[PATCH][ARM] use vsel instruction for floating point conditional moves in ARMv8

[PATCH][ARM] use vsel instruction for floating point conditional moves in ARMv8

[Kyrylo Tkachov]

[-- Attachment #1: Type: text/plain, Size: 1759 bytes --]

Hi all,
This patch uses the new ARMv8 AArch32 vsel instruction to implement
conditional moves of floating point numbers.
For example, an instruction of the form:
	vsel<cond>.f32  s0, s1, s2
means
	s0 := cond ? s1 : s2

This can be useful, among other places, in Thumb2 because it doesn't require
an enclosing IT block.
A small catch: The condition code used in vsel can only be one of {GE, GT,
EQ, VS}. If we want to use their negations {LT, LE, NE, VC} we just flip the
source operands.
A new predicate is introduced that checks that the comparison yields an ARM
condition code in the set {GE, GT, EQ, VS, LT, LE, NE, VC}.

New compilation tests are added. They pass on a model and no new regressions
on arm-none-eabi with qemu.

Ok for trunk?

Thanks,
Kyrill

gcc/ChangeLog

2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>

	* config/arm/arm.md (f_sels, f_seld): New types.
	(*cmov<mode>): New pattern.
	* config/arm/predicates.md (arm_vsel_comparison_operator): New
predicate.	


gcc/testsuite/ChangeLog

2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>

	* gcc.target/arm/vseleqdf.c: New test.
	* gcc.target/arm/vseleqsf.c: Likewise.
	* gcc.target/arm/vselgedf.c: Likewise.
	* gcc.target/arm/vselgesf.c: Likewise.
	* gcc.target/arm/vselgtdf.c: Likewise.
	* gcc.target/arm/vselgtsf.c: Likewise.
	* gcc.target/arm/vselledf.c: Likewise.
	* gcc.target/arm/vsellesf.c: Likewise.
	* gcc.target/arm/vselltdf.c: Likewise.
	* gcc.target/arm/vselltsf.c: Likewise.
	* gcc.target/arm/vselnedf.c: Likewise.
	* gcc.target/arm/vselnesf.c: Likewise.
	* gcc.target/arm/vselvcdf.c: Likewise.
	* gcc.target/arm/vselvcsf.c: Likewise.
	* gcc.target/arm/vselvsdf.c: Likewise.
	* gcc.target/arm/vselvssf.c: Likewise.

[-- Attachment #2: vsel_full.txt --]
[-- Type: text/plain, Size: 11281 bytes --]

diff --git a/gcc/config/arm/arm.md b/gcc/config/arm/arm.md
index ac507ef..a6bd8bc 100644
--- a/gcc/config/arm/arm.md
+++ b/gcc/config/arm/arm.md
@@ -388,6 +388,8 @@
   f_2_r,\
   r_2_f,\
   f_cvt,\
+  f_sels,\
+  f_seld,\
   branch,\
   call,\
   load_byte,\
@@ -8140,6 +8142,39 @@
   }"
 )
 
+(define_insn "*cmov<mode>"
+    [(set (match_operand:SDF 0 "s_register_operand" "=<F_constraint>")
+	(if_then_else:SDF (match_operator 1 "arm_vsel_comparison_operator"
+			  [(match_operand 2 "cc_register" "") (const_int 0)])
+			  (match_operand:SDF 3 "s_register_operand"
+			                      "<F_constraint>")
+			  (match_operand:SDF 4 "s_register_operand"
+			                      "<F_constraint>")))]
+  "TARGET_HARD_FLOAT && TARGET_FPU_ARMV8 <vfp_double_cond>"
+  "*
+  {
+    enum arm_cond_code code = maybe_get_arm_condition_code (operands[1]);
+    switch (code)
+      {
+      case ARM_GE:
+      case ARM_GT:
+      case ARM_EQ:
+      case ARM_VS:
+        return \"vsel%d1.<V_if_elem>\\t%<V_reg>0, %<V_reg>3, %<V_reg>4\";
+      case ARM_LT:
+      case ARM_LE:
+      case ARM_NE:
+      case ARM_VC:
+        return \"vsel%D1.<V_if_elem>\\t%<V_reg>0, %<V_reg>4, %<V_reg>3\";
+      default:
+        gcc_unreachable ();
+      }
+    return \"\";
+  }"
+  [(set_attr "conds" "use")
+   (set_attr "type" "f_sel<vfp_type>")]
+)
+
 (define_insn "*movsicc_insn"
   [(set (match_operand:SI 0 "s_register_operand" "=r,r,r,r,r,r,r,r")
 	(if_then_else:SI
diff --git a/gcc/config/arm/predicates.md b/gcc/config/arm/predicates.md
index 8f49450..898f558 100644
--- a/gcc/config/arm/predicates.md
+++ b/gcc/config/arm/predicates.md
@@ -270,6 +270,18 @@
 (define_special_predicate "lt_ge_comparison_operator"
   (match_code "lt,ge"))
 
