Message ID | 20221128130539.2124727-1-hongtao.liu@intel.com |
---|---|
State | New |
Headers | show |
Series | [x86] Fix unrecognizable insn due to illegal immediate_operand (const_int 255) of QImode. | expand |
On Mon, Nov 28, 2022 at 9:06 PM liuhongt <hongtao.liu@intel.com> wrote: > > For __builtin_ia32_vec_set_v16qi (a, -1, 2) with > !flag_signed_char. it's transformed to > __builtin_ia32_vec_set_v16qi (_4, 255, 2) in the gimple, > and expanded to (const_int 255) in the rtl. But for immediate_operand, > it expects (const_int 255) to be signed extended to > (const_int -1). The mismatch caused an unrecognizable insn error. > > expand_expr_real_1 generates (const_int 255) without considering the target mode. > I guess it's on purpose, so I'll leave that alone and only change the expander > in the backend. After applying convert_modes to (const_int 255), > it's transformed to (const_int -1) which fix the issue. > > Bootstrapped and regtested x86_64-pc-linux-gnu{-m32,}. > Ok for trunk(and backport to GCC-10/11/12 release branches)? Drop this patch since it's not a complete solution, there're also other QI builtins which is not handled. > > gcc/ChangeLog: > > PR target/107863 > * config/i386/i386-expand.cc (ix86_expand_vec_set_builtin): > Convert op1 to target mode whenever mode mismatch. > > gcc/testsuite/ChangeLog: > > * gcc.target/i386/pr107863.c: New test. > --- > gcc/config/i386/i386-expand.cc | 2 +- > gcc/testsuite/gcc.target/i386/pr107863.c | 8 ++++++++ > 2 files changed, 9 insertions(+), 1 deletion(-) > create mode 100644 gcc/testsuite/gcc.target/i386/pr107863.c > > diff --git a/gcc/config/i386/i386-expand.cc b/gcc/config/i386/i386-expand.cc > index 0373c3614a4..c639ee3a9f7 100644 > --- a/gcc/config/i386/i386-expand.cc > +++ b/gcc/config/i386/i386-expand.cc > @@ -12475,7 +12475,7 @@ ix86_expand_vec_set_builtin (tree exp) > op1 = expand_expr (arg1, NULL_RTX, mode1, EXPAND_NORMAL); > elt = get_element_number (TREE_TYPE (arg0), arg2); > > - if (GET_MODE (op1) != mode1 && GET_MODE (op1) != VOIDmode) > + if (GET_MODE (op1) != mode1) > op1 = convert_modes (mode1, GET_MODE (op1), op1, true); > > op0 = force_reg (tmode, op0); > diff --git a/gcc/testsuite/gcc.target/i386/pr107863.c b/gcc/testsuite/gcc.target/i386/pr107863.c > new file mode 100644 > index 00000000000..99fd85d9765 > --- /dev/null > +++ b/gcc/testsuite/gcc.target/i386/pr107863.c > @@ -0,0 +1,8 @@ > +/* { dg-do compile } */ > +/* { dg-options "-mavx2 -O" } */ > + > +typedef char v16qi __attribute__((vector_size(16))); > + > +v16qi foo(v16qi a){ > + return __builtin_ia32_vec_set_v16qi (a, -1, 2); > +} > -- > 2.27.0 >
On Mon, Nov 28, 2022 at 11:04 PM Hongtao Liu <crazylht@gmail.com> wrote: > > On Mon, Nov 28, 2022 at 9:06 PM liuhongt <hongtao.liu@intel.com> wrote: > > > > For __builtin_ia32_vec_set_v16qi (a, -1, 2) with > > !flag_signed_char. it's transformed to > > __builtin_ia32_vec_set_v16qi (_4, 255, 2) in the gimple, > > and expanded to (const_int 255) in the rtl. But for immediate_operand, > > it expects (const_int 255) to be signed extended to > > (const_int -1). The mismatch caused an