Thead has the XTHEADBB extension which has a lot of overlap with Zbb. I
made the incorrect assumption that XTHEADBS would largely be like Zbs
when generalizing Shreya's work.
As a result we can't use the operation synthesis code for IOR/XOR
because we don't have binv/bset like capabilities. I should have double
checked on XTHEADBS, my bad.
Anyway, the fix is trivial. Don't allow bset/binv based on XTHEADBS.
Already spun in my tester. Spinning in the pre-commit CI system now.
jeff
PR target/120223
gcc/
* config/riscv/riscv.cc (synthesize_ior_xor): XTHEADBS does not have
single bit manipulations.
gcc/testsuite/
* gcc.target/riscv/pr120223.c: New test.
diff --git a/gcc/config/riscv/riscv.cc b/gcc/config/riscv/riscv.cc
index 8b77a3539bc..deb0a8dc0c9 100644
--- a/gcc/config/riscv/riscv.cc
+++ b/gcc/config/riscv/riscv.cc
@@ -14229,7 +14229,7 @@ synthesize_ior_xor (rtx_code code, rtx operands[3])
{
/* Trivial cases that don't need synthesis. */
if (SMALL_OPERAND (INTVAL (operands[2]))
- || ((TARGET_ZBS || TARGET_XTHEADBS || TARGET_ZBKB)
+ || ((TARGET_ZBS || TARGET_ZBKB)
&& single_bit_mask_operand (operands[2], word_mode)))
return false;
@@ -14266,7 +14266,7 @@ synthesize_ior_xor (rtx_code code, rtx operands[3])
/* If we're flipping all but a small number of bits we can pre-flip
the outliers, then flip all the bits, which would restore those
bits that were pre-flipped. */
- if ((TARGET_ZBS || TARGET_XTHEADBS || TARGET_ZBKB)
+ if ((TARGET_ZBS || TARGET_ZBKB)
&& budget < 0
&& code == XOR
&& popcount_hwi (~INTVAL (operands[2])) < original_budget)
diff --git a/gcc/testsuite/gcc.target/riscv/pr120223.c
b/gcc/testsuite/gcc.target/riscv/pr120223.c
new file mode 100644
index 00000000000..fae21b6d1ec
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/pr120223.c
@@ -0,0 +1,4 @@
+/* { dg-do compile } */
+/* { dg-options "-mcpu=thead-c906" } */
+long foo(long x) { return x ^ 0x80000000; }
+