+;; The vsel instruction only accepts the ARM condition codes listed below.
+(define_special_predicate "arm_vsel_comparison_operator"
+  (and (match_operand 0 "expandable_comparison_operator")
+       (match_test "maybe_get_arm_condition_code (op) == ARM_GE
+                    || maybe_get_arm_condition_code (op) == ARM_GT
+                    || maybe_get_arm_condition_code (op) == ARM_EQ
+                    || maybe_get_arm_condition_code (op) == ARM_VS
+                    || maybe_get_arm_condition_code (op) == ARM_LT
+                    || maybe_get_arm_condition_code (op) == ARM_LE
+                    || maybe_get_arm_condition_code (op) == ARM_NE
+                    || maybe_get_arm_condition_code (op) == ARM_VC")))
+
 (define_special_predicate "noov_comparison_operator"
   (match_code "lt,ge,eq,ne"))
 
diff --git a/gcc/testsuite/gcc.target/arm/vseleqdf.c b/gcc/testsuite/gcc.target/arm/vseleqdf.c
new file mode 100644
index 0000000..86e147b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vseleqdf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+double
+foo (double x, double y)
+{
+  volatile int i = 0;
+  return i == 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vseleq.f64\td\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vseleqsf.c b/gcc/testsuite/gcc.target/arm/vseleqsf.c
new file mode 100644
index 0000000..120f44b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vseleqsf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+float
+foo (float x, float y)
+{
+  volatile int i = 0;
+  return i == 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vseleq.f32\ts\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselgedf.c b/gcc/testsuite/gcc.target/arm/vselgedf.c
new file mode 100644
index 0000000..cea08d1
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselgedf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+double
+foo (double x, double y)
+{
+  volatile int i = 0;
+  return i >= 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselge.f64\td\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselgesf.c b/gcc/testsuite/gcc.target/arm/vselgesf.c
new file mode 100644
index 0000000..86f2a04
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselgesf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+float
+foo (float x, float y)
+{
+  volatile int i = 0;
+  return i >= 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselge.f32\ts\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselgtdf.c b/gcc/testsuite/gcc.target/arm/vselgtdf.c
new file mode 100644
index 0000000..2c4a6ba
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselgtdf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+double
+foo (double x, double y)
+{
+  volatile int i = 0;
+  return i > 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselgt.f64\td\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselgtsf.c b/gcc/testsuite/gcc.target/arm/vselgtsf.c
new file mode 100644
index 0000000..388e74c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselgtsf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+float
+foo (float x, float y)
+{
+  volatile int i = 0;
+  return i > 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselgt.f32\ts\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselledf.c b/gcc/testsuite/gcc.target/arm/vselledf.c
new file mode 100644
index 0000000..088dc04
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselledf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+double
+foo (double x, double y)
+{
+  volatile int i = 0;
+  return i <= 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselgt.f64\td\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vsellesf.c b/gcc/testsuite/gcc.target/arm/vsellesf.c
new file mode 100644
index 0000000..d0afdbc
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vsellesf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+float
+foo (float x, float y)
+{
+  volatile int i = 0;
+  return i <= 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselgt.f32\ts\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselltdf.c b/gcc/testsuite/gcc.target/arm/vselltdf.c
new file mode 100644
index 0000000..fbcb9ea
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselltdf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+double
+foo (double x, double y)
+{
+  volatile int i = 0;
+  return i < 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselge.f64\td\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselltsf.c b/gcc/testsuite/gcc.target/arm/vselltsf.c
new file mode 100644
index 0000000..959dab7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselltsf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+float
+foo (float x, float y)
+{
+  volatile int i = 0;
+  return i < 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselge.f32\ts\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselnedf.c b/gcc/testsuite/gcc.target/arm/vselnedf.c
new file mode 100644
index 0000000..cf67f29
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselnedf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+double
+foo (double x, double y)
+{
+  volatile int i = 0;
+  return i != 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vseleq.f64\td\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselnesf.c b/gcc/testsuite/gcc.target/arm/vselnesf.c
new file mode 100644
index 0000000..2e16423
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselnesf.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+float
+foo (float x, float y)
+{
+  volatile int i = 0;
+  return i != 0 ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vseleq.f32\ts\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselvcdf.c b/gcc/testsuite/gcc.target/arm/vselvcdf.c
new file mode 100644
index 0000000..7f30270
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselvcdf.c
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+double
+foo (double x, double y)
+{
+  return !__builtin_isunordered (x, y) ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselvs.f64\td\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselvcsf.c b/gcc/testsuite/gcc.target/arm/vselvcsf.c
new file mode 100644
index 0000000..1bb7369
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselvcsf.c
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+float
+foo (float x, float y)
+{
+  return !__builtin_isunordered (x, y) ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselvs.f32\ts\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselvsdf.c b/gcc/testsuite/gcc.target/arm/vselvsdf.c
new file mode 100644
index 0000000..83ad5bf
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselvsdf.c
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+double
+foo (double x, double y)
+{
+  return __builtin_isunordered (x, y) ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselvs.f64\td\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/vselvssf.c b/gcc/testsuite/gcc.target/arm/vselvssf.c
new file mode 100644
index 0000000..7d76289
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/vselvssf.c
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2" } */
+/* { dg-add-options arm_v8_vfp } */
+
+float
+foo (float x, float y)
+{
+  return __builtin_isunordered (x, y) ? x : y;
+}
+
+/* { dg-final { scan-assembler-times "vselvs.f32\ts\[0-9\]+" 1 } } */

