aco: Fix subgroup_id intrinsic on GFX10.3+.
authorTimur Kristóf <timur.kristof@gmail.com>
Tue, 4 Jul 2023 12:28:45 +0000 (14:28 +0200)
committerMarge Bot <emma+marge@anholt.net>
Fri, 14 Jul 2023 12:17:24 +0000 (12:17 +0000)
Change this to match how it works in the LLVM backend.

Cc: mesa-stable
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24005>

src/amd/compiler/aco_instruction_selection.cpp

index 84a7ee0..b349f15 100644 (file)
@@ -8235,9 +8235,10 @@ visit_intrinsic(isel_context* ctx, nir_intrinsic_instr* instr)
    }
    case nir_intrinsic_load_subgroup_id: {
       if (ctx->stage.hw == AC_HW_COMPUTE_SHADER) {
+         const unsigned bfe_const =
+            ctx->program->gfx_level >= GFX10_3 ? (0x14u | 0x5u << 16) : (0x6u | (0x6u << 16));
          bld.sop2(aco_opcode::s_bfe_u32, Definition(get_ssa_temp(ctx, &instr->dest.ssa)),
-                  bld.def(s1, scc), get_arg(ctx, ctx->args->tg_size),
-                  Operand::c32(0x6u | (0x6u << 16)));
+                  bld.def(s1, scc), get_arg(ctx, ctx->args->tg_size), Operand::c32(bfe_const));
       } else if (ctx->stage.hw == AC_HW_NEXT_GEN_GEOMETRY_SHADER) {
          /* Get the id of the current wave within the threadgroup (workgroup) */
          bld.sop2(aco_opcode::s_bfe_u32, Definition(get_ssa_temp(ctx, &instr->dest.ssa)),