[PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

[PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

[Tamar Christina]

[-- Attachment #1: Type: text/plain, Size: 1323 bytes --]

Hi All, 

This patch adjusts the cost model so that when both sdiv and udiv are possible
it prefers udiv over sdiv. This was done by making sdiv slightly more expensive
instead of making udiv cheaper to keep the baseline costs of a division the same
as before.

Similar to aarch64 this patch along with my other two related mid-end changes
makes a big difference in division by constants.

Given:

int f2(int x)
{
  return ((x * x) % 300) + ((x * x) / 300);
}

we now generate

f2:
	mul	r3, r0, r0
	mov	r0, r3
	ldr	r1, .L3
	umull	r2, r3, r0, r1
	lsr	r2, r3, #5
	add	r3, r2, r2, lsl #2
	rsb	r3, r3, r3, lsl #4
	sub	r0, r0, r3, lsl #2
	add	r0, r0, r2
	bx	lr

as opposed to

f2:
	mul	r3, r0, r0
	mov	r0, r3
	ldr	r3, .L4
	push	{r4, r5}
	smull	r4, r5, r0, r3
	asr	r3, r0, #31
	rsb	r3, r3, r5, asr #5
	add	r2, r3, r3, lsl #2
	rsb	r2, r2, r2, lsl #4
	sub	r0, r0, r2, lsl #2
	add	r0, r0, r3
	pop	{r4, r5}
	bx	lr

Bootstrapped and reg tested on arm-none-eabi
with no regressions.

OK for trunk?

Thanks,
Tamar


gcc/
2017-05-02  Tamar Christina  <tamar.christina@arm.com>

	* config/arm/arm.c (arm_rtx_costs_internal): Make sdiv more expensive than udiv.


gcc/testsuite/
2017-05-02  Tamar Christina  <tamar.christina@arm.com>

	* gcc.target/arm/sdiv_costs_1.c: New.

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: sdiv-arm.patch --]
[-- Type: text/x-patch; name="sdiv-arm.patch", Size: 1911 bytes --]

diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
index b24143e32e2f100000f3b150f7ed0df4fabb3cc8..ecc7688b1db6309a4dd694a8e254e64abe14d7e3 100644
--- a/gcc/config/arm/arm.c
+++ b/gcc/config/arm/arm.c
@@ -9258,6 +9258,8 @@ arm_rtx_costs_internal (rtx x, enum rtx_code code, enum rtx_code outer_code,
 	*cost += COSTS_N_INSNS (speed_p ? extra_cost->mult[0].idiv : 0);
       else
 	*cost = LIBCALL_COST (2);
+
+      *cost += (code == DIV ? 1 : 0);
       return false;	/* All arguments must be in registers.  */
 
     case MOD:
@@ -9280,7 +9282,7 @@ arm_rtx_costs_internal (rtx x, enum rtx_code code, enum rtx_code outer_code,
 
     /* Fall-through.  */
     case UMOD:
-      *cost = LIBCALL_COST (2);
+      *cost = LIBCALL_COST (2) + (code == MOD ? 1 : 0);
       return false;	/* All arguments must be in registers.  */
 
     case ROTATE:
diff --git a/gcc/testsuite/gcc.target/arm/sdiv_costs_1.c b/gcc/testsuite/gcc.target/arm/sdiv_costs_1.c
new file mode 100644
index 0000000000000000000000000000000000000000..76086ab9ce28fceb37a4e8a615a38923fa7b985a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/sdiv_costs_1.c
@@ -0,0 +1,38 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 -march=armv8-a" } */
+
+/* Both sdiv and udiv can be used here, so prefer udiv.  */
+int f1 (unsigned char *p)
+{
+  return 100 / p[1];
+}
+
+int f2 (unsigned char *p, unsigned short x)
+{
+  return x / p[0];
+}
+
+int f3 (unsigned char *p, int x)
+{
+  x &= 0x7fffffff;
+  return x / p[0];
+}
+
+int f5 (unsigned char *p, unsigned short x)
+{
+  return x % p[0];
+}
+
+/* This should only generate signed divisions.  */
+int f4 (unsigned char *p)
+{
+  return -100 / p[1];
+}
+
+int f6 (unsigned char *p, short x)
+{
+  return x % p[0];
+}
+
+/* { dg-final { scan-assembler-times "udiv\tr\[0-9\]+, r\[0-9\]+, r\[0-9\]+" 4 } } */
+/* { dg-final { scan-assembler-times "sdiv\tr\[0-9\]+, r\[0-9\]+, r\[0-9\]+" 2 } } */