RE: [PATCH][ARM] use vsel instruction for floating point conditional moves in ARMv8

[Kyrylo Tkachov]

Ping
http://gcc.gnu.org/ml/gcc-patches/2013-01/msg01418.html

Probably for 4.9 now...

Thanks,
Kyrill

> -----Original Message-----
> From: gcc-patches-owner@gcc.gnu.org [mailto:gcc-patches-
> owner@gcc.gnu.org] On Behalf Of Kyrylo Tkachov
> Sent: 30 January 2013 09:25
> To: gcc-patches@gcc.gnu.org
> Cc: Ramana Radhakrishnan; Richard Earnshaw
> Subject: [PATCH][ARM] use vsel instruction for floating point
> conditional moves in ARMv8
> 
> Hi all,
> This patch uses the new ARMv8 AArch32 vsel instruction to implement
> conditional moves of floating point numbers.
> For example, an instruction of the form:
> 	vsel<cond>.f32  s0, s1, s2
> means
> 	s0 := cond ? s1 : s2
> 
> This can be useful, among other places, in Thumb2 because it doesn't
> require
> an enclosing IT block.
> A small catch: The condition code used in vsel can only be one of {GE,
> GT,
> EQ, VS}. If we want to use their negations {LT, LE, NE, VC} we just
> flip the
> source operands.
> A new predicate is introduced that checks that the comparison yields an
> ARM
> condition code in the set {GE, GT, EQ, VS, LT, LE, NE, VC}.
> 
> New compilation tests are added. They pass on a model and no new
> regressions
> on arm-none-eabi with qemu.
> 
> Ok for trunk?
> 
> Thanks,
> Kyrill
> 
> gcc/ChangeLog
> 
> 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
> 
> 	* config/arm/arm.md (f_sels, f_seld): New types.
> 	(*cmov<mode>): New pattern.
> 	* config/arm/predicates.md (arm_vsel_comparison_operator): New
> predicate.
> 
> 
> gcc/testsuite/ChangeLog
> 
> 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
> 
> 	* gcc.target/arm/vseleqdf.c: New test.
> 	* gcc.target/arm/vseleqsf.c: Likewise.
> 	* gcc.target/arm/vselgedf.c: Likewise.
> 	* gcc.target/arm/vselgesf.c: Likewise.
> 	* gcc.target/arm/vselgtdf.c: Likewise.
> 	* gcc.target/arm/vselgtsf.c: Likewise.
> 	* gcc.target/arm/vselledf.c: Likewise.
> 	* gcc.target/arm/vsellesf.c: Likewise.
> 	* gcc.target/arm/vselltdf.c: Likewise.
> 	* gcc.target/arm/vselltsf.c: Likewise.
> 	* gcc.target/arm/vselnedf.c: Likewise.
> 	* gcc.target/arm/vselnesf.c: Likewise.
> 	* gcc.target/arm/vselvcdf.c: Likewise.
> 	* gcc.target/arm/vselvcsf.c: Likewise.
> 	* gcc.target/arm/vselvsdf.c: Likewise.
> 	* gcc.target/arm/vselvssf.c: Likewise.

