[i386] Prefer INT_SSE_REGS for SSE_FLOAT_MODE_P in preferred_reload_class.
Commit Message
The patch helps reload to choose GENENRAL_REGS alternatives for
SSE_FLOAT_MODE and enabled optimization like
- vmovd %xmm0, -4(%rsp)
- movl $1, %eax
- addl -4(%rsp), %eax
+ movd %xmm0, %eax
+ addl $1, %eax
Bootstrapped anf regtested on x86_64-pc-linux-gnu{-m32,} and
x86_64-pc-linux-gnu{-m32\ march=cascadelake,\ -march=cadcadelake}.
No big performace impact is abserved for SPEC2017 on ICX/CLX with both
Ofast -march=native -flto -funroll-loops and -O2 -mtune=generic options.
Ok for trunk?
gcc/ChangeLog:
PR target/95740
* config/i386/i386.c (ix86_preferred_reload_class): Prefer
INT_SSE_REGS for SSE_FLOAT_MODE_P.
* config/i386/i386.h (INT_SSE_CLASS_P): New.
gcc/testsuite/ChangeLog:
* gcc.target/i386/pr95740.c: New test.
---
gcc/config/i386/i386.c | 5 +++--
gcc/config/i386/i386.h | 2 ++
gcc/testsuite/gcc.target/i386/pr95740.c | 26 +++++++++++++++++++++++++
3 files changed, 31 insertions(+), 2 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr95740.c
Comments
On Thu, Dec 2, 2021 at 4:27 PM liuhongt <hongtao.liu@intel.com> wrote:
>
> The patch helps reload to choose GENENRAL_REGS alternatives for
> SSE_FLOAT_MODE and enabled optimization like
>
> - vmovd %xmm0, -4(%rsp)
> - movl $1, %eax
> - addl -4(%rsp), %eax
> + movd %xmm0, %eax
> + addl $1, %eax
>
> Bootstrapped anf regtested on x86_64-pc-linux-gnu{-m32,} and
> x86_64-pc-linux-gnu{-m32\ march=cascadelake,\ -march=cadcadelake}.
>
> No big performace impact is abserved for SPEC2017 on ICX/CLX with both
> Ofast -march=native -flto -funroll-loops and -O2 -mtune=generic options.
>
> Ok for trunk?
>
> gcc/ChangeLog:
>
> PR target/95740
> * config/i386/i386.c (ix86_preferred_reload_class): Prefer
> INT_SSE_REGS for SSE_FLOAT_MODE_P.
> * config/i386/i386.h (INT_SSE_CLASS_P): New.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/i386/pr95740.c: New test.
> ---
> gcc/config/i386/i386.c | 5 +++--
> gcc/config/i386/i386.h | 2 ++
> gcc/testsuite/gcc.target/i386/pr95740.c | 26 +++++++++++++++++++++++++
> 3 files changed, 31 insertions(+), 2 deletions(-)
> create mode 100644 gcc/testsuite/gcc.target/i386/pr95740.c
>
> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> index 80fee627358..977af1c31a7 100644
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -19194,9 +19194,10 @@ ix86_preferred_reload_class (rtx x, reg_class_t regclass)
> return NO_REGS;
> }
>
> - /* Prefer SSE regs only, if we can use them for math. */
> + /* Prefer INT_SSE_REGS, enable reload from SSE register to GENERAL_REGS,
> + refer to PR95740. */
> if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> - return SSE_CLASS_P (regclass) ? regclass : NO_REGS;