Re: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

[Tamar Christina]

Ping
________________________________________
From: gcc-patches-owner@gcc.gnu.org <gcc-patches-owner@gcc.gnu.org> on behalf of Tamar Christina <Tamar.Christina@arm.com>
Sent: Tuesday, May 2, 2017 4:37:12 PM
To: GCC Patches
Cc: nd; Kyrylo Tkachov; Ramana Radhakrishnan; Richard Earnshaw; nickc@redhat.com
Subject: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

Hi All,

This patch adjusts the cost model so that when both sdiv and udiv are possible
it prefers udiv over sdiv. This was done by making sdiv slightly more expensive
instead of making udiv cheaper to keep the baseline costs of a division the same
as before.

Similar to aarch64 this patch along with my other two related mid-end changes
makes a big difference in division by constants.

Given:

int f2(int x)
{
  return ((x * x) % 300) + ((x * x) / 300);
}

we now generate

f2:
        mul     r3, r0, r0
        mov     r0, r3
        ldr     r1, .L3
        umull   r2, r3, r0, r1
        lsr     r2, r3, #5
        add     r3, r2, r2, lsl #2
        rsb     r3, r3, r3, lsl #4
        sub     r0, r0, r3, lsl #2
        add     r0, r0, r2
        bx      lr

as opposed to

f2:
        mul     r3, r0, r0
        mov     r0, r3
        ldr     r3, .L4
        push    {r4, r5}
        smull   r4, r5, r0, r3
        asr     r3, r0, #31
        rsb     r3, r3, r5, asr #5
        add     r2, r3, r3, lsl #2
        rsb     r2, r2, r2, lsl #4
        sub     r0, r0, r2, lsl #2
        add     r0, r0, r3
        pop     {r4, r5}
        bx      lr

Bootstrapped and reg tested on arm-none-eabi
with no regressions.

OK for trunk?

Thanks,
Tamar


gcc/
2017-05-02  Tamar Christina  <tamar.christina@arm.com>

        * config/arm/arm.c (arm_rtx_costs_internal): Make sdiv more expensive than udiv.


gcc/testsuite/
2017-05-02  Tamar Christina  <tamar.christina@arm.com>

        * gcc.target/arm/sdiv_costs_1.c: New.

Re: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

[Tamar Christina]

Ping
________________________________________
From: gcc-patches-owner@gcc.gnu.org <gcc-patches-owner@gcc.gnu.org> on behalf of Tamar Christina <Tamar.Christina@arm.com>
Sent: Monday, May 15, 2017 9:32:55 AM
To: GCC Patches
Cc: nd; Kyrylo Tkachov; Ramana Radhakrishnan; Richard Earnshaw; nickc@redhat.com
Subject: Re: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

Ping
________________________________________
From: gcc-patches-owner@gcc.gnu.org <gcc-patches-owner@gcc.gnu.org> on behalf of Tamar Christina <Tamar.Christina@arm.com>
Sent: Tuesday, May 2, 2017 4:37:12 PM
To: GCC Patches
Cc: nd; Kyrylo Tkachov; Ramana Radhakrishnan; Richard Earnshaw; nickc@redhat.com
Subject: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

Hi All,

This patch adjusts the cost model so that when both sdiv and udiv are possible
it prefers udiv over sdiv. This was done by making sdiv slightly more expensive
instead of making udiv cheaper to keep the baseline costs of a division the same
as before.

Similar to aarch64 this patch along with my other two related mid-end changes
makes a big difference in division by constants.

Given:

int f2(int x)
{
  return ((x * x) % 300) + ((x * x) / 300);
}

we now generate

f2:
        mul     r3, r0, r0
        mov     r0, r3
        ldr     r1, .L3
        umull   r2, r3, r0, r1
        lsr     r2, r3, #5
        add     r3, r2, r2, lsl #2
        rsb     r3, r3, r3, lsl #4
        sub     r0, r0, r3, lsl #2
        add     r0, r0, r2
        bx      lr

as opposed to

f2:
        mul     r3, r0, r0
        mov     r0, r3
        ldr     r3, .L4
        push    {r4, r5}
        smull   r4, r5, r0, r3
        asr     r3, r0, #31
        rsb     r3, r3, r5, asr #5
        add     r2, r3, r3, lsl #2
        rsb     r2, r2, r2, lsl #4
        sub     r0, r0, r2, lsl #2
        add     r0, r0, r3
        pop     {r4, r5}
        bx      lr

Bootstrapped and reg tested on arm-none-eabi
with no regressions.

OK for trunk?

