aco/ra: handle empty def_reg interval in get_regs_for_copies

If def_reg is empty, then def_reg.lo() may be lower than bounds.lo() if
we're moving VGPRs and info.bounds will be invalid.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17493>

2 files changed, 18 insertions, 1 deletions

diff --git a/src/amd/compiler/aco_register_allocation.cpp b/src/amd/compiler/aco_register_allocation.cpp
index 9058b14b057..dd48d650320 100644
--- a/src/amd/compiler/aco_register_allocation.cpp
+++ b/src/amd/compiler/aco_register_allocation.cpp
@@ -749,6 +749,7 @@ adjust_max_used_regs(ra_ctx& ctx, RegClass rc, unsigned reg)
    if (rc.type() == RegType::vgpr) {
       assert(reg >= 256);
       uint16_t hi = reg - 256 + size - 1;
+      assert(hi <= 255);
       ctx.max_used_vgpr = std::max(ctx.max_used_vgpr, hi);
    } else if (reg + rc.size() <= max_addressible_sgpr) {
       uint16_t hi = reg + size - 1;
@@ -1117,7 +1118,11 @@ get_regs_for_copies(ra_ctx& ctx, RegisterFile& reg_file,
             }
          }
       }
-      if (!res.second) {
+      if (!res.second && !def_reg.size) {
+         /* If this is before definitions are handled, def_reg may be an empty interval. */
+         info.bounds = bounds;
+         res = get_reg_simple(ctx, reg_file, info);
+      } else if (!res.second) {
          /* Try to find space within the bounds but outside of the definition */
          info.bounds = PhysRegInterval::from_until(bounds.lo(), MIN2(def_reg.lo(), bounds.hi()));
          res = get_reg_simple(ctx, reg_file, info);
diff --git a/src/amd/compiler/tests/test_regalloc.cpp b/src/amd/compiler/tests/test_regalloc.cpp
index 702dfe29541..da4f1e205ce 100644
--- a/src/amd/compiler/tests/test_regalloc.cpp
+++ b/src/amd/compiler/tests/test_regalloc.cpp
@@ -149,6 +149,18 @@ BEGIN_TEST(regalloc.precolor.vector.collect)
    finish_ra_test(ra_test_policy());
 END_TEST
 
+BEGIN_TEST(regalloc.precolor.vgpr_move)
+   //>> v1: %tmp0:v[0], v1: %tmp1:v[1] = p_startpgm
+   if (!setup_cs("v1 v1", GFX10))
+      return;
+
+   //! v1: %tmp0_2:v[1], v1: %tmp1_2:v[0] = p_parallelcopy %tmp0:v[0], %tmp1:v[1]
+   //! p_unit_test %tmp0_2:v[1], %tmp1_2:v[0]
+   bld.pseudo(aco_opcode::p_unit_test, inputs[0], Operand(inputs[1], PhysReg(256)));
+
+   finish_ra_test(ra_test_policy());
+END_TEST
+
 BEGIN_TEST(regalloc.scratch_sgpr.create_vector)
    if (!setup_cs("v1 s1", GFX7))
       return;