Re: [PATCH][ARM] use vsel instruction for floating point conditional moves in ARMv8

[Ramana Radhakrishnan]

On 01/30/13 09:24, Kyrylo Tkachov wrote:
> Hi all,
> This patch uses the new ARMv8 AArch32 vsel instruction to implement
> conditional moves of floating point numbers.
> For example, an instruction of the form:
> 	vsel<cond>.f32  s0, s1, s2
> means
> 	s0 := cond ? s1 : s2
>
> This can be useful, among other places, in Thumb2 because it doesn't require
> an enclosing IT block.
> A small catch: The condition code used in vsel can only be one of {GE, GT,
> EQ, VS}. If we want to use their negations {LT, LE, NE, VC} we just flip the
> source operands.
> A new predicate is introduced that checks that the comparison yields an ARM
> condition code in the set {GE, GT, EQ, VS, LT, LE, NE, VC}.
>
> New compilation tests are added. They pass on a model and no new regressions
> on arm-none-eabi with qemu.

>
> Ok for trunk?

Ok for stage1 4.9.

ramana

>
> Thanks,
> Kyrill
>
> gcc/ChangeLog
>
> 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
>
> 	* config/arm/arm.md (f_sels, f_seld): New types.
> 	(*cmov<mode>): New pattern.
> 	* config/arm/predicates.md (arm_vsel_comparison_operator): New
> predicate.	
>
>
> gcc/testsuite/ChangeLog
>
> 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
>
> 	* gcc.target/arm/vseleqdf.c: New test.
> 	* gcc.target/arm/vseleqsf.c: Likewise.
> 	* gcc.target/arm/vselgedf.c: Likewise.
> 	* gcc.target/arm/vselgesf.c: Likewise.
> 	* gcc.target/arm/vselgtdf.c: Likewise.
> 	* gcc.target/arm/vselgtsf.c: Likewise.
> 	* gcc.target/arm/vselledf.c: Likewise.
> 	* gcc.target/arm/vsellesf.c: Likewise.
> 	* gcc.target/arm/vselltdf.c: Likewise.
> 	* gcc.target/arm/vselltsf.c: Likewise.
> 	* gcc.target/arm/vselnedf.c: Likewise.
> 	* gcc.target/arm/vselnesf.c: Likewise.
> 	* gcc.target/arm/vselvcdf.c: Likewise.
> 	* gcc.target/arm/vselvcsf.c: Likewise.
> 	* gcc.target/arm/vselvsdf.c: Likewise.
> 	* gcc.target/arm/vselvssf.c: Likewise.
>

RE: [PATCH][ARM] use vsel instruction for floating point conditional moves in ARMv8

[Kyrylo Tkachov]

> -----Original Message-----
> From: Ramana Radhakrishnan
> Sent: 18 February 2013 11:51
> To: Kyrylo Tkachov
> Cc: gcc-patches@gcc.gnu.org; Richard Earnshaw
> Subject: Re: [PATCH][ARM] use vsel instruction for floating point
> conditional moves in ARMv8
> 
> On 01/30/13 09:24, Kyrylo Tkachov wrote:
> > Hi all,
> > This patch uses the new ARMv8 AArch32 vsel instruction to implement
> > conditional moves of floating point numbers.
> > For example, an instruction of the form:
> > 	vsel<cond>.f32  s0, s1, s2
> > means
> > 	s0 := cond ? s1 : s2
> >
> > This can be useful, among other places, in Thumb2 because it doesn't
> require
> > an enclosing IT block.
> > A small catch: The condition code used in vsel can only be one of
> {GE, GT,
> > EQ, VS}. If we want to use their negations {LT, LE, NE, VC} we just
> flip the
> > source operands.
> > A new predicate is introduced that checks that the comparison yields
> an ARM
> > condition code in the set {GE, GT, EQ, VS, LT, LE, NE, VC}.
> >
> > New compilation tests are added. They pass on a model and no new
> regressions
> > on arm-none-eabi with qemu.
> 
> >
> > Ok for trunk?
> 
> Ok for stage1 4.9.

Hi Ramana,

Thanks for the review.
Re-tested on arm-none-eabi against current trunk and applied as r197052.

> 
> Ramana

Thanks,
Kyrill
> 
> >
> > Thanks,
> > Kyrill
> >
> > gcc/ChangeLog
> >
> > 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
> >
> > 	* config/arm/arm.md (f_sels, f_seld): New types.
> > 	(*cmov<mode>): New pattern.
> > 	* config/arm/predicates.md (arm_vsel_comparison_operator): New
> > predicate.
> >
> >
> > gcc/testsuite/ChangeLog
> >
> > 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
> >
> > 	* gcc.target/arm/vseleqdf.c: New test.
> > 	* gcc.target/arm/vseleqsf.c: Likewise.
> > 	* gcc.target/arm/vselgedf.c: Likewise.
> > 	* gcc.target/arm/vselgesf.c: Likewise.
> > 	* gcc.target/arm/vselgtdf.c: Likewise.
> > 	* gcc.target/arm/vselgtsf.c: Likewise.
> > 	* gcc.target/arm/vselledf.c: Likewise.
> > 	* gcc.target/arm/vsellesf.c: Likewise.
> > 	* gcc.target/arm/vselltdf.c: Likewise.
> > 	* gcc.target/arm/vselltsf.c: Likewise.
> > 	* gcc.target/arm/vselnedf.c: Likewise.
> > 	* gcc.target/arm/vselnesf.c: Likewise.
> > 	* gcc.target/arm/vselvcdf.c: Likewise.
> > 	* gcc.target/arm/vselvcsf.c: Likewise.
> > 	* gcc.target/arm/vselvsdf.c: Likewise.
> > 	* gcc.target/arm/vselvssf.c: Likewise.
> >

Re: [PATCH][ARM] use vsel instruction for floating point conditional moves in ARMv8

[Matthew Gretton-Dann]

Would it be possible for this patch and the others Kyrylo has recently done 
for the new ARMv8 AArch32 instructions to be backported to 4.8?

In particular I'm refering to:

http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00994.html (trunk r197052)
http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00874.html (trunk r197051)
http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00873.html (trunk r197046)
http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00652.html (trunk r197040 and 
197041)

Thanks,

Matt


On 25/03/13 15:21, Kyrylo Tkachov wrote:
>> -----Original Message-----
>> From: Ramana Radhakrishnan
>> Sent: 18 February 2013 11:51
>> To: Kyrylo Tkachov
>> Cc: gcc-patches@gcc.gnu.org; Richard Earnshaw
>> Subject: Re: [PATCH][ARM] use vsel instruction for floating point
>> conditional moves in ARMv8
>>
>> On 01/30/13 09:24, Kyrylo Tkachov wrote:
>>> Hi all,
>>> This patch uses the new ARMv8 AArch32 vsel instruction to implement
>>> conditional moves of floating point numbers.
>>> For example, an instruction of the form:
>>> 	vsel<cond>.f32  s0, s1, s2
>>> means
>>> 	s0 := cond ? s1 : s2
>>>
>>> This can be useful, among other places, in Thumb2 because it doesn't
>> require
>>> an enclosing IT block.
>>> A small catch: The condition code used in vsel can only be one of
>> {GE, GT,
>>> EQ, VS}. If we want to use their negations {LT, LE, NE, VC} we just
>> flip the
>>> source operands.
>>> A new predicate is introduced that checks that the comparison yields
>> an ARM
>>> condition code in the set {GE, GT, EQ, VS, LT, LE, NE, VC}.
>>>
>>> New compilation tests are added. They pass on a model and no new
>> regressions
>>> on arm-none-eabi with qemu.
>>
>>>
>>> Ok for trunk?
>>
>> Ok for stage1 4.9.
>
> Hi Ramana,
>
> Thanks for the review.
> Re-tested on arm-none-eabi against current trunk and applied as r197052.
>
>>
>> Ramana
>
> Thanks,
> Kyrill
>>
>>>
>>> Thanks,
>>> Kyrill
>>>
>>> gcc/ChangeLog
>>>
>>> 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
>>>
>>> 	* config/arm/arm.md (f_sels, f_seld): New types.
>>> 	(*cmov<mode>): New pattern.
>>> 	* config/arm/predicates.md (arm_vsel_comparison_operator): New
>>> predicate.
>>>
>>>
>>> gcc/testsuite/ChangeLog
>>>
>>> 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
>>>
>>> 	* gcc.target/arm/vseleqdf.c: New test.
>>> 	* gcc.target/arm/vseleqsf.c: Likewise.
>>> 	* gcc.target/arm/vselgedf.c: Likewise.
>>> 	* gcc.target/arm/vselgesf.c: Likewise.
>>> 	* gcc.target/arm/vselgtdf.c: Likewise.
>>> 	* gcc.target/arm/vselgtsf.c: Likewise.
>>> 	* gcc.target/arm/vselledf.c: Likewise.
>>> 	* gcc.target/arm/vsellesf.c: Likewise.
>>> 	* gcc.target/arm/vselltdf.c: Likewise.
>>> 	* gcc.target/arm/vselltsf.c: Likewise.
>>> 	* gcc.target/arm/vselnedf.c: Likewise.
>>> 	* gcc.target/arm/vselnesf.c: Likewise.
>>> 	* gcc.target/arm/vselvcdf.c: Likewise.
>>> 	* gcc.target/arm/vselvcsf.c: Likewise.
>>> 	* gcc.target/arm/vselvsdf.c: Likewise.
>>> 	* gcc.target/arm/vselvssf.c: Likewise.
>>>
>
>
>
>


