i965/vec4: Optimize packUnorm4x8().

Reduces the number of instructions needed to implement packUnorm4x8()
from 11 -> 6.

3 files changed, 27 insertions, 4 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
index 61ea2079140..d7a2a916387 100644
--- a/src/mesa/drivers/dri/i965/brw_shader.cpp
+++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
@@ -83,12 +83,12 @@ brw_lower_packing_builtins(struct brw_context *brw,
            | LOWER_UNPACK_SNORM_2x16
            | LOWER_PACK_UNORM_2x16
            | LOWER_UNPACK_UNORM_2x16
-           | LOWER_PACK_SNORM_4x8
-           | LOWER_PACK_UNORM_4x8;
+           | LOWER_PACK_SNORM_4x8;
 
    if (shader_type == MESA_SHADER_FRAGMENT) {
       ops |= LOWER_UNPACK_UNORM_4x8
-           | LOWER_UNPACK_SNORM_4x8;
+           | LOWER_UNPACK_SNORM_4x8
+           | LOWER_PACK_UNORM_4x8;
    }
 
    if (brw->gen >= 7) {
diff --git a/src/mesa/drivers/dri/i965/brw_vec4.h b/src/mesa/drivers/dri/i965/brw_vec4.h
index 5b3ef8a2822..150e20a0387 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4.h
+++ b/src/mesa/drivers/dri/i965/brw_vec4.h
@@ -510,6 +510,7 @@ public:
    void emit_unpack_half_2x16(dst_reg dst, src_reg src0);
    void emit_unpack_unorm_4x8(const dst_reg &dst, src_reg src0);
    void emit_unpack_snorm_4x8(const dst_reg &dst, src_reg src0);
+   void emit_pack_unorm_4x8(const dst_reg &dst, const src_reg &src0);
 
    uint32_t gather_channel(ir_texture *ir, uint32_t sampler);
    src_reg emit_mcs_fetch(ir_texture *ir, src_reg coordinate, src_reg sampler);
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
index b6ace86cfdb..34f96070250 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
@@ -516,6 +516,26 @@ vec4_visitor::emit_unpack_snorm_4x8(const dst_reg &dst, src_reg src0)
 }
 
 void
+vec4_visitor::emit_pack_unorm_4x8(const dst_reg &dst, const src_reg &src0)
+{
+   dst_reg saturated(this, glsl_type::vec4_type);
+   vec4_instruction *inst = emit(MOV(saturated, src0));
+   inst->saturate = true;
+
+   dst_reg scaled(this, glsl_type::vec4_type);
+   emit(MUL(scaled, src_reg(saturated), src_reg(255.0f)));
+
+   dst_reg rounded(this, glsl_type::vec4_type);
+   emit(RNDE(rounded, src_reg(scaled)));
+
+   dst_reg u(this, glsl_type::uvec4_type);
+   emit(MOV(u, src_reg(rounded)));
+
+   src_reg bytes(u);
+   emit(VEC4_OPCODE_PACK_BYTES, dst, bytes);
+}
+
+void
 vec4_visitor::visit_instructions(const exec_list *list)
 {
    foreach_in_list(ir_instruction, ir, list) {
@@ -1802,10 +1822,12 @@ vec4_visitor::visit(ir_expression *ir)
    case ir_unop_unpack_snorm_4x8:
       emit_unpack_snorm_4x8(result_dst, op[0]);
       break;
+   case ir_unop_pack_unorm_4x8:
+      emit_pack_unorm_4x8(result_dst, op[0]);
+      break;
    case ir_unop_pack_snorm_2x16:
    case ir_unop_pack_snorm_4x8:
    case ir_unop_pack_unorm_2x16:
-   case ir_unop_pack_unorm_4x8:
    case ir_unop_unpack_snorm_2x16:
    case ir_unop_unpack_unorm_2x16:
       unreachable("not reached: should be handled by lower_packing_builtins");