Message ID | ff9638d7-0f97-b5e7-f57a-40f1dd721ce9@linux.ibm.com |
---|---|
State | New |
Headers | show |
Series | [rs6000] Disable gimple fold for float or double vec_minmax when fast-math is not set | expand |
Hi, Hao Neither the inlined patch nor the attached patch seem to contain the change to rs6000-call.c. I only see the new testcases. Please resend the complete patch. Thanks David On Mon, Nov 1, 2021 at 2:48 AM HAO CHEN GUI <guihaoc@linux.ibm.com> wrote: > > Hi, > > This patch disables gimple folding for VSX_BUILTIN_XVMINDP, VSX_BUILTIN_XVMAXDP, ALTIVEC_BUILTIN_VMINFP and ALTIVEC_BUILTIN_VMAXFP when fast-math is not set. With the gimple folding is enabled, the four built-ins will be implemented by c-type instructions - xs[min|max]cdp on P9 and P10 if they can be converted to scalar comparisons. While they are implemented by xv[min|max][s|d]p on P8 and P7 as P8 and P7 don't have corresponding scalar comparison instructions. The patch binds these four built-ins to xv[min|max][s|d]p when fast-math is not set. The two new test cases illustrate it. > > ALTIVEC_BUILTIN_VMINFP and ALTIVEC_BUILTIN_VMAXFP are not implemented by vminfp or vmaxfp. > > rs6000-builtin.def:BU_ALTIVEC_2 (VMAXFP, "vmaxfp", CONST, smaxv4sf3) > > rs6000-builtin.def:BU_ALTIVEC_2 (VMINFP, "vminfp", CONST, sminv4sf3) > > Bootstrapped and tested on powerpc64le-linux with no regressions. Is this okay for trunk? Any recommendations? Thanks a lot. > > > ChangeLog > > 2021-11-01 Haochen Gui <guihaoc@linux.ibm.com> > > gcc/ > * config/rs6000/rs6000-call.c (rs6000_gimple_fold_builtin): Disable > gimple fold for VSX_BUILTIN_XVMINDP, ALTIVEC_BUILTIN_VMINFP, > VSX_BUILTIN_XVMAXDP, ALTIVEC_BUILTIN_VMAXFP when fast-math is not > set. > > gcc/testsuite/ > * gcc.target/powerpc/vec-minmax-1.c: New test. > * gcc.target/powerpc/vec-minmax-2.c: Likewise. > > > patch.diff > > diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c > new file mode 100644 > index 00000000000..e238659c9be > --- /dev/null > +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c > @@ -0,0 +1,52 @@ > +/* { dg-require-effective-target powerpc_p9vector_ok } */ > +/* { dg-options "-O2 -mdejagnu-cpu=power9" } */ > +/* { dg-final { scan-assembler-times {\mxvmaxdp\M} 1 } } */ > +/* { dg-final { scan-assembler-times {\mxvmaxsp\M} 1 } } */ > +/* { dg-final { scan-assembler-times {\mxvmindp\M} 1 } } */ > +/* { dg-final { scan-assembler-times {\mxvminsp\M} 1 } } */ > + > +/* This test verifies that float or double vec_min/max are bound to > + xv[min|max][d|s]p instructions when fast-math is not set. */ > + > + > +#include <altivec.h> > + > +#ifdef _BIG_ENDIAN > + const int PREF_D = 0; > +#else > + const int PREF_D = 1; > +#endif > + > +double vmaxd (double a, double b) > +{ > + vector double va = vec_promote (a, PREF_D); > + vector double vb = vec_promote (b, PREF_D); > + return vec_extract (vec_max (va, vb), PREF_D); > +} > + > +double vmind (double a, double b) > +{ > + vector double va = vec_promote (a, PREF_D); > + vector double vb = vec_promote (b, PREF_D); > + return vec_extract (vec_min (va, vb), PREF_D); > +} > + > +#ifdef _BIG_ENDIAN > + const int PREF_F = 0; > +#else > + const int PREF_F = 3; > +#endif > + > +float vmaxf (float a, float b) > +{ > + vector float va = vec_promote (a, PREF_F); > + vector float vb = vec_promote (b, PREF_F); > + return vec_extract (vec_max (va, vb), PREF_F); > +} > + > +float vminf (float a, float b) > +{ > + vector float va = vec_promote (a, PREF_F); > + vector float vb = vec_promote (b, PREF_F); > + return vec_extract (vec_min (va, vb), PREF_F); > +} > diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c > new file mode 100644 > index 00000000000..149275d8709 > --- /dev/null > +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c > @@ -0,0 +1,50 @@ > +/* { dg-require-effective-target powerpc_p9vector_ok } */ > +/* { dg-options "-O2 -mdejagnu-cpu=power9 -ffast-math" } */ > +/* { dg-final { scan-assembler-times {\mxsmaxcdp\M} 2 } } */ > +/* { dg-final { scan-assembler-times {\mxsmincdp\M} 2 } } */ > + > +/* This test verifies that float or double vec_min/max can be converted > + to scalar comparison when fast-math is set. */ > + > + > +#include <altivec.h> > + > +#ifdef _BIG_ENDIAN > + const int PREF_D = 0; > +#else > + const int PREF_D = 1; > +#endif > + > +double vmaxd (double a, double b) > +{ > + vector double va = vec_promote (a, PREF_D); > + vector double vb = vec_promote (b, PREF_D); > + return vec_extract (vec_max (va, vb), PREF_D); > +} > + > +double vmind (double a, double b) > +{ > + vector double va = vec_promote (a, PREF_D); > + vector double vb = vec_promote (b, PREF_D); > + return vec_extract (vec_min (va, vb), PREF_D); > +} > + > +#ifdef _BIG_ENDIAN > + const int PREF_F = 0; > +#else > + const int PREF_F = 3; > +#endif > + > +float vmaxf (float a, float b) > +{ > + vector float va = vec_promote (a, PREF_F); > + vector float vb = vec_promote (b, PREF_F); > + return vec_extract (vec_max (va, vb), PREF_F); > +} > + > +float vminf (float a, float b) > +{ > + vector float va = vec_promote (a, PREF_F); > + vector float vb = vec_promote (b, PREF_F); > + return vec_extract (vec_min (va, vb), PREF_F); > +}
David, My patch file was broken. I am sorry for it. Here is the correct one. Thanks a lot. ChangeLog 2021-11-01 Haochen Gui <guihaoc@linux.ibm.com> gcc/ * config/rs6000/rs6000-call.c (rs6000_gimple_fold_builtin): Disable gimple fold for VSX_BUILTIN_XVMINDP, ALTIVEC_BUILTIN_VMINFP, VSX_BUILTIN_XVMAXDP, ALTIVEC_BUILTIN_VMAXFP when fast-math is not set. gcc/testsuite/ * gcc.target/powerpc/vec-minmax-1.c: New test. * gcc.target/powerpc/vec-minmax-2.c: Likewise. patch.diff diff --git a/gcc/config/rs6000/rs6000-call.c b/gcc/config/rs6000/rs6000-call.c index 7d485480225..a8e193a0089 100644 --- a/gcc/config/rs6000/rs6000-call.c +++ b/gcc/config/rs6000/rs6000-call.c @@ -12333,6 +12333,14 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) return true; /* flavors of vec_min. */ case VSX_BUILTIN_XVMINDP: + case ALTIVEC_BUILTIN_VMINFP: + { + lhs = gimple_call_lhs (stmt); + tree type = TREE_TYPE (lhs); + if (HONOR_NANS (type)) + return false; + gcc_fallthrough (); + } case P8V_BUILTIN_VMINSD: case P8V_BUILTIN_VMINUD: case ALTIVEC_BUILTIN_VMINSB: @@ -12341,7 +12349,6 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) case ALTIVEC_BUILTIN_VMINUB: case ALTIVEC_BUILTIN_VMINUH: case ALTIVEC_BUILTIN_VMINUW: - case ALTIVEC_BUILTIN_VMINFP: arg0 = gimple_call_arg (stmt, 0); arg1 = gimple_call_arg (stmt, 1); lhs = gimple_call_lhs (stmt); @@ -12351,6 +12358,14 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) return true; /* flavors of vec_max. */ case VSX_BUILTIN_XVMAXDP: + case ALTIVEC_BUILTIN_VMAXFP: + { + lhs = gimple_call_lhs (stmt); + tree type = TREE_TYPE (lhs); + if (HONOR_NANS (type)) + return false; + gcc_fallthrough (); + } case P8V_BUILTIN_VMAXSD: case P8V_BUILTIN_VMAXUD: case ALTIVEC_BUILTIN_VMAXSB: @@ -12359,7 +12374,6 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) case ALTIVEC_BUILTIN_VMAXUB: case ALTIVEC_BUILTIN_VMAXUH: case ALTIVEC_BUILTIN_VMAXUW: - case ALTIVEC_BUILTIN_VMAXFP: arg0 = gimple_call_arg (stmt, 0); arg1 = gimple_call_arg (stmt, 1); lhs = gimple_call_lhs (stmt); diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c new file mode 100644 index 00000000000..e238659c9be --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c @@ -0,0 +1,52 @@ +/* { dg-require-effective-target powerpc_p9vector_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power9" } */ +/* { dg-final { scan-assembler-times {\mxvmaxdp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvmaxsp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvmindp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvminsp\M} 1 } } */ + +/* This test verifies that float or double vec_min/max are bound to + xv[min|max][d|s]p instructions when fast-math is not set. */ + + +#include <altivec.h> + +#ifdef _BIG_ENDIAN + const int PREF_D = 0; +#else + const int PREF_D = 1; +#endif + +double