Hi, When comparing two vectors, the type of vector was used as the result of the condition result. This meant that for floating point comparisons, each value would either be `0.0' or `-1.0' reinterpreted as an integer, not the expected integral bitmask values `0' and `-1'.
Instead, use the comparison type determined by truth_type_for as the result of the comparison. If a reinterpret is later required by the final conversion for generating CmpExp, it is still only going to reinterpret one integer kind as another. Bootstrapped and regression tested on x86_64-linux-gnu/-m32, and committed to mainline. Regards, Iain. --- PR d/108945 gcc/d/ChangeLog: * d-codegen.cc (build_boolop): Evaluate vector comparison as the truth_type_for vector type. gcc/testsuite/ChangeLog: * gdc.dg/pr108945.d: New test. --- gcc/d/d-codegen.cc | 9 ++++----- gcc/testsuite/gdc.dg/pr108945.d | 12 ++++++++++++ 2 files changed, 16 insertions(+), 5 deletions(-) create mode 100644 gcc/testsuite/gdc.dg/pr108945.d diff --git a/gcc/d/d-codegen.cc b/gcc/d/d-codegen.cc index 0e8e07366ee..5a041927ec9 100644 --- a/gcc/d/d-codegen.cc +++ b/gcc/d/d-codegen.cc @@ -1453,13 +1453,12 @@ build_boolop (tree_code code, tree arg0, tree arg1) { /* Build a vector comparison. VEC_COND_EXPR <e1 op e2, { -1, -1, -1, -1 }, { 0, 0, 0, 0 }>; */ - tree type = TREE_TYPE (arg0); - tree cmptype = truth_type_for (type); + tree cmptype = truth_type_for (TREE_TYPE (arg0)); tree cmp = fold_build2_loc (input_location, code, cmptype, arg0, arg1); - return fold_build3_loc (input_location, VEC_COND_EXPR, type, cmp, - build_minus_one_cst (type), - build_zero_cst (type)); + return fold_build3_loc (input_location, VEC_COND_EXPR, cmptype, cmp, + build_minus_one_cst (cmptype), + build_zero_cst (cmptype)); } if (code == EQ_EXPR || code == NE_EXPR) diff --git a/gcc/testsuite/gdc.dg/pr108945.d b/gcc/testsuite/gdc.dg/pr108945.d new file mode 100644 index 00000000000..03b9de8e758 --- /dev/null +++ b/gcc/testsuite/gdc.dg/pr108945.d @@ -0,0 +1,12 @@ +// { dg-options "-fdump-tree-gimple" } +// { dg-additional-options "-mavx" { target avx_runtime } } +// { dg-do compile { target { avx_runtime || vect_sizes_16B_8B } } } + +alias f4 = __vector(float[4]); + +auto pr108945(f4 a, f4 b) +{ + return a < b; +} + +// { dg-final { scan-tree-dump-not "VEC_COND_EXPR" "gimple" } } -- 2.37.2