X-Git-Url: https://git.libre-soc.org/?p=mesa.git;a=blobdiff_plain;f=src%2Fbroadcom%2Fcompiler%2Fv3d_compiler.h;h=29057bdf4df03247af574c21c399f56c2f2fd0d8;hp=5bc514ce0a0d7b9ba582c0fc523b0678cc8c6273;hb=6e68f743953d79f2beb2ac1fc34331f016d1b7c7;hpb=110f14d4b4bdee779a35c26e3224a9d28eb81fa7

diff --git a/src/broadcom/compiler/v3d_compiler.h b/src/broadcom/compiler/v3d_compiler.h
index 5bc514ce0a0..29057bdf4df 100644
--- a/src/broadcom/compiler/v3d_compiler.h
+++ b/src/broadcom/compiler/v3d_compiler.h
@@ -69,8 +69,6 @@ enum qfile {
          * or physical registers later.
          */
         QFILE_TEMP,
-        QFILE_TLB,
-        QFILE_TLBU,
 
         /**
          * VPM reads use this with an index value to say what part of the VPM
@@ -104,6 +102,11 @@ static inline struct qreg vir_reg(enum qfile file, uint32_t index)
         return (struct qreg){file, index};
 }
 
+static inline struct qreg vir_magic_reg(uint32_t index)
+{
+        return (struct qreg){QFILE_MAGIC, index};
+}
+
 static inline struct qreg vir_nop_reg(void)
 {
         return (struct qreg){QFILE_NULL, 0};
@@ -278,17 +281,18 @@ enum quniform_contents {
         QUNIFORM_SHARED_OFFSET,
 };
 
-static inline uint32_t v3d_tmu_config_data_create(uint32_t unit, uint32_t value)
+static inline uint32_t v3d_unit_data_create(uint32_t unit, uint32_t value)
 {
+        assert(value < (1 << 24));
         return unit << 24 | value;
 }
 
-static inline uint32_t v3d_tmu_config_data_get_unit(uint32_t data)
+static inline uint32_t v3d_unit_data_get_unit(uint32_t data)
 {
         return data >> 24;
 }
 
-static inline uint32_t v3d_tmu_config_data_get_value(uint32_t data)
+static inline uint32_t v3d_unit_data_get_offset(uint32_t data)
 {
         return data & 0xffffff;
 }
@@ -314,25 +318,6 @@ static inline uint8_t v3d_slot_get_component(struct v3d_varying_slot slot)
         return slot.slot_and_component & 3;
 }
 
-struct v3d_ubo_range {
-        /**
-         * offset in bytes from the start of the ubo where this range is
-         * uploaded.
-         *
-         * Only set once used is set.
-         */
-        uint32_t dst_offset;
-
-        /**
-         * offset in bytes from the start of the gallium uniforms where the
-         * data comes from.
-         */
-        uint32_t src_offset;
-
-        /** size in bytes of this ubo range */
-        uint32_t size;
-};
-
 struct v3d_key {
         void *shader_state;
         struct {
@@ -370,6 +355,15 @@ struct v3d_fs_key {
          */
         uint8_t int_color_rb;
         uint8_t uint_color_rb;
+
+        /* Color format information per render target. Only set when logic
+         * operations are enabled.
+         */
+        struct {
+                enum pipe_format format;
+                const uint8_t *swizzle;
+        } color_fmt[V3D_MAX_DRAW_BUFFERS];
+
         uint8_t alpha_test_func;
         uint8_t logicop_func;
         uint32_t point_sprite_mask;
@@ -380,8 +374,8 @@ struct v3d_fs_key {
 struct v3d_vs_key {
         struct v3d_key base;
 
