ir3: Moves with shared destination are always legal

I got this wrong before because I missed the need for (ss), once that
was fixed then a move from anything to a shared register is legal,
include non-shared registers, as long as all active channels have the
same value.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22075>

6 files changed, 5 insertions, 39 deletions

diff --git a/src/freedreno/ir3/disasm-a3xx.c b/src/freedreno/ir3/disasm-a3xx.c
index 0d64bcab2c2..f72583fff07 100644
--- a/src/freedreno/ir3/disasm-a3xx.c
+++ b/src/freedreno/ir3/disasm-a3xx.c
@@ -196,7 +196,6 @@ static const struct opc_info {
    OPC(1, OPC_ELECT_MACRO,  elect.macro),
    OPC(1, OPC_READ_COND_MACRO, read_cond.macro),
    OPC(1, OPC_READ_FIRST_MACRO, read_first.macro),
-   OPC(1, OPC_SWZ_SHARED_MACRO, swz_shared.macro),
    OPC(1, OPC_SCAN_MACRO, scan.macro),
    OPC(1, OPC_SCAN_CLUSTERS_MACRO, scan_clusters.macro),
    OPC(1, OPC_SHPS_MACRO, shps.macro),
diff --git a/src/freedreno/ir3/instr-a3xx.h b/src/freedreno/ir3/instr-a3xx.h
index c928bd42acd..b56292efcbd 100644
--- a/src/freedreno/ir3/instr-a3xx.h
+++ b/src/freedreno/ir3/instr-a3xx.h
@@ -124,8 +124,7 @@ typedef enum {
    OPC_ELECT_MACRO     = _OPC(1, 53),
    OPC_READ_COND_MACRO = _OPC(1, 54),
    OPC_READ_FIRST_MACRO = _OPC(1, 55),
-   OPC_SWZ_SHARED_MACRO = _OPC(1, 56),
-   OPC_SHPS_MACRO       = _OPC(1, 57),
+   OPC_SHPS_MACRO       = _OPC(1, 56),
 
    /* Macros that expand to a loop */
    OPC_SCAN_MACRO      = _OPC(1, 58),
diff --git a/src/freedreno/ir3/ir3.h b/src/freedreno/ir3/ir3.h
index 2caad8ea8e2..2c510382f3f 100644
--- a/src/freedreno/ir3/ir3.h
+++ b/src/freedreno/ir3/ir3.h
@@ -998,7 +998,6 @@ is_subgroup_cond_mov_macro(struct ir3_instruction *instr)
    case OPC_ELECT_MACRO:
    case OPC_READ_COND_MACRO:
    case OPC_READ_FIRST_MACRO:
-   case OPC_SWZ_SHARED_MACRO:
    case OPC_SCAN_MACRO:
    case OPC_SCAN_CLUSTERS_MACRO:
       return true;
diff --git a/src/freedreno/ir3/ir3_lower_parallelcopy.c b/src/freedreno/ir3/ir3_lower_parallelcopy.c
index 7d8b52a5d64..5962142120c 100644
--- a/src/freedreno/ir3/ir3_lower_parallelcopy.c
+++ b/src/freedreno/ir3/ir3_lower_parallelcopy.c
@@ -162,14 +162,7 @@ do_swap(struct ir3_compiler *compiler, struct ir3_instruction *instr,
       do_xor(instr, src_num, src_num, dst_num, entry->flags);
       do_xor(instr, dst_num, dst_num, src_num, entry->flags);
    } else {
-      /* Use a macro for shared regs because any shared reg writes need to
-       * be wrapped in a getone block to work correctly. Writing shared regs
-       * with multiple threads active does not work, even if they all return
-       * the same value.
-       */
-      unsigned opc =
-         (entry->flags & IR3_REG_SHARED) ? OPC_SWZ_SHARED_MACRO : OPC_SWZ;
-      struct ir3_instruction *swz = ir3_instr_create(instr->block, opc, 2, 2);
+      struct ir3_instruction *swz = ir3_instr_create(instr->block, OPC_SWZ, 2, 2);
       ir3_dst_create(swz, dst_num, entry->flags);
       ir3_dst_create(swz, src_num, entry->flags);
       ir3_src_create(swz, src_num, entry->flags);
@@ -253,10 +246,7 @@ do_copy(struct ir3_compiler *compiler, struct ir3_instruction *instr,
    unsigned src_num = ra_physreg_to_num(entry->src.reg, entry->flags);
    unsigned dst_num = ra_physreg_to_num(entry->dst, entry->flags);
 