vmaxd (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_max (va, vb), PREF_D); +} + +double vmind (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_min (va, vb), PREF_D); +} + +#ifdef _BIG_ENDIAN + const int PREF_F = 0; +#else + const int PREF_F = 3; +#endif + +float vmaxf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_max (va, vb), PREF_F); +} + +float vminf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_min (va, vb), PREF_F); +} diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c new file mode 100644 index 00000000000..149275d8709 --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c @@ -0,0 +1,50 @@ +/* { dg-require-effective-target powerpc_p9vector_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power9 -ffast-math" } */ +/* { dg-final { scan-assembler-times {\mxsmaxcdp\M} 2 } } */ +/* { dg-final { scan-assembler-times {\mxsmincdp\M} 2 } } */ + +/* This test verifies that float or double vec_min/max can be converted + to scalar comparison when fast-math is set. */ + + +#include <altivec.h> + +#ifdef _BIG_ENDIAN + const int PREF_D = 0; +#else + const int PREF_D = 1; +#endif + +double vmaxd (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_max (va, vb), PREF_D); +} + +double vmind (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_min (va, vb), PREF_D); +} + +#ifdef _BIG_ENDIAN + const int PREF_F = 0; +#else + const int PREF_F = 3; +#endif + +float vmaxf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_max (va, vb), PREF_F); +} + +float vminf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_min (va, vb), PREF_F); +} On 2/11/2021 上午 8:21, David Edelsohn wrote: > Hi, Hao > > Neither the inlined patch nor the attached patch seem to contain the > change to rs6000-call.c. I only see the new testcases. > > Please resend the complete patch. > > Thanks David > > On Mon, Nov 1, 2021 at 2:48 AM HAO CHEN GUI <guihaoc@linux.ibm.com> wrote: >> Hi, >> >> This patch disables gimple folding for VSX_BUILTIN_XVMINDP, VSX_BUILTIN_XVMAXDP, ALTIVEC_BUILTIN_VMINFP and ALTIVEC_BUILTIN_VMAXFP when fast-math is not set. With the gimple folding is enabled, the four built-ins will be implemented by c-type instructions - xs[min|max]cdp on P9 and P10 if they can be converted to scalar comparisons. While they are implemented by xv[min|max][s|d]p on P8 and P7 as P8 and P7 don't have corresponding scalar comparison instructions. The patch binds these four built-ins to xv[min|max][s|d]p when fast-math is not set. The two new test cases illustrate it. >> >> ALTIVEC_BUILTIN_VMINFP and ALTIVEC_BUILTIN_VMAXFP are not implemented by vminfp or vmaxfp. >> >> rs6000-builtin.def:BU_ALTIVEC_2 (VMAXFP, "vmaxfp", CONST, smaxv4sf3) >> >> rs6000-builtin.def:BU_ALTIVEC_2 (VMINFP, "vminfp", CONST, sminv4sf3) >> >> Bootstrapped and tested on powerpc64le-linux with no regressions. Is this okay for trunk? Any recommendations? Thanks a lot. >> >> >> ChangeLog >> >> 2021-11-01 Haochen Gui <guihaoc@linux.ibm.com> >> >> gcc/ >> * config/rs6000/rs6000-call.c (rs6000_gimple_fold_builtin): Disable >> gimple fold for VSX_BUILTIN_XVMINDP, ALTIVEC_BUILTIN_VMINFP, >> VSX_BUILTIN_XVMAXDP, ALTIVEC_BUILTIN_VMAXFP when fast-math is not >> set. >> >> gcc/testsuite/ >> * gcc.target/powerpc/vec-minmax-1.c: New test. >> * gcc.target/powerpc/vec-minmax-2.c: Likewise. >> >> >> patch.diff >> >> diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c >> new file mode 100644 >> index 00000000000..e238659c9be >> --- /dev/null >> +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c >> @@ -0,0 +1,52 @@ >> +/* { dg-require-effective-target powerpc_p9vector_ok } */ >> +/* { dg-options "-O2 -mdejagnu-cpu=power9" } */ >> +/* { dg-final { scan-assembler-times {\mxvmaxdp\M} 1 } } */ >> +/* { dg-final { scan-assembler-times {\mxvmaxsp\M} 1 } } */ >> +/* { dg-final { scan-assembler-times {\mxvmindp\M} 1 } } */ >> +/* { dg-final { scan-assembler-times {\mxvminsp\M} 1 } } */ >> + >> +/* This test verifies that float or double vec_min/max are bound to >> + xv[min|max][d|s]p instructions when fast-math is not set. */ >> + >> + >> +#include <altivec.h> >> + >> +#ifdef _BIG_ENDIAN >> + const int PREF_D = 0; >> +#else >> + const int PREF_D = 1; >> +#endif >> + >> +double vmaxd (double a, double b) >> +{ >> + vector double va = vec_promote (a, PREF_D); >> + vector double vb = vec_promote (b, PREF_D); >> + return vec_extract (vec_max (va, vb), PREF_D); >> +} >> + >> +double vmind (double a, double b) >> +{ >> + vector double va = vec_promote (a, PREF_D); >> + vector double vb = vec_promote (b, PREF_D); >> + return vec_extract (vec_min (va, vb), PREF_D); >> +} >> + >> +#ifdef _BIG_ENDIAN >> + const int PREF_F = 0; >> +#else >> + const int PREF_F = 3; >> +#endif >> + >> +float vmaxf (float a, float b) >> +{ >> + vector float va = vec_promote (a, PREF_F); >> + vector float vb = vec_promote (b, PREF_F); >> + return vec_extract (vec_max (va, vb), PREF_F); >> +} >> + >> +float vminf (float a, float b) >> +{ >> + vector float va = vec_promote (a, PREF_F); >> + vector float vb = vec_promote (b, PREF_F); >> + return vec_extract (vec_min (va, vb), PREF_F); >> +} >> diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c >> new file mode 100644 >> index 00000000000..149275d8709 >> --- /dev/null >> +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c >> @@ -0,0 +1,50 @@ >> +/* { dg-require-effective-target powerpc_p9vector_ok } */ >> +/* { dg-options "-O2 -mdejagnu-cpu=power9 -ffast-math" } */ >> +/* { dg-final { scan-assembler-times {\mxsmaxcdp\M} 2 } } */ >> +/* { dg-final { scan-assembler-times {\mxsmincdp\M} 2 } } */ >> + >> +/* This test verifies that float or double vec_min/max can be converted >> + to scalar comparison when fast-math is set. */ >> + >> + >> +#include <altivec.h> >> + >> +#ifdef _BIG_ENDIAN >> + const int PREF_D = 0; >> +#else >> + const int PREF_D = 1; >> +#endif >> + >> +double vmaxd (double a, double b) >> +{ >> + vector double va = vec_promote (a, PREF_D); >> + vector double vb = vec_promote (b, PREF_D); >> + return vec_extract (vec_max (va, vb), PREF_D); >> +} >> + >> +double vmind (double a, double b) >> +{ >> + vector double va = vec_promote (a, PREF_D); >> + vector double vb = vec_promote (b, PREF_D); >> + return vec_extract (vec_min (va, vb), PREF_D); >> +} >> + >> +#ifdef _BIG_ENDIAN >> + const int PREF_F = 0; >> +#else >> + const int PREF_F = 3; >> +#endif >> + >> +float vmaxf (float a, float b) >> +{ >> + vector float va = vec_promote (a, PREF_F); >> + vector float vb = vec_promote (b, PREF_F); >> + return vec_extract (vec_max (va, vb), PREF_F); >> +} >> + >> +float vminf (float a, float b) >> +{ >> + vector float va = vec_promote (a, PREF_F); >> + vector float vb = vec_promote (b, PREF_F); >> + return vec_extract (vec_min (va, vb), PREF_F); >> +} 2021-11-01 Haochen Gui <guihaoc@linux.ibm.com> gcc/ * config/rs6000/rs6000-call.c (rs6000_gimple_fold_builtin): Disable gimple fold for VSX_BUILTIN_XVMINDP, ALTIVEC_BUILTIN_VMINFP, VSX_BUILTIN_XVMAXDP, ALTIVEC_BUILTIN_VMAXFP when fast-math is not set. gcc/testsuite/ * gcc.target/powerpc/vec-minmax-1.c: New test. * gcc.target/powerpc/vec-minmax-2.c: Likewise. diff --git a/gcc/config/rs6000/rs6000-call.c b/gcc/config/rs6000/rs6000-call.c index 7d485480225..a8e193a0089 100644 --- a/gcc/config/rs6000/rs6000-call.c +++ b/gcc/config/rs6000/rs6000-call.c @@ -12333,6 +12333,14 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) return true; /* flavors of vec_min. */ case VSX_BUILTIN_XVMINDP: + case ALTIVEC_BUILTIN_VMINFP: + { + lhs = gimple_call_lhs (stmt); + tree type = TREE_TYPE (lhs); + if (HONOR_NANS (type)) + return false; + gcc_fallthrough (); + } case P8V_BUILTIN_VMINSD: case P8V_BUILTIN_VMINUD: case ALTIVEC_BUILTIN_VMINSB: @@ -12341,7 +12349,6 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) case ALTIVEC_BUILTIN_VMINUB: case ALTIVEC_BUILTIN_VMINUH: case ALTIVEC_BUILTIN_VMINUW: - case ALTIVEC_BUILTIN_VMINFP: arg0 = gimple_call_arg (stmt, 0); arg1 = gimple_call_arg (stmt, 1); lhs = gimple_call_lhs (stmt); @@ -12351,6 +12358,14 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) return true; /* flavors of vec_max. */ case VSX_BUILTIN_XVMAXDP: + case ALTIVEC_BUILTIN_VMAXFP: + { + lhs = gimple_call_lhs (stmt); + tree type = TREE_TYPE (lhs); + if (HONOR_NANS (type)) + return false; + gcc_fallthrough (); + } case P8V_BUILTIN_VMAXSD: case P8V_BUILTIN_VMAXUD: case ALTIVEC_BUILTIN_VMAXSB: @@ -12359,7 +12374,6 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) case ALTIVEC_BUILTIN_VMAXUB: case ALTIVEC_BUILTIN_VMAXUH: case ALTIVEC_BUILTIN_VMAXUW: - case ALTIVEC_BUILTIN_VMAXFP: arg0 = gimple_call_arg (stmt, 0); arg1 = gimple_call_arg (stmt, 1); lhs = gimple_call_lhs (stmt); diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c new file mode 100644 index 00000000000..e238659c9be --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c @@ -0,0 +1,52 @@ +/* { dg-require-effective-target powerpc_p9vector_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power9" } */ +/* { dg-final { scan-assembler-times {\mxvmaxdp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvmaxsp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvmindp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvminsp\M} 1 } } */ + +/* This test verifies that float or double vec_min/max are bound to + xv[min|max][d|s]p instructions when fast-math is not set. */ + + +#include <altivec.h> + +#ifdef _BIG_ENDIAN + const int PREF_D = 0; +#else + const int PREF_D = 1; +#endif + +double vmaxd (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_max (va, vb), PREF_D); +} + +double vmind (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_min (va, vb), PREF_D); +} + +#ifdef _BIG_ENDIAN + const int PREF_F = 0; +#else + const int PREF_F = 3; +#endif + +float vmaxf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_max (va, vb), PREF_F); +} + +float vminf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_min (va, vb), PREF_F); +} diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c new file mode 100644 index 00000000000..149275d8709 --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c @@ -0,0 +1,50 @@ +/* { dg-require-effective-target powerpc_p9vector_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power9 -ffast-math" } */ +/* { dg-final { scan-assembler-times {\mxsmaxcdp\M} 2 } } */ +/* { dg-final { scan-assembler-times {\mxsmincdp\M} 2 } } */ + +/* This test verifies that float or double vec_min/max can be converted + to scalar comparison when fast-math is set. */ + + +#include <altivec.h> + +#ifdef _BIG_ENDIAN + const int PREF_D = 0; +#else + const int PREF_D = 1; +#endif + +double vmaxd (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_max (va, vb), PREF_D); +} + +double vmind (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_min (va, vb), PREF_D); +} + +#ifdef _BIG_ENDIAN + const int PREF_F = 0; +#else + const int PREF_F = 3; +#endif + +float vmaxf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_max (va, vb), PREF_F); +} + +float vminf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_min (va, vb), PREF_F); +}
On Mon, Nov 1, 2021 at 10:40 PM HAO CHEN GUI <guihaoc@linux.ibm.com> wrote: > > David, > > My patch file was broken. I am sorry for it. Here is the correct one. Thanks a lot. > > ChangeLog > > 2021-11-01 Haochen Gui <guihaoc@linux.ibm.com> > > gcc/ > * config/rs6000/rs6000-call.c (rs6000_gimple_fold_builtin): Disable > gimple fold for VSX_BUILTIN_XVMINDP, ALTIVEC_BUILTIN_VMINFP, > VSX_BUILTIN_XVMAXDP, ALTIVEC_BUILTIN_VMAXFP when fast-math is not > set. > > gcc/testsuite/ > * gcc.target/powerpc/vec-minmax-1.c: New test. > * gcc.target/powerpc/vec-minmax-2.c: Likewise. This is okay. The default DejaGNU test action is compile, but it's a good idea to include the dg-do line to be clear and document the intention. Thanks, David
