On 2025/4/5 5:09 PM, Daniel Henrique Barboza wrote:
On 3/29/25 11:44 AM, Max Chou wrote:
According to the v spec, a vector register cannot be used to provide
source
operands with more than one EEW for a single instruction.
Signed-off-by: Max Chou <max.c...@sifive.com>
---
target/riscv/insn_trans/trans_rvv.c.inc | 29 +++++++++++++++++++++++++
1 file changed, 29 insertions(+)
diff --git a/target/riscv/insn_trans/trans_rvv.c.inc
b/target/riscv/insn_trans/trans_rvv.c.inc
index e630f8661e1..70c19c49ae4 100644
--- a/target/riscv/insn_trans/trans_rvv.c.inc
+++ b/target/riscv/insn_trans/trans_rvv.c.inc
@@ -379,6 +379,35 @@ static bool vext_check_ld_index(DisasContext *s,
int vd, int vs2,
return ret;
}
+/*
+ * Check whether a vector register is used to provide source
operands with
+ * more than one EEW for the vector instruction.
+ * Returns true if the instruction has valid encoding
+ * Returns false if encoding violates the mismatched input EEWs
constraint
+ */
+static bool vext_check_input_eew(DisasContext *s, int vs1, uint8_t
eew_vs1,
+ int vs2, uint8_t eew_vs2, int vm)
+{
+ bool is_valid = true;
+ int8_t emul_vs1 = eew_vs1 - s->sew + s->lmul;
+ int8_t emul_vs2 = eew_vs2 - s->sew + s->lmul;
+
+ /* When vm is 0, vs1 & vs2(EEW!=1) group can't overlap v0
(EEW=1) */
+ if ((vs1 != -1 && !require_vm(vm, vs1)) ||
+ (vs2 != -1 && !require_vm(vm, vs2))) {
+ is_valid = false;
+ }
+
+ /* When eew_vs1 != eew_vs2, check whether vs1 and vs2 are
overlapped */
+ if ((vs1 != -1 && vs2 != -1) && (eew_vs1 != eew_vs2) &&
+ is_overlapped(vs1, 1 << MAX(emul_vs1, 0),
+ vs2, 1 << MAX(emul_vs2, 0))) {
+ is_valid = false;
+ }
+
+ return is_valid;
+}
+
Code LGTM but the patch won't compile on its own because there's no
callers for
it:
In file included from ../target/riscv/translate.c:1182:
../target/riscv/insn_trans/trans_rvv.c.inc:388:13: error:
‘vext_check_input_eew’ defined but not used [-Werror=unused-function]
388 | static bool vext_check_input_eew(DisasContext *s, int vs1,
uint8_t eew_vs1,
| ^~~~~~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors
ninja: build stopped: subcommand failed.
We want each patch to be "buildable" and with test passing to make our
lives easier
when doing bisects.
You can merge this patch with patch 4 to introduce the new function
and add its first
callers. Thanks,
Daniel
Thanks for the suggestion. Will fix this issue at v3.
Max
static bool vext_check_ss(DisasContext *s, int vd, int vs, int vm)
{
return require_vm(vm, vd) &&