-   /* Similar to the swap case, we have to use a macro for shared regs. */
-   unsigned opc =
-      (entry->flags & IR3_REG_SHARED) ? OPC_READ_FIRST_MACRO : OPC_MOV;
-   struct ir3_instruction *mov = ir3_instr_create(instr->block, opc, 1, 1);
+   struct ir3_instruction *mov = ir3_instr_create(instr->block, OPC_MOV, 1, 1);
    ir3_dst_create(mov, dst_num, entry->flags);
    if (entry->src.flags & (IR3_REG_IMMED | IR3_REG_CONST))
       ir3_src_create(mov, INVALID_REG, (entry->flags & IR3_REG_HALF) | entry->src.flags);
diff --git a/src/freedreno/ir3/ir3_lower_subgroups.c b/src/freedreno/ir3/ir3_lower_subgroups.c
index d949ed536a2..bd4048d63f4 100644
--- a/src/freedreno/ir3/ir3_lower_subgroups.c
+++ b/src/freedreno/ir3/ir3_lower_subgroups.c
@@ -263,7 +263,6 @@ lower_instr(struct ir3 *ir, struct ir3_block **block, struct ir3_instruction *in
    case OPC_ALL_MACRO:
    case OPC_ELECT_MACRO:
    case OPC_READ_COND_MACRO:
-   case OPC_SWZ_SHARED_MACRO:
    case OPC_SCAN_MACRO:
    case OPC_SCAN_CLUSTERS_MACRO:
       break;
@@ -411,16 +410,10 @@ lower_instr(struct ir3 *ir, struct ir3_block **block, struct ir3_instruction *in
    } else {
       /* For ballot, the destination must be initialized to 0 before we do
        * the movmsk because the condition may be 0 and then the movmsk will
-       * be skipped. Because it's a shared register we have to wrap the
-       * initialization in a getone block.
+       * be skipped.
        */
       if (instr->opc == OPC_BALLOT_MACRO) {
-         struct ir3_block *then_block =
-            create_if(ir, before_block, after_block, OPC_GETONE, NULL);
-         mov_immed(instr->dsts[0], then_block, 0);
-         after_block->reconvergence_point = true;
-         before_block = after_block;
-         after_block = split_block(ir, before_block, instr);
+         mov_immed(instr->dsts[0], before_block, 0);
       }
 
       struct ir3_instruction *condition = NULL;
@@ -450,7 +443,6 @@ lower_instr(struct ir3 *ir, struct ir3_block **block, struct ir3_instruction *in
          branch_opc = OPC_BALL;
          break;
       case OPC_ELECT_MACRO:
-      case OPC_SWZ_SHARED_MACRO:
          after_block->reconvergence_point = true;
          branch_opc = OPC_GETONE;
          break;
@@ -490,18 +482,6 @@ lower_instr(struct ir3 *ir, struct ir3_block **block, struct ir3_instruction *in
          break;
       }
 
-      case OPC_SWZ_SHARED_MACRO: {
-         struct ir3_instruction *swz =
-            ir3_instr_create(then_block, OPC_SWZ, 2, 2);
-         ir3_dst_create(swz, instr->dsts[0]->num, instr->dsts[0]->flags);
-         ir3_dst_create(swz, instr->dsts[1]->num, instr->dsts[1]->flags);
-         ir3_src_create(swz, instr->srcs[0]->num, instr->srcs[0]->flags);
-         ir3_src_create(swz, instr->srcs[1]->num, instr->srcs[1]->flags);
-         swz->cat1.dst_type = swz->cat1.src_type = TYPE_U32;
-         swz->repeat = 1;
-         break;
-      }
-
       default:
          unreachable("bad opcode");
       }
diff --git a/src/freedreno/ir3/ir3_reconvergence.c b/src/freedreno/ir3/ir3_reconvergence.c
index bddd733efca..ddf1dfeeb0e 100644
--- a/src/freedreno/ir3/ir3_reconvergence.c
+++ b/src/freedreno/ir3/ir3_reconvergence.c
@@ -303,7 +303,6 @@ ir3_calc_reconvergence(struct ir3_shader_variant *so)
          case OPC_READ_COND_MACRO:
          case OPC_ELECT_MACRO:
          case OPC_READ_FIRST_MACRO:
-         case OPC_SWZ_SHARED_MACRO:
             so->branchstack = MAX2(so->branchstack, rc_level + 1);
             break;
          default: