For aarch64, arm, loongarch64, mips, we can drop rotl.
For ppc, s390x we can drop rotr.
Only x86, riscv, tci have both rotl and rotr.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Many host architectures do not implement both rotate right
and rotate left and require the compiler to negate the
shift count to rotate the opposite direction. We have been
requiring the backend to perform this transformation.
Do this during opcode expansion so that the next patch
can drop support where possible in the backend.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
For TCI, we're losing type information in the interpreter.
Introduce a tci-specific opcode to handle the difference.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Rename to INDEX_op_rems to emphasize signed inputs,
and mirroring INDEX_op_remu_*.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
For TCI, we're losing type information in the interpreter.
Introduce a tci-specific opcode to handle the difference.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Rename to INDEX_op_divs2 to emphasize signed inputs,
and mirroring INDEX_op_divu2_*. Document the opcode.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
For TCI, we're losing type information in the interpreter.
Introduce a tci-specific opcode to handle the difference.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Rename to INDEX_op_divs to emphasize signed inputs,
and mirroring INDEX_op_divu_*.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
For TCI, we're losing type information in the interpreter.
Introduce a tci-specific opcode to handle the difference.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Create a special subclass for sub, because two backends can
support "subtract from immediate". Drop all backend support
for an immediate as the second operand, as we transform sub
to add during optimize.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
In 7536b82d28 we lost the rI constraint that allowed the use of
RSB to perform reg = imm - reg. At the same time, drop support
for reg = reg - imm, which is now transformed generically to
addition, and need not be handled by the backend.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The instruction set does not implement nor with immediate.
There is no reason to pretend that we do.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
At the same time, drop all backend support for immediate
operands, as we now transform orc to or during optimize.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
