_saeonly_sd_mask_operand5" "sd" "%r5" "%r7")
 (define_subst_attr "round_saeonly_op2" "round_saeonly" "" "%r2")
@@ -143,12 +154,17 @@ 
 (define_subst_attr "round_saeonly_prefix" "round_saeonly" "vex" "evex")
 (define_subst_attr "round_saeonly_mask_op2" "round_saeonly" "" "<round_saeonly_mask_operand2>")
 (define_subst_attr "round_saeonly_mask_op3" "round_saeonly" "" "<round_saeonly_mask_operand3>")
+(define_subst_attr "round_saeonly_mask_op4" "round_saeonly" "" "<round_saeonly_mask_operand4>")
 (define_subst_attr "round_saeonly_mask_scalar_merge_op4" "round_saeonly" "" "<round_saeonly_mask_scalar_merge_operand4>")
 (define_subst_attr "round_saeonly_sd_mask_op5" "round_saeonly" "" "<round_saeonly_sd_mask_operand5>")
 (define_subst_attr "round_saeonly_constraint" "round_saeonly" "vm" "v")
 (define_subst_attr "round_saeonly_constraint2" "round_saeonly" "m" "v")
 (define_subst_attr "round_saeonly_nimm_predicate" "round_saeonly" "nonimmediate_operand" "register_operand")
-(define_subst_attr "round_saeonly_mode512bit_condition" "round_saeonly" "1" "(<MODE>mode == V16SFmode || <MODE>mode == V8DFmode)")
+(define_subst_attr "round_saeonly_mode512bit_condition" "round_saeonly" "1" "(<MODE>mode == V16SFmode
+									      || <MODE>mode == V8DFmode
+									      || <MODE>mode == V8DImode
+									      || <MODE>mode == V16SImode)")
+(define_subst_attr "round_saeonly_modev8sf_condition" "round_saeonly" "1" "(<MODE>mode == V8SFmode)")
 
 (define_subst "round_saeonly"
   [(set (match_operand:SUBST_A 0)
@@ -196,3 +212,19 @@ 
    (match_dup 4)
    (match_dup 5)
    (unspec [(match_operand:SI 6 "const48_operand")] UNSPEC_EMBEDDED_ROUNDING)])
+
+(define_subst_attr "mask_expand4_name" "mask_expand4" "" "_mask")
+(define_subst_attr "mask_expand4_args" "mask_expand4" "" ", operands[4], operands[5]")
+
+(define_subst "mask_expand4"
+  [(match_operand:SUBST_V 0)
+   (match_operand:SUBST_V 1)
+   (match_operand:SUBST_V 2)
+   (match_operand:SI 3)]
+   "TARGET_AVX512VL"
+   [(match_dup 0)
+    (match_dup 1)
+    (match_dup 2)
+    (match_dup 3)
+    (match_operand:SUBST_V 4 "vector_move_operand")
+    (match_operand:<avx512fmaskmode> 5 "register_operand")])