David, Thanks for your comments. I refined it according to your advice. ChangeLog 2021-11-01 Haochen Gui <guihaoc@linux.ibm.com> gcc/ * config/rs6000/rs6000-call.c (rs6000_gimple_fold_builtin): Disable gimple fold for VSX_BUILTIN_XVMINDP, ALTIVEC_BUILTIN_VMINFP, VSX_BUILTIN_XVMAXDP, ALTIVEC_BUILTIN_VMAXFP when fast-math is not set. gcc/testsuite/ * gcc.target/powerpc/vec-minmax-1.c: New test. * gcc.target/powerpc/vec-minmax-2.c: Likewise. patch.diff diff --git a/gcc/config/rs6000/rs6000-call.c b/gcc/config/rs6000/rs6000-call.c index 7d485480225..a8e193a0089 100644 --- a/gcc/config/rs6000/rs6000-call.c +++ b/gcc/config/rs6000/rs6000-call.c @@ -12333,6 +12333,14 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) return true; /* flavors of vec_min. */ case VSX_BUILTIN_XVMINDP: + case ALTIVEC_BUILTIN_VMINFP: + { + lhs = gimple_call_lhs (stmt); + tree type = TREE_TYPE (lhs); + if (HONOR_NANS (type)) + return false; + gcc_fallthrough (); + } case P8V_BUILTIN_VMINSD: case P8V_BUILTIN_VMINUD: case ALTIVEC_BUILTIN_VMINSB: @@ -12341,7 +12349,6 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) case ALTIVEC_BUILTIN_VMINUB: case ALTIVEC_BUILTIN_VMINUH: case ALTIVEC_BUILTIN_VMINUW: - case ALTIVEC_BUILTIN_VMINFP: arg0 = gimple_call_arg (stmt, 0); arg1 = gimple_call_arg (stmt, 1); lhs = gimple_call_lhs (stmt); @@ -12351,6 +12358,14 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) return true; /* flavors of vec_max. */ case VSX_BUILTIN_XVMAXDP: + case ALTIVEC_BUILTIN_VMAXFP: + { + lhs = gimple_call_lhs (stmt); + tree type = TREE_TYPE (lhs); + if (HONOR_NANS (type)) + return false; + gcc_fallthrough (); + } case P8V_BUILTIN_VMAXSD: case P8V_BUILTIN_VMAXUD: case ALTIVEC_BUILTIN_VMAXSB: @@ -12359,7 +12374,6 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) case ALTIVEC_BUILTIN_VMAXUB: case ALTIVEC_BUILTIN_VMAXUH: case ALTIVEC_BUILTIN_VMAXUW: - case ALTIVEC_BUILTIN_VMAXFP: arg0 = gimple_call_arg (stmt, 0); arg1 = gimple_call_arg (stmt, 1); lhs = gimple_call_lhs (stmt); diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c new file mode 100644 index 00000000000..3f6e3760075 --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c @@ -0,0 +1,53 @@ +/* { dg-do compile } */ +/* { dg-require-effective-target powerpc_p9vector_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power9" } */ +/* { dg-final { scan-assembler-times {\mxvmaxdp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvmaxsp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvmindp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvminsp\M} 1 } } */ + +/* This test verifies that float or double vec_min/max are bound to + xv[min|max][d|s]p instructions when fast-math is not set. */ + + +#include <altivec.h> + +#ifdef _BIG_ENDIAN + const int PREF_D = 0; +#else + const int PREF_D = 1; +#endif + +double vmaxd (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_max (va, vb), PREF_D); +} + +double vmind (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_min (va, vb), PREF_D); +} + +#ifdef _BIG_ENDIAN + const int PREF_F = 0; +#else + const int PREF_F = 3; +#endif + +float vmaxf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_max (va, vb), PREF_F); +} + +float vminf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_min (va, vb), PREF_F); +} diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c new file mode 100644 index 00000000000..b27bebba846 --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c @@ -0,0 +1,51 @@ +/* { dg-do compile } */ +/* { dg-require-effective-target powerpc_p9vector_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power9 -ffast-math" } */ +/* { dg-final { scan-assembler-times {\mxsmaxcdp\M} 2 } } */ +/* { dg-final { scan-assembler-times {\mxsmincdp\M} 2 } } */ + +/* This test verifies that float or double vec_min/max can be converted + to scalar comparison when fast-math is set. */ + + +#include <altivec.h> + +#ifdef _BIG_ENDIAN + const int PREF_D = 0; +#else + const int PREF_D = 1; +#endif + +double vmaxd (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_max (va, vb), PREF_D); +} + +double vmind (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_min (va, vb), PREF_D); +} + +#ifdef _BIG_ENDIAN + const int PREF_F = 0; +#else + const int PREF_F = 3; +#endif + +float vmaxf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_max (va, vb), PREF_F); +} + +float vminf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_min (va, vb), PREF_F); +} On 2/11/2021 下午 9:12, David Edelsohn wrote: > On Mon, Nov 1, 2021 at 10:40 PM HAO CHEN GUI <guihaoc@linux.ibm.com> wrote: >> David, >> >> My patch file was broken. I am sorry for it. Here is the correct one. Thanks a lot. >> >> ChangeLog >> >> 2021-11-01 Haochen Gui <guihaoc@linux.ibm.com> >> >> gcc/ >> * config/rs6000/rs6000-call.c (rs6000_gimple_fold_builtin): Disable >> gimple fold for VSX_BUILTIN_XVMINDP, ALTIVEC_BUILTIN_VMINFP, >> VSX_BUILTIN_XVMAXDP, ALTIVEC_BUILTIN_VMAXFP when fast-math is not >> set. >> >> gcc/testsuite/ >> * gcc.target/powerpc/vec-minmax-1.c: New test. >> * gcc.target/powerpc/vec-minmax-2.c: Likewise. > This is okay. > > The default DejaGNU test action is compile, but it's a good idea to > include the dg-do line to be clear and document the intention. > > Thanks, David 2021-11-01 Haochen Gui <guihaoc@linux.ibm.com> gcc/ * config/rs6000/rs6000-call.c (rs6000_gimple_fold_builtin): Disable gimple fold for VSX_BUILTIN_XVMINDP, ALTIVEC_BUILTIN_VMINFP, VSX_BUILTIN_XVMAXDP, ALTIVEC_BUILTIN_VMAXFP when fast-math is not set. gcc/testsuite/ * gcc.target/powerpc/vec-minmax-1.c: New test. * gcc.target/powerpc/vec-minmax-2.c: Likewise. diff --git a/gcc/config/rs6000/rs6000-call.c b/gcc/config/rs6000/rs6000-call.c index 7d485480225..a8e193a0089 100644 --- a/gcc/config/rs6000/rs6000-call.c +++ b/gcc/config/rs6000/rs6000-call.c @@ -12333,6 +12333,14 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) return true; /* flavors of vec_min. */ case VSX_BUILTIN_XVMINDP: + case ALTIVEC_BUILTIN_VMINFP: + { + lhs = gimple_call_lhs (stmt); + tree type = TREE_TYPE (lhs); + if (HONOR_NANS (type)) + return false; + gcc_fallthrough (); + } case P8V_BUILTIN_VMINSD: case P8V_BUILTIN_VMINUD: case ALTIVEC_BUILTIN_VMINSB: @@ -12341,7 +12349,6 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) case ALTIVEC_BUILTIN_VMINUB: case ALTIVEC_BUILTIN_VMINUH: case ALTIVEC_BUILTIN_VMINUW: - case ALTIVEC_BUILTIN_VMINFP: arg0 = gimple_call_arg (stmt, 0); arg1 = gimple_call_arg (stmt, 1); lhs = gimple_call_lhs (stmt); @@ -12351,6 +12358,14 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) return true; /* flavors of vec_max. */ case VSX_BUILTIN_XVMAXDP: + case ALTIVEC_BUILTIN_VMAXFP: + { + lhs = gimple_call_lhs (stmt); + tree type = TREE_TYPE (lhs); + if (HONOR_NANS (type)) + return false; + gcc_fallthrough (); + } case P8V_BUILTIN_VMAXSD: case P8V_BUILTIN_VMAXUD: case ALTIVEC_BUILTIN_VMAXSB: @@ -12359,7 +12374,6 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi) case ALTIVEC_BUILTIN_VMAXUB: case ALTIVEC_BUILTIN_VMAXUH: case ALTIVEC_BUILTIN_VMAXUW: - case ALTIVEC_BUILTIN_VMAXFP: arg0 = gimple_call_arg (stmt, 0); arg1 = gimple_call_arg (stmt, 1); lhs = gimple_call_lhs (stmt); diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c new file mode 100644 index 00000000000..3f6e3760075 --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c @@ -0,0 +1,53 @@ +/* { dg-do compile } */ +/* { dg-require-effective-target powerpc_p9vector_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power9" } */ +/* { dg-final { scan-assembler-times {\mxvmaxdp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvmaxsp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvmindp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvminsp\M} 1 } } */ + +/* This test verifies that float or double vec_min/max are bound to + xv[min|max][d|s]p instructions when fast-math is not set. */ + + +#include <altivec.h> + +#ifdef _BIG_ENDIAN + const int PREF_D = 0; +#else + const int PREF_D = 1; +#endif + +double vmaxd (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_max (va, vb), PREF_D); +} + +double vmind (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_min (va, vb), PREF_D); +} + +#ifdef _BIG_ENDIAN + const