-- 
Matthew Gretton-Dann
Toolchain Working Group, Linaro

Re: [PATCH][ARM] use vsel instruction for floating point conditional moves in ARMv8

[Richard Biener]

On Wed, 3 Apr 2013, Matthew Gretton-Dann wrote:

> Would it be possible for this patch and the others Kyrylo has recently done
> for the new ARMv8 AArch32 instructions to be backported to 4.8?
> 
> In particular I'm refering to:
> 
> http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00994.html (trunk r197052)
> http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00874.html (trunk r197051)
> http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00873.html (trunk r197046)
> http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00652.html (trunk r197040 and
> 197041)

Those seem to be new features and not regression fixes and thus are
not appropriate for a release branch.

Richard.

> Thanks,
> 
> Matt
> 
> 
> On 25/03/13 15:21, Kyrylo Tkachov wrote:
> > > -----Original Message-----
> > > From: Ramana Radhakrishnan
> > > Sent: 18 February 2013 11:51
> > > To: Kyrylo Tkachov
> > > Cc: gcc-patches@gcc.gnu.org; Richard Earnshaw
> > > Subject: Re: [PATCH][ARM] use vsel instruction for floating point
> > > conditional moves in ARMv8
> > > 
> > > On 01/30/13 09:24, Kyrylo Tkachov wrote:
> > > > Hi all,
> > > > This patch uses the new ARMv8 AArch32 vsel instruction to implement
> > > > conditional moves of floating point numbers.
> > > > For example, an instruction of the form:
> > > > 	vsel<cond>.f32  s0, s1, s2
> > > > means
> > > > 	s0 := cond ? s1 : s2
> > > > 
> > > > This can be useful, among other places, in Thumb2 because it doesn't
> > > require
> > > > an enclosing IT block.
> > > > A small catch: The condition code used in vsel can only be one of
> > > {GE, GT,
> > > > EQ, VS}. If we want to use their negations {LT, LE, NE, VC} we just
> > > flip the
> > > > source operands.
> > > > A new predicate is introduced that checks that the comparison yields
> > > an ARM
> > > > condition code in the set {GE, GT, EQ, VS, LT, LE, NE, VC}.
> > > > 
> > > > New compilation tests are added. They pass on a model and no new
> > > regressions
> > > > on arm-none-eabi with qemu.
> > > 
> > > > 
> > > > Ok for trunk?
> > > 
> > > Ok for stage1 4.9.
> > 
> > Hi Ramana,
> > 
> > Thanks for the review.
> > Re-tested on arm-none-eabi against current trunk and applied as r197052.
> > 
> > > 
> > > Ramana
> > 
> > Thanks,
> > Kyrill
> > > 
> > > > 
> > > > Thanks,
> > > > Kyrill
> > > > 
> > > > gcc/ChangeLog
> > > > 
> > > > 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
> > > > 
> > > > 	* config/arm/arm.md (f_sels, f_seld): New types.
> > > > 	(*cmov<mode>): New pattern.
> > > > 	* config/arm/predicates.md (arm_vsel_comparison_operator): New
> > > > predicate.
> > > > 
> > > > 
> > > > gcc/testsuite/ChangeLog
> > > > 
> > > > 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
> > > > 
> > > > 	* gcc.target/arm/vseleqdf.c: New test.
> > > > 	* gcc.target/arm/vseleqsf.c: Likewise.
> > > > 	* gcc.target/arm/vselgedf.c: Likewise.
> > > > 	* gcc.target/arm/vselgesf.c: Likewise.
> > > > 	* gcc.target/arm/vselgtdf.c: Likewise.
> > > > 	* gcc.target/arm/vselgtsf.c: Likewise.
> > > > 	* gcc.target/arm/vselledf.c: Likewise.
> > > > 	* gcc.target/arm/vsellesf.c: Likewise.
> > > > 	* gcc.target/arm/vselltdf.c: Likewise.
> > > > 	* gcc.target/arm/vselltsf.c: Likewise.
> > > > 	* gcc.target/arm/vselnedf.c: Likewise.
> > > > 	* gcc.target/arm/vselnesf.c: Likewise.
> > > > 	* gcc.target/arm/vselvcdf.c: Likewise.
> > > > 	* gcc.target/arm/vselvcsf.c: Likewise.
> > > > 	* gcc.target/arm/vselvsdf.c: Likewise.
> > > > 	* gcc.target/arm/vselvssf.c: Likewise.
> > > > 
> > 
> > 
> > 
> > 
> 
> 
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE / SUSE Labs
SUSE LINUX Products GmbH - Nuernberg - AG Nuernberg - HRB 16746
GF: Jeff Hawn, Jennifer Guild, Felix Imend