Thanks,
Tamar


gcc/
2017-05-02  Tamar Christina  <tamar.christina@arm.com>

        * config/arm/arm.c (arm_rtx_costs_internal): Make sdiv more expensive than udiv.


gcc/testsuite/
2017-05-02  Tamar Christina  <tamar.christina@arm.com>

        * gcc.target/arm/sdiv_costs_1.c: New.

Re: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

[Kyrill Tkachov]

On 02/05/17 16:37, Tamar Christina wrote:
> Hi All,
>
> This patch adjusts the cost model so that when both sdiv and udiv are possible
> it prefers udiv over sdiv. This was done by making sdiv slightly more expensive
> instead of making udiv cheaper to keep the baseline costs of a division the same
> as before.
>
> Similar to aarch64 this patch along with my other two related mid-end changes
> makes a big difference in division by constants.
>
> Given:
>
> int f2(int x)
> {
>   return ((x * x) % 300) + ((x * x) / 300);
> }
>
> we now generate
>
> f2:
>         mul     r3, r0, r0
>         mov     r0, r3
>         ldr     r1, .L3
>         umull   r2, r3, r0, r1
>         lsr     r2, r3, #5
>         add     r3, r2, r2, lsl #2
>         rsb     r3, r3, r3, lsl #4
>         sub     r0, r0, r3, lsl #2
>         add     r0, r0, r2
>         bx      lr
>
> as opposed to
>
> f2:
>         mul     r3, r0, r0
>         mov     r0, r3
>         ldr     r3, .L4
>         push    {r4, r5}
>         smull   r4, r5, r0, r3
>         asr     r3, r0, #31
>         rsb     r3, r3, r5, asr #5
>         add     r2, r3, r3, lsl #2
>         rsb     r2, r2, r2, lsl #4
>         sub     r0, r0, r2, lsl #2
>         add     r0, r0, r3
>         pop     {r4, r5}
>         bx      lr
>
> Bootstrapped and reg tested on arm-none-eabi
> with no regressions.
>
> OK for trunk?
>
> Thanks,
> Tamar
>
>
> gcc/
> 2017-05-02  Tamar Christina  <tamar.christina@arm.com>
>
>         * config/arm/arm.c (arm_rtx_costs_internal): Make sdiv more expensive than udiv.
>
>
> gcc/testsuite/
> 2017-05-02  Tamar Christina  <tamar.christina@arm.com>
>
>         * gcc.target/arm/sdiv_costs_1.c: New.

diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
index b24143e32e2f100000f3b150f7ed0df4fabb3cc8..ecc7688b1db6309a4dd694a8e254e64abe14d7e3 100644
--- a/gcc/config/arm/arm.c
+++ b/gcc/config/arm/arm.c
@@ -9258,6 +9258,8 @@ arm_rtx_costs_internal (rtx x, enum rtx_code code, enum rtx_code outer_code,
  	*cost += COSTS_N_INSNS (speed_p ? extra_cost->mult[0].idiv : 0);
        else
  	*cost = LIBCALL_COST (2);
+
+      *cost += (code == DIV ? 1 : 0);
        return false;	/* All arguments must be in registers.  */
  

We usually try to avoid adjusting the costs in units other than COSTS_N_INSNS.
Would adding COSTS_N_INSNS (1) here work?
If so, could you also add a comment here to describe why we're adjusting the cost.

      case MOD:
@@ -9280,7 +9282,7 @@ arm_rtx_costs_internal (rtx x, enum rtx_code code, enum rtx_code outer_code,
  
      /* Fall-through.  */
      case UMOD:
-      *cost = LIBCALL_COST (2);
+      *cost = LIBCALL_COST (2) + (code == MOD ? 1 : 0);

Same here.

Thanks,
Kyrill

RE: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

[Tamar Christina]

Hi Kyrill,

> diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c index
> b24143e32e2f100000f3b150f7ed0df4fabb3cc8..ecc7688b1db6309a4dd694a8e
> 254e64abe14d7e3 100644
> --- a/gcc/config/arm/arm.c
> +++ b/gcc/config/arm/arm.c
> @@ -9258,6 +9258,8 @@ arm_rtx_costs_internal (rtx x, enum rtx_code
> code, enum rtx_code outer_code,
>   	*cost += COSTS_N_INSNS (speed_p ? extra_cost->mult[0].idiv : 0);
>         else
>   	*cost = LIBCALL_COST (2);
> +
> +      *cost += (code == DIV ? 1 : 0);
>         return false;	/* All arguments must be in registers.  */
> 
> 
> We usually try to avoid adjusting the costs in units other than
> COSTS_N_INSNS.
> Would adding COSTS_N_INSNS (1) here work?
> If so, could you also add a comment here to describe why we're adjusting the
> cost.