int PREF_F = 0; +#else + const int PREF_F = 3; +#endif + +float vmaxf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_max (va, vb), PREF_F); +} + +float vminf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_min (va, vb), PREF_F); +} diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c new file mode 100644 index 00000000000..b27bebba846 --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c @@ -0,0 +1,51 @@ +/* { dg-do compile } */ +/* { dg-require-effective-target powerpc_p9vector_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power9 -ffast-math" } */ +/* { dg-final { scan-assembler-times {\mxsmaxcdp\M} 2 } } */ +/* { dg-final { scan-assembler-times {\mxsmincdp\M} 2 } } */ + +/* This test verifies that float or double vec_min/max can be converted + to scalar comparison when fast-math is set. */ + + +#include <altivec.h> + +#ifdef _BIG_ENDIAN + const int PREF_D = 0; +#else + const int PREF_D = 1; +#endif + +double vmaxd (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_max (va, vb), PREF_D); +} + +double vmind (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_min (va, vb), PREF_D); +} + +#ifdef _BIG_ENDIAN + const int PREF_F = 0; +#else + const int PREF_F = 3; +#endif + +float vmaxf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_max (va, vb), PREF_F); +} + +float vminf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_min (va, vb), PREF_F); +}
diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c new file mode 100644 index 00000000000..e238659c9be --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-1.c @@ -0,0 +1,52 @@ +/* { dg-require-effective-target powerpc_p9vector_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power9" } */ +/* { dg-final { scan-assembler-times {\mxvmaxdp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvmaxsp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvmindp\M} 1 } } */ +/* { dg-final { scan-assembler-times {\mxvminsp\M} 1 } } */ + +/* This test verifies that float or double vec_min/max are bound to + xv[min|max][d|s]p instructions when fast-math is not set. */ + + +#include <altivec.h> + +#ifdef _BIG_ENDIAN + const int PREF_D = 0; +#else + const int PREF_D = 1; +#endif + +double vmaxd (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_max (va, vb), PREF_D); +} + +double vmind (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_min (va, vb), PREF_D); +} + +#ifdef _BIG_ENDIAN + const int PREF_F = 0; +#else + const int PREF_F = 3; +#endif + +float vmaxf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_max (va, vb), PREF_F); +} + +float vminf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_min (va, vb), PREF_F); +} diff --git a/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c new file mode 100644 index 00000000000..149275d8709 --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/vec-minmax-2.c @@ -0,0 +1,50 @@ +/* { dg-require-effective-target powerpc_p9vector_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power9 -ffast-math" } */ +/* { dg-final { scan-assembler-times {\mxsmaxcdp\M} 2 } } */ +/* { dg-final { scan-assembler-times {\mxsmincdp\M} 2 } } */ + +/* This test verifies that float or double vec_min/max can be converted + to scalar comparison when fast-math is set. */ + + +#include <altivec.h> + +#ifdef _BIG_ENDIAN + const int PREF_D = 0; +#else + const int PREF_D = 1; +#endif + +double vmaxd (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_max (va, vb), PREF_D); +} + +double vmind (double a, double b) +{ + vector double va = vec_promote (a, PREF_D); + vector double vb = vec_promote (b, PREF_D); + return vec_extract (vec_min (va, vb), PREF_D); +} + +#ifdef _BIG_ENDIAN + const int PREF_F = 0; +#else + const int PREF_F = 3; +#endif + +float vmaxf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_max (va, vb), PREF_F); +} + +float vminf (float a, float b) +{ + vector float va = vec_promote (a, PREF_F); + vector float vb = vec_promote (b, PREF_F); + return vec_extract (vec_min (va, vb), PREF_F); +}