RE: [PATCH][ARM] use vsel instruction for floating point conditional moves in ARMv8

[Kyrylo Tkachov]

> -----Original Message-----
> From: Richard Biener [mailto:rguenther@suse.de]
> Sent: 04 April 2013 08:46
> To: Matthew Gretton-Dann
> Cc: gcc-patches@gcc.gnu.org; Kyrylo Tkachov; Ramana Radhakrishnan;
> Richard Earnshaw
> Subject: Re: [PATCH][ARM] use vsel instruction for floating point
> conditional moves in ARMv8
> 
> On Wed, 3 Apr 2013, Matthew Gretton-Dann wrote:
> 
> > Would it be possible for this patch and the others Kyrylo has
> recently done
> > for the new ARMv8 AArch32 instructions to be backported to 4.8?
> >
> > In particular I'm refering to:
> >
> > http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00994.html (trunk
> r197052)
> > http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00874.html (trunk
> r197051)
> > http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00873.html (trunk
> r197046)
> > http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00652.html (trunk
> r197040 and
> > 197041)
> 
> Those seem to be new features and not regression fixes and thus are
> not appropriate for a release branch.

The last one:
http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00652.html
is actually a bug fix for an ICE with vcond that can be triggered
in 4.8, 4.7 and 4.6 (PR 56720) so could be worth backporting.


Thanks,
Kyrill

> 
> Richard.
> 
> > Thanks,
> >
> > Matt
> >
> >
> > On 25/03/13 15:21, Kyrylo Tkachov wrote:
> > > > -----Original Message-----
> > > > From: Ramana Radhakrishnan
> > > > Sent: 18 February 2013 11:51
> > > > To: Kyrylo Tkachov
> > > > Cc: gcc-patches@gcc.gnu.org; Richard Earnshaw
> > > > Subject: Re: [PATCH][ARM] use vsel instruction for floating point
> > > > conditional moves in ARMv8
> > > >
> > > > On 01/30/13 09:24, Kyrylo Tkachov wrote:
> > > > > Hi all,
> > > > > This patch uses the new ARMv8 AArch32 vsel instruction to
> implement
> > > > > conditional moves of floating point numbers.
> > > > > For example, an instruction of the form:
> > > > > 	vsel<cond>.f32  s0, s1, s2
> > > > > means
> > > > > 	s0 := cond ? s1 : s2
> > > > >
> > > > > This can be useful, among other places, in Thumb2 because it
> doesn't
> > > > require
> > > > > an enclosing IT block.
> > > > > A small catch: The condition code used in vsel can only be one
> of
> > > > {GE, GT,
> > > > > EQ, VS}. If we want to use their negations {LT, LE, NE, VC} we
> just
> > > > flip the
> > > > > source operands.
> > > > > A new predicate is introduced that checks that the comparison
> yields
> > > > an ARM
> > > > > condition code in the set {GE, GT, EQ, VS, LT, LE, NE, VC}.
> > > > >
> > > > > New compilation tests are added. They pass on a model and no
> new
> > > > regressions
> > > > > on arm-none-eabi with qemu.
> > > >
> > > > >
> > > > > Ok for trunk?
> > > >
> > > > Ok for stage1 4.9.
> > >
> > > Hi Ramana,
> > >
> > > Thanks for the review.
> > > Re-tested on arm-none-eabi against current trunk and applied as
> r197052.
> > >
> > > >
> > > > Ramana
> > >
> > > Thanks,
> > > Kyrill
> > > >
> > > > >
> > > > > Thanks,
> > > > > Kyrill
> > > > >
> > > > > gcc/ChangeLog
> > > > >
> > > > > 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
> > > > >
> > > > > 	* config/arm/arm.md (f_sels, f_seld): New types.
> > > > > 	(*cmov<mode>): New pattern.
> > > > > 	* config/arm/predicates.md (arm_vsel_comparison_operator):
> New
> > > > > predicate.
> > > > >
> > > > >
> > > > > gcc/testsuite/ChangeLog
> > > > >
> > > > > 2013-01-30  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
> > > > >
> > > > > 	* gcc.target/arm/vseleqdf.c: New test.
> > > > > 	* gcc.target/arm/vseleqsf.c: Likewise.
> > > > > 	* gcc.target/arm/vselgedf.c: Likewise.
> > > > > 	* gcc.target/arm/vselgesf.c: Likewise.
> > > > > 	* gcc.target/arm/vselgtdf.c: Likewise.
> > > > > 	* gcc.target/arm/vselgtsf.c: Likewise.
> > > > > 	* gcc.target/arm/vselledf.c: Likewise.
> > > > > 	* gcc.target/arm/vsellesf.c: Likewise.
> > > > > 	* gcc.target/arm/vselltdf.c: Likewise.
> > > > > 	* gcc.target/arm/vselltsf.c: Likewise.
> > > > > 	* gcc.target/arm/vselnedf.c: Likewise.
> > > > > 	* gcc.target/arm/vselnesf.c: Likewise.
> > > > > 	* gcc.target/arm/vselvcdf.c: Likewise.
> > > > > 	* gcc.target/arm/vselvcsf.c: Likewise.
> > > > > 	* gcc.target/arm/vselvsdf.c: Likewise.
> > > > > 	* gcc.target/arm/vselvssf.c: Likewise.
> > > > >
> > >
> > >
> > >
> > >
> >
> >
> >
> 
> --
> Richard Biener <rguenther@suse.de>
> SUSE / SUSE Labs
> SUSE LINUX Products GmbH - Nuernberg - AG Nuernberg - HRB 16746
> GF: Jeff Hawn, Jennifer Guild, Felix Imend

