In this PR we tried to create a widening multiply of two 3-bit numbers, but that isn't a widening multiply at the optab/rtl level, since both the input and output still have the same mode.
We could trap this either in is_widening_mult_p or (as the patch does) in the routines that actually ask for an optab. The latter seemed more natural since is_widening_mult_p doesn't otherwise care about modes. Tested on aarch64-linux-gnu, x86_64-linux-gnu and powerpc64-linux-gnu. OK to install? Richard 2017-11-03 Richard Sandiford <richard.sandif...@linaro.org> Alan Hayward <alan.hayw...@arm.com> David Sherwood <david.sherw...@arm.com> gcc/ PR tree-optimization/82816 * tree-ssa-math-opts.c (convert_mult_to_widen): Return false if the modes of the two types are the same. (convert_plusminus_to_widen): Likewise. gcc/testsuite/ * gcc.c-torture/compile/pr82816.c: New test. Index: gcc/tree-ssa-math-opts.c =================================================================== --- gcc/tree-ssa-math-opts.c 2017-11-01 12:29:40.203534002 +0000 +++ gcc/tree-ssa-math-opts.c 2017-11-03 11:18:03.046411241 +0000 @@ -3259,6 +3259,9 @@ convert_mult_to_widen (gimple *stmt, gim to_mode = SCALAR_INT_TYPE_MODE (type); from_mode = SCALAR_INT_TYPE_MODE (type1); + if (to_mode == from_mode) + return false; + from_unsigned1 = TYPE_UNSIGNED (type1); from_unsigned2 = TYPE_UNSIGNED (type2); @@ -3449,6 +3452,9 @@ convert_plusminus_to_widen (gimple_stmt_ to_mode = SCALAR_TYPE_MODE (type); from_mode = SCALAR_TYPE_MODE (type1); + if (to_mode == from_mode) + return false; + from_unsigned1 = TYPE_UNSIGNED (type1); from_unsigned2 = TYPE_UNSIGNED (type2); optype = type1; Index: gcc/testsuite/gcc.c-torture/compile/pr82816.c =================================================================== --- /dev/null 2017-11-03 10:40:07.002381728 +0000 +++ gcc/testsuite/gcc.c-torture/compile/pr82816.c 2017-11-03 11:18:03.045411265 +0000 @@ -0,0 +1,12 @@ +struct A +{ + int b:3; +} d, e; + +int c; + +void f () +{ + char g = d.b * e.b; + c = g; +}