| Message ID | 00e101d8740c$e785e110$b691a330$@nextmovesoftware.com |
|---|---|
| State | New |
| Headers | show |
| Series | PR rtl-optimization/7061: Complex number arguments on x86_64-like ABIs. | expand |
On 5/30/2022 4:06 AM, Roger Sayle wrote: > This patch addresses the issue in comment #6 of PR rtl-optimization/7061 > (a four digit PR number) from 2006 where on x86_64 complex number arguments > are unconditionally spilled to the stack. > > For the test cases below: > float re(float _Complex a) { return __real__ a; } > float im(float _Complex a) { return __imag__ a; } > > GCC with -O2 currently generates: > > re: movq %xmm0, -8(%rsp) > movss -8(%rsp), %xmm0 > ret > im: movq %xmm0, -8(%rsp) > movss -4(%rsp), %xmm0 > ret > > with this patch we now generate: > > re: ret > im: movq %xmm0, %rax > shrq $32, %rax > movd %eax, %xmm0 > ret > > [Technically, this shift can be performed on %xmm0 in a single > instruction, but the backend needs to be taught to do that, the > important bit is that the SCmode argument isn't written to the > stack]. > > The patch itself is to emit_group_store where just before RTL > expansion commits to writing to the stack, we check if the store > group consists of a single scalar integer register that holds > a complex mode value; on x86_64 SCmode arguments are passed in > DImode registers. If this is the case, we can use a SUBREG to > "view_convert" the integer to the equivalent complex mode. > > An interesting corner case that showed up during testing is that > x86_64 also passes HCmode arguments in DImode registers(!), i.e. > using modes of different sizes. This is easily handled/supported > by first converting to an integer mode of the correct size, and > then generating a complex mode SUBREG of this. This is similar > in concept to the patch I proposed here: > https://gcc.gnu.org/pipermail/gcc-patches/2022-February/590139.html > which was almost (but not quite) approved here: > https://gcc.gnu.org/pipermail/gcc-patches/2022-March/591139.html Yea, sorry. Too much to do at the new job. Trying to work my way through queued up stuff now... > > This patch has been tested on x86_64-pc-linux-gnu with make bootstrap > and make -k check, both with and without --target_board=unix{-m32}, > with no new failures. Ok for mainline? > > > 2020-05-30 Roger Sayle <roger@nextmovesoftware.com> > > gcc/ChangeLog > PR rtl-optimization/7061 > * expr.cc (emit_group_stote): For groups that consist of a single > scalar integer register that hold a complex mode value, use > gen_lowpart to generate a SUBREG to "view_convert" to the complex > mode. For modes of different sizes, first convert to an integer > mode of the appropriate size. > > gcc/testsuite/ChangeLog > PR rtl-optimization/7061 > * gcc.target/i386/pr7061-1.c: New test case. > * gcc.target/i386/pr7061-2.c: New test case. OK jeff
diff --git a/gcc/expr.cc b/gcc/expr.cc index 7197996..c9df206 100644 --- a/gcc/expr.cc +++ b/gcc/expr.cc @@ -2803,10 +2803,26 @@ emit_group_store (rtx orig_dst, rtx src, tree type ATTRIBUTE_UNUSED, { machine_mode dest_mode = GET_MODE (dest); machine_mode tmp_mode = GET_MODE (tmps[i]); + scalar_int_mode imode; gcc_assert (known_eq (bytepos, 0) && XVECLEN (src, 0)); - if (GET_MODE_ALIGNMENT (dest_mode) + if (finish == 1 + && REG_P (tmps[i]) + && COMPLEX_MODE_P (dest_mode) + && SCALAR_INT_MODE_P (tmp_mode) + && int_mode_for_mode (dest_mode).exists (&imode)) + { + if (tmp_mode != imode) + { + rtx tmp = gen_reg_rtx (imode); + emit_move_insn (tmp, gen_lowpart (imode, tmps[i])); + dst = gen_lowpart (dest_mode, tmp); + } + else + dst = gen_lowpart (dest_mode, tmps[i]); + } + else if (GET_MODE_ALIGNMENT (dest_mode) >= GET_MODE_ALIGNMENT (tmp_mode)) { dest = assign_stack_temp (dest_mode, diff --git a/gcc/testsuite/gcc.target/i386/pr7061-1.c b/gcc/testsuite/gcc.target/i386/pr7061-1.c new file mode 100644 index 0000000..ce5f6b2 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr7061-1.c @@ -0,0 +1,4 @@ +/* { dg-do compile { target { ! ia32 } } } */ +/* { dg-options "-O2" } */ +float re(float _Complex a) { return __real__ a; } +/* { dg-final { scan-assembler-not "mov" } } */ diff --git a/gcc/testsuite/gcc.target/i386/pr7061-2.c b/gcc/testsuite/gcc.target/i386/pr7061-2.c new file mode 100644 index 0000000..ac33340 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr7061-2.c @@ -0,0 +1,5 @@ +/* { dg-do compile { target { ! ia32 } } } */ +/* { dg-options "-O2" } */ +float im(float _Complex a) { return __imag__ a; } +/* { dg-final { scan-assembler-not "movss" } } */ +/* { dg-final { scan-assembler-not "rsp" } } */
This patch addresses the issue in comment #6 of PR rtl-optimization/7061 (a four digit PR number) from 2006 where on x86_64 complex number arguments are unconditionally spilled to the stack. For the test cases below: float re(float _Complex a) { return __real__ a; } float im(float _Complex a) { return __imag__ a; } GCC with -O2 currently generates: re: movq %xmm0, -8(%rsp) movss -8(%rsp), %xmm0 ret im: movq %xmm0, -8(%rsp) movss -4(%rsp), %xmm0 ret with this patch we now generate: re: ret im: movq %xmm0, %rax shrq $32, %rax movd %eax, %xmm0 ret [Technically, this shift can be performed on %xmm0 in a single instruction, but the backend needs to be taught to do that, the important bit is that the SCmode argument isn't written to the stack]. The patch itself is to emit_group_store where just before RTL expansion commits to writing to the stack, we check if the store group consists of a single scalar integer register that holds a complex mode value; on x86_64 SCmode arguments are passed in DImode registers. If this is the case, we can use a SUBREG to "view_convert" the integer to the equivalent complex mode. An interesting corner case that showed up during testing is that x86_64 also passes HCmode arguments in DImode registers(!), i.e. using modes of different sizes. This is easily handled/supported by first converting to an integer mode of the correct size, and then generating a complex mode SUBREG of this. This is similar in concept to the patch I proposed here: https://gcc.gnu.org/pipermail/gcc-patches/2022-February/590139.html which was almost (but not quite) approved here: https://gcc.gnu.org/pipermail/gcc-patches/2022-March/591139.html This patch has been tested on x86_64-pc-linux-gnu with make bootstrap and make -k check, both with and without --target_board=unix{-m32}, with no new failures. Ok for mainline? 2020-05-30 Roger Sayle <roger@nextmovesoftware.com> gcc/ChangeLog PR rtl-optimization/7061 * expr.cc (emit_group_stote): For groups that consist of a single scalar integer register that hold a complex mode value, use gen_lowpart to generate a SUBREG to "view_convert" to the complex mode. For modes of different sizes, first convert to an integer mode of the appropriate size. gcc/testsuite/ChangeLog PR rtl-optimization/7061 * gcc.target/i386/pr7061-1.c: New test case. * gcc.target/i386/pr7061-2.c: New test case. Thanks in advance, Roger --