gallivm: don't lower local invocation index in frontend

The frontend can't handle variable block sizes properly,
so just lower it here in the backend.

Fixes CTS basic local_linear_id + get_linear_ids

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12285>

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_nir.c b/src/gallium/auxiliary/gallivm/lp_bld_nir.c
index 47466b6..5884f0a 100644 (file)
--- a/src/gallium/auxiliary/gallivm/lp_bld_nir.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_nir.c
@@ -1695,6 +1695,7 @@ static void visit_intrinsic(struct lp_build_nir_context *bld_base,
    case nir_intrinsic_load_first_vertex:
    case nir_intrinsic_load_workgroup_id:
    case nir_intrinsic_load_local_invocation_id:
+   case nir_intrinsic_load_local_invocation_index:
    case nir_intrinsic_load_num_workgroups:
    case nir_intrinsic_load_invocation_id:
    case nir_intrinsic_load_front_face:

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_nir_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_nir_soa.c
index 65c3422..eb4e0f4 100644 (file)
--- a/src/gallium/auxiliary/gallivm/lp_bld_nir_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_nir_soa.c
@@ -1528,6 +1528,20 @@ static void emit_sysval_intrin(struct lp_build_nir_context *bld_base,
       for (unsigned i = 0; i < 3; i++)
          result[i] = LLVMBuildExtractValue(gallivm->builder, bld->system_values.thread_id, i, "");
       break;
+   case nir_intrinsic_load_local_invocation_index: {
+      LLVMValueRef tmp, tmp2;
+      tmp = lp_build_broadcast_scalar(&bld_base->uint_bld, LLVMBuildExtractElement(gallivm->builder, bld->system_values.block_size, lp_build_const_int32(gallivm, 1), ""));
+      tmp2 = lp_build_broadcast_scalar(&bld_base->uint_bld, LLVMBuildExtractElement(gallivm->builder, bld->system_values.block_size, lp_build_const_int32(gallivm, 0), ""));
+      tmp = lp_build_mul(&bld_base->uint_bld, tmp, tmp2);
+      tmp = lp_build_mul(&bld_base->uint_bld, tmp, LLVMBuildExtractValue(gallivm->builder, bld->system_values.thread_id, 2, ""));
+
+      tmp2 = lp_build_broadcast_scalar(&bld_base->uint_bld, LLVMBuildExtractElement(gallivm->builder, bld->system_values.block_size, lp_build_const_int32(gallivm, 0), ""));
+      tmp2 = lp_build_mul(&bld_base->uint_bld, tmp2, LLVMBuildExtractValue(gallivm->builder, bld->system_values.thread_id, 1, ""));
+      tmp = lp_build_add(&bld_base->uint_bld, tmp, tmp2);
+      tmp = lp_build_add(&bld_base->uint_bld, tmp, LLVMBuildExtractValue(gallivm->builder, bld->system_values.thread_id, 0, ""));
+      result[0] = tmp;
+      break;
+   }
    case nir_intrinsic_load_num_workgroups: {
       LLVMValueRef tmp[3];
       for (unsigned i = 0; i < 3; i++) {

diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index 0d55bb0..4e43f5a 100644 (file)
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -595,7 +595,6 @@ static const struct nir_shader_compiler_options gallivm_nir_options = {
    .max_unroll_iterations = 32,
    .use_interpolated_input_intrinsics = true,
    .lower_to_scalar = true,
-   .lower_cs_local_index_from_id = true,
    .lower_uniforms_to_ubo = true,
    .lower_vector_cmp = true,
    .lower_device_index_to_zero = true,