unrecognizable insn error. > > > > expand_expr_real_1 generates (const_int 255) without considering the target mode. > > I guess it's on purpose, so I'll leave that alone and only change the expander > > in the backend. After applying convert_modes to (const_int 255), > > it's transformed to (const_int -1) which fix the issue. > > > > Bootstrapped and regtested x86_64-pc-linux-gnu{-m32,}. > > Ok for trunk(and backport to GCC-10/11/12 release branches)? > Drop this patch since it's not a complete solution, there're also > other QI builtins which is not handled. I checked the x86 backend. __builtin_ia32_vec_set_v16qi is the only intrinsic with this issue. > > > > gcc/ChangeLog: > > > > PR target/107863 > > * config/i386/i386-expand.cc (ix86_expand_vec_set_builtin): > > Convert op1 to target mode whenever mode mismatch. > > > > gcc/testsuite/ChangeLog: > > > > * gcc.target/i386/pr107863.c: New test. > > --- > > gcc/config/i386/i386-expand.cc | 2 +- > > gcc/testsuite/gcc.target/i386/pr107863.c | 8 ++++++++ > > 2 files changed, 9 insertions(+), 1 deletion(-) > > create mode 100644 gcc/testsuite/gcc.target/i386/pr107863.c > > > > diff --git a/gcc/config/i386/i386-expand.cc b/gcc/config/i386/i386-expand.cc > > index 0373c3614a4..c639ee3a9f7 100644 > > --- a/gcc/config/i386/i386-expand.cc > > +++ b/gcc/config/i386/i386-expand.cc > > @@ -12475,7 +12475,7 @@ ix86_expand_vec_set_builtin (tree exp) > > op1 = expand_expr (arg1, NULL_RTX, mode1, EXPAND_NORMAL); > > elt = get_element_number (TREE_TYPE (arg0), arg2); > > > > - if (GET_MODE (op1) != mode1 && GET_MODE (op1) != VOIDmode) > > + if (GET_MODE (op1) != mode1) > > op1 = convert_modes (mode1, GET_MODE (op1), op1, true); > > > > op0 = force_reg (tmode, op0); > > diff --git a/gcc/testsuite/gcc.target/i386/pr107863.c b/gcc/testsuite/gcc.target/i386/pr107863.c > > new file mode 100644 > > index 00000000000..99fd85d9765 > > --- /dev/null > > +++ b/gcc/testsuite/gcc.target/i386/pr107863.c > > @@ -0,0 +1,8 @@ > > +/* { dg-do compile } */ > > +/* { dg-options "-mavx2 -O" } */ > > + > > +typedef char v16qi __attribute__((vector_size(16))); > > + > > +v16qi foo(v16qi a){ > > + return __builtin_ia32_vec_set_v16qi (a, -1, 2); > > +} > > -- > > 2.27.0 > > > > > -- > BR, > Hongtao
On Wed, Nov 30, 2022 at 3:12 AM H.J. Lu <hjl.tools@gmail.com> wrote: > > On Mon, Nov 28, 2022 at 11:04 PM Hongtao Liu <crazylht@gmail.com> wrote: > > > > On Mon, Nov 28, 2022 at 9:06 PM liuhongt <hongtao.liu@intel.com> wrote: > > > > > > For __builtin_ia32_vec_set_v16qi (a, -1, 2) with > > > !flag_signed_char. it's transformed to > > > __builtin_ia32_vec_set_v16qi (_4, 255, 2) in the gimple, > > > and expanded to (const_int 255) in the rtl. But for immediate_operand, > > > it expects (const_int 255) to be signed extended to > > > (const_int -1). The mismatch caused an unrecognizable insn error. > > > > > > expand_expr_real_1 generates (const_int 255) without considering the target mode. > > > I guess it's on purpose, so I'll leave that alone and only change the expander > > > in the backend. After applying convert_modes to (const_int 