* and SSA id.
*/
struct Temp {
- Temp() = default;
+ Temp() noexcept : id_(0), reg_class(0) {}
constexpr Temp(uint32_t id, RegClass cls) noexcept
- : id_(id), reg_class(cls) {}
+ : id_(id), reg_class(uint8_t(cls)) {}
constexpr uint32_t id() const noexcept { return id_; }
- constexpr RegClass regClass() const noexcept { return reg_class; }
+ constexpr RegClass regClass() const noexcept { return (RegClass::RC)reg_class; }
- constexpr unsigned bytes() const noexcept { return reg_class.bytes(); }
- constexpr unsigned size() const noexcept { return reg_class.size(); }
- constexpr RegType type() const noexcept { return reg_class.type(); }
- constexpr bool is_linear() const noexcept { return reg_class.is_linear(); }
+ constexpr unsigned bytes() const noexcept { return regClass().bytes(); }
+ constexpr unsigned size() const noexcept { return regClass().size(); }
+ constexpr RegType type() const noexcept { return regClass().type(); }
+ constexpr bool is_linear() const noexcept { return regClass().is_linear(); }
constexpr bool operator <(Temp other) const noexcept { return id() < other.id(); }
constexpr bool operator==(Temp other) const noexcept { return id() == other.id(); }
constexpr bool operator!=(Temp other) const noexcept { return id() != other.id(); }
private:
- uint32_t id_:24;
- RegClass reg_class;
+ uint32_t id_: 24;
+ uint32_t reg_class : 8;
};
/**
/* hardware stages (can't be OR'd, just a mask for convenience when testing multiple) */
static constexpr Stage hw_vs = 1 << 7;
-static constexpr Stage hw_es = 1 << 8; /* not on GFX9. combined into GS on GFX9 (and GFX10/legacy). */
-static constexpr Stage hw_gs = 1 << 9;
-static constexpr Stage hw_ls = 1 << 10; /* not on GFX9. combined into HS on GFX9 (and GFX10/legacy). */
-static constexpr Stage hw_hs = 1 << 11;
-static constexpr Stage hw_fs = 1 << 12;
-static constexpr Stage hw_cs = 1 << 13;
-static constexpr Stage hw_mask = 0x7f << 7;
+static constexpr Stage hw_es = 1 << 8; /* Export shader: pre-GS (VS or TES) on GFX6-8. Combined into GS on GFX9 (and GFX10/legacy). */
+static constexpr Stage hw_gs = 1 << 9; /* Geometry shader on GFX10/legacy and GFX6-9. */
+static constexpr Stage hw_ngg_gs = 1 << 10; /* Geometry shader on GFX10/NGG. */
+static constexpr Stage hw_ls = 1 << 11; /* Local shader: pre-TCS (VS) on GFX6-8. Combined into HS on GFX9 (and GFX10/legacy). */
+static constexpr Stage hw_hs = 1 << 12; /* Hull shader: TCS on GFX6-8. Merged VS and TCS on GFX9-10. */
+static constexpr Stage hw_fs = 1 << 13;
+static constexpr Stage hw_cs = 1 << 14;
+static constexpr Stage hw_mask = 0xff << 7;
/* possible settings of Program::stage */
static constexpr Stage vertex_vs = sw_vs | hw_vs;
static constexpr Stage tess_eval_vs = sw_tes | hw_vs;
static constexpr Stage gs_copy_vs = sw_gs_copy | hw_vs;
/* GFX10/NGG */
-static constexpr Stage ngg_vertex_gs = sw_vs | hw_gs;
-static constexpr Stage ngg_vertex_geometry_gs = sw_vs | sw_gs | hw_gs;
-static constexpr Stage ngg_tess_eval_geometry_gs = sw_tes | sw_gs | hw_gs;
-static constexpr Stage ngg_vertex_tess_control_hs = sw_vs | sw_tcs | hw_hs;
+static constexpr Stage ngg_vertex_gs = sw_vs | hw_ngg_gs;
+static constexpr Stage ngg_vertex_geometry_gs = sw_vs | sw_gs | hw_ngg_gs;
+static constexpr Stage ngg_tess_eval_gs = sw_tes | hw_ngg_gs;
+static constexpr Stage ngg_tess_eval_geometry_gs = sw_tes | sw_gs | hw_ngg_gs;
/* GFX9 (and GFX10 if NGG isn't used) */
static constexpr Stage vertex_geometry_gs = sw_vs | sw_gs | hw_gs;
static constexpr Stage vertex_tess_control_hs = sw_vs | sw_tcs | hw_hs;