> + return INT_SSE_CLASS_P (regclass) ? regclass : NO_REGS;
>
> /* Generally when we see PLUS here, it's the function invariant
> (plus soft-fp const_int). Which can only be computed into general
> diff --git a/gcc/config/i386/i386.h b/gcc/config/i386/i386.h
> index 2fda1e0686e..ec90e47904b 100644
> --- a/gcc/config/i386/i386.h
> +++ b/gcc/config/i386/i386.h
> @@ -1283,6 +1283,8 @@ enum reg_class
> reg_class_subset_p ((CLASS), FLOAT_REGS)
> #define SSE_CLASS_P(CLASS) \
> reg_class_subset_p ((CLASS), ALL_SSE_REGS)
> +#define INT_SSE_CLASS_P(CLASS) \
> + reg_class_subset_p ((CLASS), INT_SSE_REGS)
> #define MMX_CLASS_P(CLASS) \
> ((CLASS) == MMX_REGS)
> #define MASK_CLASS_P(CLASS) \
> diff --git a/gcc/testsuite/gcc.target/i386/pr95740.c b/gcc/testsuite/gcc.target/i386/pr95740.c
> new file mode 100644
> index 00000000000..9bc7b862787
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr95740.c
> @@ -0,0 +1,26 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-options "-msse2 -O2 -mtune-ctrl=use_incdec -masm=att -mfpmath=sse" } */
> +/* { dg-final { scan-assembler-times {(?n)movd[\t ]*%xmm0.*%eax} 1 } } */
> +/* { dg-final { scan-assembler-times {(?n)incl[\t ]*%eax} 1 } } */
> +/* { dg-final { scan-assembler-times {(?n)movq[\t ]*%xmm0.*%rax} 1 } } */
> +/* { dg-final { scan-assembler-times {(?n)incq[\t ]*%rax} 1 } } */
> +
> +int
> +foo (float a)
> +{
> + union{
> + int b;
> + float a;}u;
> + u.a = a;
> + return u.b + 1;
> +}
> +
> +long long
> +foo1 (double a)
> +{
> + union{
> + long long b;
> + double a;}u;
> + u.a = a;
> + return u.b + 1;
> +}
> --
> 2.18.1
>
On Thu, Dec 2, 2021 at 9:36 AM Hongtao Liu <crazylht@gmail.com> wrote:
>
> On Thu, Dec 2, 2021 at 4:27 PM liuhongt <hongtao.liu@intel.com> wrote:
> >
> > The patch helps reload to choose GENENRAL_REGS alternatives for
> > SSE_FLOAT_MODE and enabled optimization like
> >
> > - vmovd %xmm0, -4(%rsp)
> > - movl $1, %eax
> > - addl -4(%rsp), %eax
> > + movd %xmm0, %eax
> > + addl $1, %eax
> >
> > Bootstrapped anf regtested on x86_64-pc-linux-gnu{-m32,} and
> > x86_64-pc-linux-gnu{-m32\ march=cascadelake,\ -march=cadcadelake}.
> >
> > No big performace impact is abserved for SPEC2017 on ICX/CLX with both
> > Ofast -march=native -flto -funroll-loops and -O2 -mtune=generic options.
> >
> > Ok for trunk?
Please also consider TARGET_INTER_UNIT_MOVES_TO_VEC and
TARGET_INTER_UNIT_MOVES_FROM_VEC.
Uros.
> >
> > gcc/ChangeLog:
> >
> > PR target/95740
> > * config/i386/i386.c (ix86_preferred_reload_class): Prefer
> > INT_SSE_REGS for SSE_FLOAT_MODE_P.
> > * config/i386/i386.h (INT_SSE_CLASS_P): New.
> >
> > gcc/testsuite/ChangeLog:
> >
> > * gcc.target/i386/pr95740.c: New test.