It would, I'm just slightly worried it might end up generating different code for DIV then.
The reason I have used a unit smaller than COSTS_N_INSNS it so that it should have any real
Impact on any other optimization as the cost is likely treated as an integer? It's only for things that
Compare the costs values between signed and unsigned would the small unit make a difference.

Since I think the compiler still has some hard coded cost limits somewhere it may be an issue, but I'm not
100% certain. I can make the change though.

> 
>       case MOD:
> @@ -9280,7 +9282,7 @@ arm_rtx_costs_internal (rtx x, enum rtx_code
> code, enum rtx_code outer_code,
> 
>       /* Fall-through.  */
>       case UMOD:
> -      *cost = LIBCALL_COST (2);
> +      *cost = LIBCALL_COST (2) + (code == MOD ? 1 : 0);
> 
> Same here.
> 
> Thanks,
> Kyrill
> 

Re: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

[Tamar Christina]

[-- Attachment #1: Type: text/plain, Size: 2111 bytes --]

Hi Kyrill,

I have updated the patch and regtested on arm-none-linux-gnueabihf.

OK for trunk?

Thanks,
Tamar
________________________________________
From: Tamar Christina
Sent: Wednesday, June 7, 2017 11:15:49 AM
To: Kyrill Tkachov; GCC Patches
Cc: nd; Ramana Radhakrishnan; Richard Earnshaw; nickc@redhat.com
Subject: RE: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

Hi Kyrill,

> diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c index
> b24143e32e2f100000f3b150f7ed0df4fabb3cc8..ecc7688b1db6309a4dd694a8e
> 254e64abe14d7e3 100644
> --- a/gcc/config/arm/arm.c
> +++ b/gcc/config/arm/arm.c
> @@ -9258,6 +9258,8 @@ arm_rtx_costs_internal (rtx x, enum rtx_code
> code, enum rtx_code outer_code,
>       *cost += COSTS_N_INSNS (speed_p ? extra_cost->mult[0].idiv : 0);
>         else
>       *cost = LIBCALL_COST (2);
> +
> +      *cost += (code == DIV ? 1 : 0);
>         return false; /* All arguments must be in registers.  */
>
>
> We usually try to avoid adjusting the costs in units other than
> COSTS_N_INSNS.
> Would adding COSTS_N_INSNS (1) here work?
> If so, could you also add a comment here to describe why we're adjusting the
> cost.

It would, I'm just slightly worried it might end up generating different code for DIV then.
The reason I have used a unit smaller than COSTS_N_INSNS it so that it should have any real
Impact on any other optimization as the cost is likely treated as an integer? It's only for things that
Compare the costs values between signed and unsigned would the small unit make a difference.

Since I think the compiler still has some hard coded cost limits somewhere it may be an issue, but I'm not
100% certain. I can make the change though.

>
>       case MOD:
> @@ -9280,7 +9282,7 @@ arm_rtx_costs_internal (rtx x, enum rtx_code
> code, enum rtx_code outer_code,
>
>       /* Fall-through.  */
>       case UMOD:
> -      *cost = LIBCALL_COST (2);
> +      *cost = LIBCALL_COST (2) + (code == MOD ? 1 : 0);
>
> Same here.
>
> Thanks,
> Kyrill
>


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: arm-div-costs.patch --]
[-- Type: text/x-patch; name="arm-div-costs.patch", Size: 2167 bytes --]

diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
index b24143e32e2f100000f3b150f7ed0df4fabb3cc8..442d12de4dcff50484229e2d27e65d78c3fd6b37 100644
--- a/gcc/config/arm/arm.c
+++ b/gcc/config/arm/arm.c
@@ -9258,6 +9258,10 @@ arm_rtx_costs_internal (rtx x, enum rtx_code code, enum rtx_code outer_code,
 	*cost += COSTS_N_INSNS (speed_p ? extra_cost->mult[0].idiv : 0);
       else
 	*cost = LIBCALL_COST (2);
+
+      /* Make the cost of sdiv more expensive so when both sdiv and udiv are
+	 possible udiv is prefered.  */
+      *cost += (code == DIV ? COSTS_N_INSNS (1) : 0);
       return false;	/* All arguments must be in registers.  */
 
     case MOD:
@@ -9280,7 +9284,9 @@ arm_rtx_costs_internal (rtx x, enum rtx_code code, enum rtx_code outer_code,
 
