Mesa (master): spirv: Implement SPV_KHR_shader_clock

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: f20cea016214ed27d50a6974fa108d360fda1da8
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=f20cea016214ed27d50a6974fa108d360fda1da8

Author: Caio Marcelo de Oliveira Filho <caio.oliveira at intel.com>
Date:   Fri Apr 19 12:15:16 2019 -0700

spirv: Implement SPV_KHR_shader_clock

We only have the subgroup variant in NIR (equivalent to clockARB), so
only support that for now.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin at intel.com>

---

 src/compiler/shader_info.h        |  1 +
 src/compiler/spirv/spirv_to_nir.c | 35 +++++++++++++++++++++++++++++++++++
 2 files changed, 36 insertions(+)

diff --git a/src/compiler/shader_info.h b/src/compiler/shader_info.h
index 8b0176f0438..740b75b756a 100644
--- a/src/compiler/shader_info.h
+++ b/src/compiler/shader_info.h
@@ -60,6 +60,7 @@ struct spirv_supported_capabilities {
    bool post_depth_coverage;
    bool runtime_descriptor_array;
    bool float_controls;
+   bool shader_clock;
    bool shader_viewport_index_layer;
    bool stencil_export;
    bool storage_8bit;
diff --git a/src/compiler/spirv/spirv_to_nir.c b/src/compiler/spirv/spirv_to_nir.c
index edffcb0a53f..abfd8eaef70 100644
--- a/src/compiler/spirv/spirv_to_nir.c
+++ b/src/compiler/spirv/spirv_to_nir.c
@@ -3652,6 +3652,10 @@ vtn_handle_preamble_instruction(struct vtn_builder *b, SpvOp opcode,
          spv_check_supported(demote_to_helper_invocation, cap);
          break;
 
+      case SpvCapabilityShaderClockKHR:
+         spv_check_supported(shader_clock, cap);
+         break;
+
       default:
          vtn_fail("Unhandled capability: %s (%u)",
                   spirv_capability_to_string(cap), cap);
@@ -4554,6 +4558,37 @@ vtn_handle_body_instruction(struct vtn_builder *b, SpvOp opcode,
       break;
    }
 
+   case SpvOpReadClockKHR: {
+      assert(vtn_constant_uint(b, w[3]) == SpvScopeSubgroup);
+
+      /* Operation supports two result types: uvec2 and uint64_t.  The NIR
+       * intrinsic gives uvec2, so pack the result for the other case.
+       */
+      nir_intrinsic_instr *intrin =
+         nir_intrinsic_instr_create(b->nb.shader, nir_intrinsic_shader_clock);
+      nir_ssa_dest_init(&intrin->instr, &intrin->dest, 2, 32, NULL);
+      nir_builder_instr_insert(&b->nb, &intrin->instr);
+
+      struct vtn_type *type = vtn_value(b, w[1], vtn_value_type_type)->type;
+      const struct glsl_type *dest_type = type->type;
+      nir_ssa_def *result;
+
+      if (glsl_type_is_vector(dest_type)) {
+         assert(dest_type == glsl_vector_type(GLSL_TYPE_UINT, 2));
+         result = &intrin->dest.ssa;
+      } else {
+         assert(glsl_type_is_scalar(dest_type));
+         assert(glsl_get_base_type(dest_type) == GLSL_TYPE_UINT64);
+         result = nir_pack_64_2x32(&b->nb, &intrin->dest.ssa);
+      }
+
+      struct vtn_value *val = vtn_push_value(b, w[2], vtn_value_type_ssa);
+      val->type = type;
+      val->ssa = vtn_create_ssa_value(b, dest_type);
+      val->ssa->def = result;
+      break;
+   }
+
    default:
       vtn_fail_with_opcode("Unhandled opcode", opcode);
    }