[PATCH v2] RISC-V: Fix moving data from V_REGS to GR_REGS by scalar move.

When the ELEN is 32 in rv64, it can move a V2SI vector register into a DI GPR.
But the extract result in low-part must be zero-extended.

The following gimples are from pr111391-2.c:
_25 = (char) d.1_23;
_17 = {_25, _25, _25, _25, _25, _25, _25, _25};
a_26 = VIEW_CONVERT_EXPR<long int>(_17);

The final assembly:
vsetivli        zero,2,e32,m1,ta,ma
vslidedown.vi   v2,v4,1
vmv.x.s s1,v4 -> s1 must be zero-extended to prevent the bit 31 of v4[0] is 1
vmv.x.s a4,v2
slli    a5,a4,32
or      s1,a5,s1

gcc/ChangeLog:

	* config/riscv/riscv.cc (riscv_legitimize_move): Append extend.

gcc/testsuite/ChangeLog:

	* gcc.target/riscv/rvv/autovec/pr111391-2.c: Add expected asm.

diff --git a/gcc/config/riscv/riscv.cc b/gcc/config/riscv/riscv.cc
index d5de76c342e0c19fbc7fa50f546a6f242bc69750..1440d81e5ecf0cb9fd67e98e9e91348ce435f418 100644 (file)
--- a/gcc/config/riscv/riscv.cc
+++ b/gcc/config/riscv/riscv.cc
@@ -3723,6 +3723,12 @@ riscv_legitimize_move (machine_mode mode, rtx dest, rtx src)
              riscv_vector::emit_vec_extract (result, v,
                                              gen_int_mode (index + i, Pmode));
 
+             /* The low-part must be zero-extended when ELEN == 32 and
+                mode == 64.  */
+             if (num == 2 && i == 0)
+               emit_insn (gen_extend_insn (int_reg, result, mode, smode,
+                                           true));
+
              if (i == 1)
                {
                  if (UNITS_PER_WORD < mode_size)

diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr111391-2.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr111391-2.c
index 32db3a68fd37ef37cf976bbd90273fe9018dd336..e0d757a1c90415f573330b4b1102c06cf6f9d253 100644 (file)
--- a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr111391-2.c
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr111391-2.c
@@ -6,5 +6,6 @@
 /* { dg-final { scan-assembler-times {vsetivli\s+zero,\s*2,\s*e32,\s*m1,\s*t[au],\s*m[au]} 1 } } */
 /* { dg-final { scan-assembler-times {vmv\.x\.s} 2 } } */
 /* { dg-final { scan-assembler-times {vslidedown.vi\s+v[0-9]+,\s*v[0-9]+,\s*1} 1 } } */
-/* { dg-final { scan-assembler-times {slli\s+[a-x0-9]+,[a-x0-9]+,32} 1 } } */
+/* { dg-final { scan-assembler-times {slli\s+[a-x0-9]+,[a-x0-9]+,32} 2 } } */
+/* { dg-final { scan-assembler-times {srli\s+[a-x0-9]+,[a-x0-9]+,32} 1 } } */
 /* { dg-final { scan-assembler-times {or\s+[a-x0-9]+,[a-x0-9]+,[a-x0-9]+} 1 } } */