255), > > > it's transformed to (const_int -1) which fix the issue. > > > > > > Bootstrapped and regtested x86_64-pc-linux-gnu{-m32,}. > > > Ok for trunk(and backport to GCC-10/11/12 release branches)? > > Drop this patch since it's not a complete solution, there're also > > other QI builtins which is not handled. > > I checked the x86 backend. __builtin_ia32_vec_set_v16qi is the > only intrinsic with this issue. Ok, I'll commit the patch. > > > > > > > gcc/ChangeLog: > > > > > > PR target/107863 > > > * config/i386/i386-expand.cc (ix86_expand_vec_set_builtin): > > > Convert op1 to target mode whenever mode mismatch. > > > > > > gcc/testsuite/ChangeLog: > > > > > > * gcc.target/i386/pr107863.c: New test. > > > --- > > > gcc/config/i386/i386-expand.cc | 2 +- > > > gcc/testsuite/gcc.target/i386/pr107863.c | 8 ++++++++ > > > 2 files changed, 9 insertions(+), 1 deletion(-) > > > create mode 100644 gcc/testsuite/gcc.target/i386/pr107863.c > > > > > > diff --git a/gcc/config/i386/i386-expand.cc b/gcc/config/i386/i386-expand.cc > > > index 0373c3614a4..c639ee3a9f7 100644 > > > --- a/gcc/config/i386/i386-expand.cc > > > +++ b/gcc/config/i386/i386-expand.cc > > > @@ -12475,7 +12475,7 @@ ix86_expand_vec_set_builtin (tree exp) > > > op1 = expand_expr (arg1, NULL_RTX, mode1, EXPAND_NORMAL); > > > elt = get_element_number (TREE_TYPE (arg0), arg2); > > > > > > - if (GET_MODE (op1) != mode1 && GET_MODE (op1) != VOIDmode) > > > + if (GET_MODE (op1) != mode1) > > > op1 = convert_modes (mode1, GET_MODE (op1), op1, true); > > > > > > op0 = force_reg (tmode, op0); > > > diff --git a/gcc/testsuite/gcc.target/i386/pr107863.c b/gcc/testsuite/gcc.target/i386/pr107863.c > > > new file mode 100644 > > > index 00000000000..99fd85d9765 > > > --- /dev/null > > > +++ b/gcc/testsuite/gcc.target/i386/pr107863.c > > > @@ -0,0 +1,8 @@ > > > +/* { dg-do compile } */ > > > +/* { dg-options "-mavx2 -O" } */ > > > + > > > +typedef char v16qi __attribute__((vector_size(16))); > > > + > > > +v16qi foo(v16qi a){ > > > + return __builtin_ia32_vec_set_v16qi (a, -1, 2); > > > +} > > > -- > > > 2.27.0 > > > > > > > > > -- > > BR, > > Hongtao > > > > -- > H.J.
diff --git a/gcc/config/i386/i386-expand.cc b/gcc/config/i386/i386-expand.cc index 0373c3614a4..c639ee3a9f7 100644 --- a/gcc/config/i386/i386-expand.cc +++ b/gcc/config/i386/i386-expand.cc @@ -12475,7 +12475,7 @@ ix86_expand_vec_set_builtin (tree exp) op1 = expand_expr (arg1, NULL_RTX, mode1, EXPAND_NORMAL); elt = get_element_number (TREE_TYPE (arg0), arg2); - if (GET_MODE (op1) != mode1 && GET_MODE (op1) != VOIDmode) + if (GET_MODE (op1) != mode1) op1 = convert_modes (mode1, GET_MODE (op1), op1, true); op0 = force_reg (tmode, op0); diff --git a/gcc/testsuite/gcc.target/i386/pr107863.c b/gcc/testsuite/gcc.target/i386/pr107863.c new file mode 100644 index 00000000000..99fd85d9765 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr107863.c @@ -0,0 +1,8 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx2 -O" } */ + +typedef char v16qi __attribute__((vector_size(16))); + +v16qi foo(v16qi a){ + return __builtin_ia32_vec_set_v16qi (a, -1, 2); +}