     /* Fall-through.  */
     case UMOD:
-      *cost = LIBCALL_COST (2);
+      /* Make the cost of sdiv more expensive so when both sdiv and udiv are
+	 possible udiv is prefered.  */
+      *cost = LIBCALL_COST (2) + (code == MOD ? COSTS_N_INSNS (1) : 0);
       return false;	/* All arguments must be in registers.  */
 
     case ROTATE:
diff --git a/gcc/testsuite/gcc.target/arm/sdiv_costs_1.c b/gcc/testsuite/gcc.target/arm/sdiv_costs_1.c
new file mode 100644
index 0000000000000000000000000000000000000000..76086ab9ce28fceb37a4e8a615a38923fa7b985a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/sdiv_costs_1.c
@@ -0,0 +1,38 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 -march=armv8-a" } */
+
+/* Both sdiv and udiv can be used here, so prefer udiv.  */
+int f1 (unsigned char *p)
+{
+  return 100 / p[1];
+}
+
+int f2 (unsigned char *p, unsigned short x)
+{
+  return x / p[0];
+}
+
+int f3 (unsigned char *p, int x)
+{
+  x &= 0x7fffffff;
+  return x / p[0];
+}
+
+int f5 (unsigned char *p, unsigned short x)
+{
+  return x % p[0];
+}
+
+/* This should only generate signed divisions.  */
+int f4 (unsigned char *p)
+{
+  return -100 / p[1];
+}
+
+int f6 (unsigned char *p, short x)
+{
+  return x % p[0];
+}
+
+/* { dg-final { scan-assembler-times "udiv\tr\[0-9\]+, r\[0-9\]+, r\[0-9\]+" 4 } } */
+/* { dg-final { scan-assembler-times "sdiv\tr\[0-9\]+, r\[0-9\]+, r\[0-9\]+" 2 } } */

Re: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]

[Kyrill Tkachov]

Hi Tamar,

On 07/06/17 16:09, Tamar Christina wrote:
> Hi Kyrill,
>
> I have updated the patch and regtested on arm-none-linux-gnueabihf.
>
> OK for trunk?

Ok.
Thanks,
Kyrill

> Thanks,
> Tamar
> ________________________________________
> From: Tamar Christina
> Sent: Wednesday, June 7, 2017 11:15:49 AM
> To: Kyrill Tkachov; GCC Patches
> Cc: nd; Ramana Radhakrishnan; Richard Earnshaw; nickc@redhat.com
> Subject: RE: [PATCH][GCC][ARM] Adjust costs so udiv is preferred over sdiv when both are valid. [Patch (2/2)]
>
> Hi Kyrill,
>
>> diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c index
>> b24143e32e2f100000f3b150f7ed0df4fabb3cc8..ecc7688b1db6309a4dd694a8e
>> 254e64abe14d7e3 100644
>> --- a/gcc/config/arm/arm.c
>> +++ b/gcc/config/arm/arm.c
>> @@ -9258,6 +9258,8 @@ arm_rtx_costs_internal (rtx x, enum rtx_code
>> code, enum rtx_code outer_code,
>>        *cost += COSTS_N_INSNS (speed_p ? extra_cost->mult[0].idiv : 0);
>>          else
>>        *cost = LIBCALL_COST (2);
>> +
>> +      *cost += (code == DIV ? 1 : 0);
>>          return false; /* All arguments must be in registers.  */
>>
>>
>> We usually try to avoid adjusting the costs in units other than
>> COSTS_N_INSNS.
>> Would adding COSTS_N_INSNS (1) here work?
>> If so, could you also add a comment here to describe why we're adjusting the
>> cost.
> It would, I'm just slightly worried it might end up generating different code for DIV then.
> The reason I have used a unit smaller than COSTS_N_INSNS it so that it should have any real
> Impact on any other optimization as the cost is likely treated as an integer? It's only for things that
> Compare the costs values between signed and unsigned would the small unit make a difference.
>
> Since I think the compiler still has some hard coded cost limits somewhere it may be an issue, but I'm not
> 100% certain. I can make the change though.
>
>>        case MOD:
>> @@ -9280,7 +9282,7 @@ arm_rtx_costs_internal (rtx x, enum rtx_code
>> code, enum rtx_code outer_code,
>>
>>        /* Fall-through.  */
>>        case UMOD:
>> -      *cost = LIBCALL_COST (2);
>> +      *cost = LIBCALL_COST (2) + (code == MOD ? 1 : 0);
>>
>> Same here.
>>
>> Thanks,
>> Kyrill
>>