> > ---
> > gcc/config/i386/i386.c | 5 +++--
> > gcc/config/i386/i386.h | 2 ++
> > gcc/testsuite/gcc.target/i386/pr95740.c | 26 +++++++++++++++++++++++++
> > 3 files changed, 31 insertions(+), 2 deletions(-)
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr95740.c
> >
> > diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> > index 80fee627358..977af1c31a7 100644
> > --- a/gcc/config/i386/i386.c
> > +++ b/gcc/config/i386/i386.c
> > @@ -19194,9 +19194,10 @@ ix86_preferred_reload_class (rtx x, reg_class_t regclass)
> > return NO_REGS;
> > }
> >
> > - /* Prefer SSE regs only, if we can use them for math. */
> > + /* Prefer INT_SSE_REGS, enable reload from SSE register to GENERAL_REGS,
> > + refer to PR95740. */
> > if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> > - return SSE_CLASS_P (regclass) ? regclass : NO_REGS;
> > + return INT_SSE_CLASS_P (regclass) ? regclass : NO_REGS;
> >
> > /* Generally when we see PLUS here, it's the function invariant
> > (plus soft-fp const_int). Which can only be computed into general
> > diff --git a/gcc/config/i386/i386.h b/gcc/config/i386/i386.h
> > index 2fda1e0686e..ec90e47904b 100644
> > --- a/gcc/config/i386/i386.h
> > +++ b/gcc/config/i386/i386.h
> > @@ -1283,6 +1283,8 @@ enum reg_class
> > reg_class_subset_p ((CLASS), FLOAT_REGS)
> > #define SSE_CLASS_P(CLASS) \
> > reg_class_subset_p ((CLASS), ALL_SSE_REGS)
> > +#define INT_SSE_CLASS_P(CLASS) \
> > + reg_class_subset_p ((CLASS), INT_SSE_REGS)
> > #define MMX_CLASS_P(CLASS) \
> > ((CLASS) == MMX_REGS)
> > #define MASK_CLASS_P(CLASS) \
> > diff --git a/gcc/testsuite/gcc.target/i386/pr95740.c b/gcc/testsuite/gcc.target/i386/pr95740.c
> > new file mode 100644
> > index 00000000000..9bc7b862787
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr95740.c
> > @@ -0,0 +1,26 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-options "-msse2 -O2 -mtune-ctrl=use_incdec -masm=att -mfpmath=sse" } */
> > +/* { dg-final { scan-assembler-times {(?n)movd[\t ]*%xmm0.*%eax} 1 } } */
> > +/* { dg-final { scan-assembler-times {(?n)incl[\t ]*%eax} 1 } } */
> > +/* { dg-final { scan-assembler-times {(?n)movq[\t ]*%xmm0.*%rax} 1 } } */
> > +/* { dg-final { scan-assembler-times {(?n)incq[\t ]*%rax} 1 } } */
> > +
> > +int
> > +foo (float a)
> > +{
> > + union{
> > + int b;
> > + float a;}u;
> > + u.a = a;
> > + return u.b + 1;
> > +}
> > +
> > +long long
> > +foo1 (double a)
> > +{
> > + union{
> > + long long b;
> > + double a;}u;
> > + u.a = a;
> > + return u.b + 1;
> > +}
> > --
> > 2.18.1
> >
>
>
> --
> BR,
> Hongtao
@@ -19194,9 +19194,10 @@ ix86_preferred_reload_class (rtx x, reg_class_t regclass)
return NO_REGS;
}
- /* Prefer SSE regs only, if we can use them for math. */
+ /* Prefer INT_SSE_REGS, enable reload from SSE register to GENERAL_REGS,
+ refer to PR95740. */
if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
- return SSE_CLASS_P (regclass) ? regclass : NO_REGS;
+ return INT_SSE_CLASS_P (regclass) ? regclass : NO_REGS;
/* Generally when we see PLUS here, it's the function invariant
(plus soft-fp const_int). Which can only be computed into general
@@ -1283,6 +1283,8 @@ enum reg_class
reg_class_subset_p ((CLASS), FLOAT_REGS)
#define SSE_CLASS_P(CLASS) \
reg_class_subset_p ((CLASS), ALL_SSE_REGS)
+#define INT_SSE_CLASS_P(CLASS) \
+ reg_class_subset_p ((CLASS), INT_SSE_REGS)
#define MMX_CLASS_P(CLASS) \
((CLASS) == MMX_REGS)
#define MASK_CLASS_P(CLASS) \
new file mode 100644
@@ -0,0 +1,26 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-options "-msse2 -O2 -mtune-ctrl=use_incdec -masm=att -mfpmath=sse" } */
+/* { dg-final { scan-assembler-times {(?n)movd[\t ]*%xmm0.*%eax} 1 } } */
+/* { dg-final { scan-assembler-times {(?n)incl[\t ]*%eax} 1 } } */
+/* { dg-final { scan-assembler-times {(?n)movq[\t ]*%xmm0.*%rax} 1 } } */
+/* { dg-final { scan-assembler-times {(?n)incq[\t ]*%rax} 1 } } */
+
+int
+foo (float a)
+{
+ union{
+ int b;
+ float a;}u;
+ u.a = a;
+ return u.b + 1;
+}
+
+long long
+foo1 (double a)
+{
+ union{
+ long long b;
+ double a;}u;
+ u.a = a;
+ return u.b + 1;
+}