Re: [PATCH][ARM] use vsel instruction for floating point conditional moves in ARMv8

[Ramana Radhakrishnan]

On 04/04/13 08:46, Richard Biener wrote:
> On Wed, 3 Apr 2013, Matthew Gretton-Dann wrote:
>
>> Would it be possible for this patch and the others Kyrylo has recently done
>> for the new ARMv8 AArch32 instructions to be backported to 4.8?
>>
>> In particular I'm refering to:
>>
>> http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00994.html (trunk r197052)
>> http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00874.html (trunk r197051)
>> http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00873.html (trunk r197046)



>
> Those seem to be new features and not regression fixes and thus are
> not appropriate for a release branch.

Not all of them really, 
http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00652.html is really a 
target bug fix which causes mild annoyance once a while. I do believe 
that Kyryll was looking at proposing a backport of this patch (and there 
is a bugzilla entry for this one.)

I am sympathetic to the new atomics for v8-a and I've heard atleast one 
request for it privately (i.e. this one 
http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00874.html) but I'd defer 
to the RM's on that one.

regards
Ramana

Re: [PATCH][ARM] use vsel instruction for floating point conditional moves in ARMv8

[Richard Biener]

On Thu, 4 Apr 2013, Ramana Radhakrishnan wrote:

> On 04/04/13 08:46, Richard Biener wrote:
> > On Wed, 3 Apr 2013, Matthew Gretton-Dann wrote:
> > 
> > > Would it be possible for this patch and the others Kyrylo has recently
> > > done
> > > for the new ARMv8 AArch32 instructions to be backported to 4.8?
> > > 
> > > In particular I'm refering to:
> > > 
> > > http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00994.html (trunk r197052)
> > > http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00874.html (trunk r197051)
> > > http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00873.html (trunk r197046)
> 
> 
> 
> > 
> > Those seem to be new features and not regression fixes and thus are
> > not appropriate for a release branch.
> 
> Not all of them really,
> http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00652.html is really a target bug
> fix which causes mild annoyance once a while. I do believe that Kyryll was
> looking at proposing a backport of this patch (and there is a bugzilla entry
> for this one.)
> 
> I am sympathetic to the new atomics for v8-a and I've heard atleast one
> request for it privately (i.e. this one
> http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00874.html) but I'd defer to the
> RM's on that one.

Please ask more specifically then.  Note that release managers generally
defer to target maintainers for architecture specific patches.  But we
of course expect them to follow the rule of fixing regressions
and wrong-code issues only.  Exceptions are always possible but
strongly discouraged by us for primary and secondary targets where
new regressions may block a release.

Richard.