-        struct v3d_varying_slot fs_inputs[V3D_MAX_FS_INPUTS];
-        uint8_t num_fs_inputs;
+        struct v3d_varying_slot used_outputs[V3D_MAX_ANY_STAGE_INPUTS];
+        uint8_t num_used_outputs;
 
         bool is_coord;
         bool per_vertex_point_size;
@@ -510,8 +504,8 @@ struct v3d_compile {
         struct qreg *inputs;
         struct qreg *outputs;
         bool msaa_per_sample_output;
-        struct qreg color_reads[V3D_MAX_SAMPLES];
-        struct qreg sample_colors[V3D_MAX_SAMPLES];
+        struct qreg color_reads[V3D_MAX_DRAW_BUFFERS * V3D_MAX_SAMPLES * 4];
+        struct qreg sample_colors[V3D_MAX_DRAW_BUFFERS * V3D_MAX_SAMPLES * 4];
         uint32_t inputs_array_size;
         uint32_t outputs_array_size;
         uint32_t uniforms_array_size;
@@ -528,13 +522,7 @@ struct v3d_compile {
 
         bool uses_center_w;
         bool writes_z;
-
-        struct v3d_ubo_range *ubo_ranges;
-        bool *ubo_range_used;
-        uint32_t ubo_ranges_array_size;
-        /** Number of uniform areas tracked in ubo_ranges. */
-        uint32_t num_ubo_ranges;
-        uint32_t next_ubo_dst_offset;
+        bool uses_implicit_point_line_varyings;
 
         /* State for whether we're executing on each channel currently.  0 if
          * yes, otherwise a block number + 1 that the channel jumped to.
@@ -625,24 +613,13 @@ struct v3d_compile {
         uint64_t *qpu_insts;
         uint32_t qpu_inst_count;
         uint32_t qpu_inst_size;
+        uint32_t qpu_inst_stalled_count;
 
         /* For the FS, the number of varying inputs not counting the
          * point/line varyings payload
          */
         uint32_t num_inputs;
 
-        /**
-         * Number of inputs from num_inputs remaining to be queued to the read
-         * FIFO in the VS/CS.
-         */
-        uint32_t num_inputs_remaining;
-
-        /* Number of inputs currently in the read FIFO for the VS/CS */
-        uint32_t num_inputs_in_fifo;
-
-        /** Next offset in the VPM to read from in the VS/CS */
-        uint32_t vpm_read_offset;
-
         uint32_t program_id;
         uint32_t variant_id;
 
@@ -658,7 +635,12 @@ struct v3d_compile {
         struct qinst *last_thrsw;
         bool last_thrsw_at_top_level;
 
+        bool emitted_tlb_load;
+        bool lock_scoreboard_on_first_thrsw;
+
         bool failed;
+
+        bool tmu_dirty_rcl;
 };
 
 struct v3d_uniform_list {
@@ -670,9 +652,6 @@ struct v3d_uniform_list {
 struct v3d_prog_data {
         struct v3d_uniform_list uniforms;
 
-        struct v3d_ubo_range *ubo_ranges;
-        uint32_t num_ubo_ranges;
-        uint32_t ubo_size;
         uint32_t spill_size;
 
         uint8_t threads;
@@ -681,6 +660,8 @@ struct v3d_prog_data {
          * after-final-THRSW state.
          */
         bool single_seg;
+
+        bool tmu_dirty_rcl;
 };
 
 struct v3d_vs_prog_data {
@@ -726,6 +707,14 @@ struct v3d_fs_prog_data {
         bool writes_z;
         bool disable_ez;
         bool uses_center_w;
+        bool uses_implicit_point_line_varyings;
+        bool lock_scoreboard_on_first_thrsw;
+};
+
+struct v3d_compute_prog_data {
+        struct v3d_prog_data base;
+        /* Size in bytes of the workgroup's shared space. */
+        uint32_t shared_size;
 };
 
 static inline bool
@@ -734,13 +723,6 @@ vir_has_uniform(struct qinst *inst)
         return inst->uniform != ~0;
 }
 
