Running the pr104116 tests with SVE2 which implements IFN_DIV_POW2 results in
FAIL: gcc.dg/vect/pr104116-ceil-div-2.c execution test
FAIL: gcc.dg/vect/pr104116-ceil-div-pow2.c execution test
FAIL: gcc.dg/vect/pr104116-ceil-mod-2.c execution test
FAIL: gcc.dg/vect/pr104116-ceil-mod-pow2.c execution test
FAIL: gcc.dg/vect/pr104116-ceil-udiv-2.c execution test
FAIL: gcc.dg/vect/pr104116-ceil-udiv-pow2.c execution test
FAIL: gcc.dg/vect/pr104116-ceil-umod-2.c execution test
FAIL: gcc.dg/vect/pr104116-ceil-umod-pow2.c execution test
FAIL: gcc.dg/vect/pr104116-floor-div-2.c execution test
FAIL: gcc.dg/vect/pr104116-floor-div-pow2.c execution test
FAIL: gcc.dg/vect/pr104116-floor-mod-2.c execution test
FAIL: gcc.dg/vect/pr104116-floor-mod-pow2.c execution test
FAIL: gcc.dg/vect/pr104116-round-div-2.c execution test
FAIL: gcc.dg/vect/pr104116-round-div-pow2.c execution test
FAIL: gcc.dg/vect/pr104116-round-mod-2.c execution test
FAIL: gcc.dg/vect/pr104116-round-mod-pow2.c execution test
FAIL: gcc.dg/vect/pr104116-round-udiv-2.c execution test
FAIL: gcc.dg/vect/pr104116-round-udiv-pow2.c execution test
FAIL: gcc.dg/vect/pr104116-round-umod-2.c execution test
FAIL: gcc.dg/vect/pr104116-round-umod-pow2.c execution test
The first part is that add_code_for_floorceilround_divmod expects that the 6th
parameter be the remainder. This is passed correctly for most uses except for
one where instead the shifted quotient is passed instead which causes a
miscompile.
Secondly IFN_DIV_POW2 is defined for only signed types.
DEF_INTERNAL_OPTAB_FN (DIV_POW2, ECF_CONST | ECF_NOTHROW, sdiv_pow2, binary)
and the documentation says
@itemx @samp{sdiv_pow2@var{m}3}
Signed division by power-of-2 immediate. Equivalent to:
However the code does not actually check that the type is signed.
So we end up using the IFN for unsigned division as well resulting in the
failures for the unsigned testcases.
Bootstrapped Regtested on aarch64-none-linux-gnu,
arm-none-linux-gnueabihf, x86_64-pc-linux-gnu
-m32, -m64 and no issues.
Pushed.
Thanks,
Tamar
gcc/ChangeLog:
PR tree-optimization/126961
* tree-vect-patterns.cc (vect_recog_divmod_pattern): Fix remainder use
and add signedness check.
---
diff --git a/gcc/tree-vect-patterns.cc b/gcc/tree-vect-patterns.cc
index
fd1232cc75521e24bdbcfafa6d7b14a8f45e481a..ef128d75e1099f5961a2d0355d354257141c9010
100644
--- a/gcc/tree-vect-patterns.cc
+++ b/gcc/tree-vect-patterns.cc
@@ -5274,7 +5274,8 @@ vect_recog_divmod_pattern (vec_info *vinfo,
/* Check if the target supports this internal function. */
internal_fn ifn = IFN_DIV_POW2;
- if (direct_internal_fn_supported_p (ifn, vectype, OPTIMIZE_FOR_SPEED))
+ if (!TYPE_UNSIGNED (itype)
+ && direct_internal_fn_supported_p (ifn, vectype, OPTIMIZE_FOR_SPEED))
{
tree shift = build_int_cst (itype, tree_log2 (oprnd1));
@@ -5288,16 +5289,15 @@ vect_recog_divmod_pattern (vec_info *vinfo,
def_stmt
= gimple_build_assign (t1, LSHIFT_EXPR, var_div, shift);
append_pattern_def_seq (vinfo, stmt_vinfo, def_stmt);
- pattern_stmt
- = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
- MINUS_EXPR, oprnd0, t1);
+ tree r = vect_recog_temp_ssa_var (itype, NULL);
+ pattern_stmt = gimple_build_assign (r, MINUS_EXPR, oprnd0, t1);
if (is_flclrd_moddiv_p)
{
append_pattern_def_seq (vinfo, stmt_vinfo, pattern_stmt);
pattern_stmt
= add_code_for_floorceilround_divmod (vectype, vinfo,
stmt_vinfo, rhs_code,
- var_div, t1, oprnd0,
+ var_div, r, oprnd0,
oprnd1, itype);
if (pattern_stmt == NULL)
return NULL;
--
diff --git a/gcc/tree-vect-patterns.cc b/gcc/tree-vect-patterns.cc
index fd1232cc75521e24bdbcfafa6d7b14a8f45e481a..ef128d75e1099f5961a2d0355d354257141c9010 100644
--- a/gcc/tree-vect-patterns.cc
+++ b/gcc/tree-vect-patterns.cc
@@ -5274,7 +5274,8 @@ vect_recog_divmod_pattern (vec_info *vinfo,
/* Check if the target supports this internal function. */
internal_fn ifn = IFN_DIV_POW2;
- if (direct_internal_fn_supported_p (ifn, vectype, OPTIMIZE_FOR_SPEED))
+ if (!TYPE_UNSIGNED (itype)
+ && direct_internal_fn_supported_p (ifn, vectype, OPTIMIZE_FOR_SPEED))
{
tree shift = build_int_cst (itype, tree_log2 (oprnd1));
@@ -5288,16 +5289,15 @@ vect_recog_divmod_pattern (vec_info *vinfo,
def_stmt
= gimple_build_assign (t1, LSHIFT_EXPR, var_div, shift);
append_pattern_def_seq (vinfo, stmt_vinfo, def_stmt);
- pattern_stmt
- = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
- MINUS_EXPR, oprnd0, t1);
+ tree r = vect_recog_temp_ssa_var (itype, NULL);
+ pattern_stmt = gimple_build_assign (r, MINUS_EXPR, oprnd0, t1);
if (is_flclrd_moddiv_p)
{
append_pattern_def_seq (vinfo, stmt_vinfo, pattern_stmt);
pattern_stmt
= add_code_for_floorceilround_divmod (vectype, vinfo,
stmt_vinfo, rhs_code,
- var_div, t1, oprnd0,
+ var_div, r, oprnd0,
oprnd1, itype);
if (pattern_stmt == NULL)
return NULL;
