target/109944 - avoid STLF fail for V16QImode CTOR expansion

The following dispatches to V2DImode CTOR expansion instead of
using sets of (subreg:DI (reg:V16QI 146) [08]) which causes
LRA to spill DImode and reload V16QImode.  The same applies for
V8QImode or V4HImode construction from SImode parts which happens
during 32bit libgcc build.

	PR target/109944
	* config/i386/i386-expand.cc (ix86_expand_vector_init_general):
	Perform final vector composition using
	ix86_expand_vector_init_general instead of setting
	the highpart and lowpart which causes spilling.

	* gcc.target/i386/pr109944-1.c: New testcase.
	* gcc.target/i386/pr109944-2.c: Likewise.

diff --git a/gcc/config/i386/i386-expand.cc b/gcc/config/i386/i386-expand.cc
index ff3d382f1b40d0b23bc07a9f4f218374b6b20f23..19acd9c01f99b05c00351651cdb174fc04b49c6a 100644 (file)
--- a/gcc/config/i386/i386-expand.cc
+++ b/gcc/config/i386/i386-expand.cc
@@ -16367,11 +16367,12 @@ quarter:
        emit_move_insn (target, gen_lowpart (mode, words[0]));
       else if (n_words == 2)
        {
-         rtx tmp = gen_reg_rtx (mode);
-         emit_clobber (tmp);
-         emit_move_insn (gen_lowpart (tmp_mode, tmp), words[0]);
-         emit_move_insn (gen_highpart (tmp_mode, tmp), words[1]);
-         emit_move_insn (target, tmp);
+         gcc_assert (tmp_mode == DImode || tmp_mode == SImode);
+         machine_mode concat_mode = tmp_mode == DImode ? V2DImode : V2SImode;
+         rtx tmp = gen_reg_rtx (concat_mode);
+         vals = gen_rtx_PARALLEL (concat_mode, gen_rtvec_v (2, words));
+         ix86_expand_vector_init_general (false, concat_mode, tmp, vals);
+         emit_move_insn (target, gen_lowpart (mode, tmp));
        }
       else if (n_words == 4)
        {

diff --git a/gcc/testsuite/gcc.target/i386/pr109944-1.c b/gcc/testsuite/gcc.target/i386/pr109944-1.c
new file mode 100644 (file)
index 0000000..d82214d
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr109944-1.c
@@ -0,0 +1,30 @@
+/* { dg-do compile } */
+/* { dg-options "-O2" } */
+
+void foo (char * __restrict a, char *b)
+{
+  a[0] = b[0];
+  a[1] = b[16];
+  a[2] = b[32];
+  a[3] = b[48];
+  a[4] = b[64];
+  a[5] = b[80];
+  a[6] = b[96];
+  a[7] = b[112];
+  a[8] = b[128];
+  a[9] = b[144];
+  a[10] = b[160];
+  a[11] = b[176];
+  a[12] = b[192];
+  a[13] = b[208];
+  a[14] = b[224];
+  a[15] = b[240];
+}
+
+/* We do not want to generate a spill/reload for when the store is vectorized.
+        movq    %rdx, -24(%rsp)
+...
+        movq    %rax, -16(%rsp)
+        movdqa  -24(%rsp), %xmm0
+        movups  %xmm0, (%rdi)  */
+/* { dg-final { scan-assembler-not "movdq\[^\r\n\]*\[bs\]p\\\), %xmm" } } */

diff --git a/gcc/testsuite/gcc.target/i386/pr109944-2.c b/gcc/testsuite/gcc.target/i386/pr109944-2.c
new file mode 100644 (file)
index 0000000..318dfab
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr109944-2.c
@@ -0,0 +1,17 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -msse2" } */
+
+typedef char v16qi __attribute__((vector_size(16)));
+v16qi foo (char *b)
+{
+  return (v16qi){ b[0], b[16], b[32], b[48], b[64], b[80], b[96], b[112],
+      b[128], b[144], b[160], b[176], b[192], b[208], b[224], b[240] };
+}
+
+/* We do not want to generate a spill/reload
+        movq    %rdx, -24(%rsp)
+...
+        movq    %rax, -16(%rsp)
+        movdqa  -24(%rsp), %xmm0
+        movups  %xmm0, (%rdi)  */
+/* { dg-final { scan-assembler-not "movdq\[^\r\n\]*\[bs\]p\\\), %xmm" } } */