-/* Special nir_load_input intrinsic index for loading the current TLB
- * destination color.
- */
-#define V3D_NIR_TLB_COLOR_READ_INPUT		2000000000
-
-#define V3D_NIR_MS_MASK_OUTPUT			2000000000
-
 extern const nir_shader_compiler_options v3d_nir_options;
 
 const struct v3d_compiler *v3d_compiler_init(const struct v3d_device_info *devinfo);
@@ -780,6 +762,7 @@ struct qreg vir_uniform(struct v3d_compile *c,
                         enum quniform_contents contents,
                         uint32_t data);
 void vir_schedule_instructions(struct v3d_compile *c);
+void v3d_setup_spill_base(struct v3d_compile *c);
 struct v3d_qpu_instr v3d_qpu_nop(void);
 
 struct qreg vir_emit_def(struct v3d_compile *c, struct qinst *inst);
@@ -822,10 +805,13 @@ bool vir_opt_constant_folding(struct v3d_compile *c);
 bool vir_opt_copy_propagate(struct v3d_compile *c);
 bool vir_opt_dead_code(struct v3d_compile *c);
 bool vir_opt_peephole_sf(struct v3d_compile *c);
+bool vir_opt_redundant_flags(struct v3d_compile *c);
 bool vir_opt_small_immediates(struct v3d_compile *c);
 bool vir_opt_vpm(struct v3d_compile *c);
 void v3d_nir_lower_blend(nir_shader *s, struct v3d_compile *c);
 void v3d_nir_lower_io(nir_shader *s, struct v3d_compile *c);
+void v3d_nir_lower_logic_ops(nir_shader *s, struct v3d_compile *c);
+void v3d_nir_lower_scratch(nir_shader *s);
 void v3d_nir_lower_txf_ms(nir_shader *s, struct v3d_compile *c);
 void v3d_nir_lower_image_load_store(nir_shader *s);
 void vir_lower_uniforms(struct v3d_compile *c);
@@ -845,6 +831,9 @@ bool vir_init_reg_sets(struct v3d_compiler *compiler);
 
 bool v3d_gl_format_is_return_32(GLenum format);
 
+uint32_t
+v3d_get_op_for_atomic_add(nir_intrinsic_instr *instr, unsigned src);
+
 static inline bool
 quniform_contents_is_texture_p0(enum quniform_contents contents)
 {
@@ -1009,6 +998,10 @@ VIR_A_ALU0(TIDX)
 VIR_A_ALU0(EIDX)
 VIR_A_ALU1(LDVPMV_IN)
 VIR_A_ALU1(LDVPMV_OUT)
+VIR_A_ALU1(LDVPMD_IN)
+VIR_A_ALU1(LDVPMD_OUT)
+VIR_A_ALU2(LDVPMG_IN)
+VIR_A_ALU2(LDVPMG_OUT)
 VIR_A_ALU0(TMUWT)
 
 VIR_A_ALU0(FXCD)
@@ -1102,6 +1095,30 @@ vir_UMUL(struct v3d_compile *c, struct qreg src0, struct qreg src1)
         return vir_UMUL24(c, src0, src1);
 }
 
+static inline struct qreg
+vir_TLBU_COLOR_READ(struct v3d_compile *c, uint32_t config)
+{
+        assert(c->devinfo->ver >= 41); /* XXX */
+        assert((config & 0xffffff00) == 0xffffff00);
+
+        struct qinst *ldtlb = vir_add_inst(V3D_QPU_A_NOP, c->undef,
+                                           c->undef, c->undef);
+        ldtlb->qpu.sig.ldtlbu = true;
+        ldtlb->uniform = vir_get_uniform_index(c, QUNIFORM_CONSTANT, config);
+        return vir_emit_def(c, ldtlb);
+}
+
+static inline struct qreg
+vir_TLB_COLOR_READ(struct v3d_compile *c)
+{
+        assert(c->devinfo->ver >= 41); /* XXX */
+
+        struct qinst *ldtlb = vir_add_inst(V3D_QPU_A_NOP, c->undef,
+                                           c->undef, c->undef);
+        ldtlb->qpu.sig.ldtlb = true;
+        return vir_emit_def(c, ldtlb);
+}
+
 /*
 static inline struct qreg
 vir_LOAD_IMM(struct v3d_compile *c, uint32_t val)