r300g: turn fragment shader into a CB

5 files changed, 126 insertions, 129 deletions

diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index f501c1fc899..658880aebcc 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -127,7 +127,7 @@ static const float * get_rc_constant_state(
 /* Convert a normal single-precision float into the 7.16 format
  * used by the R300 fragment shader.
  */
-static uint32_t pack_float24(float f)
+uint32_t pack_float24(float f)
 {
     union {
         float fl;
@@ -158,77 +158,12 @@ static uint32_t pack_float24(float f)
     return float24;
 }
 
-unsigned r300_get_fs_atom_size(struct r300_context *r300)
-{
-    struct r300_fragment_shader *fs = r300_fs(r300);
-    unsigned imm_count = fs->shader->immediates_count;
-    struct r300_fragment_program_code *code = &fs->shader->code.code.r300;
-
-    return 19 +
-           code->alu.length * 4 +
-           (code->tex.length ? (1 + code->tex.length) : 0) +
-           (imm_count ? imm_count * 5 : 0);
-}
-
 void r300_emit_fs(struct r300_context* r300, unsigned size, void *state)
 {
     struct r300_fragment_shader *fs = r300_fs(r300);
-    struct rX00_fragment_program_code* generic_code = &fs->shader->code;
-    struct r300_fragment_program_code * code = &generic_code->code.r300;
-    unsigned i;
-    unsigned imm_count = fs->shader->immediates_count;
-    unsigned imm_first = fs->shader->externals_count;
-    unsigned imm_end = generic_code->constants.Count;
-    struct rc_constant *constants = generic_code->constants.Constants;
     CS_LOCALS(r300);
 
-    BEGIN_CS(size);
-    OUT_CS_REG(R300_US_CONFIG, code->config);
-    OUT_CS_REG(R300_US_PIXSIZE, code->pixsize);
-    OUT_CS_REG(R300_US_CODE_OFFSET, code->code_offset);
-
-    OUT_CS_REG_SEQ(R300_US_CODE_ADDR_0, 4);
-    OUT_CS_TABLE(code->code_addr, 4);
-
-    OUT_CS_REG_SEQ(R300_US_ALU_RGB_INST_0, code->alu.length);
-    for (i = 0; i < code->alu.length; i++)
-        OUT_CS(code->alu.inst[i].rgb_inst);
-
-    OUT_CS_REG_SEQ(R300_US_ALU_RGB_ADDR_0, code->alu.length);
-    for (i = 0; i < code->alu.length; i++)
-        OUT_CS(code->alu.inst[i].rgb_addr);
-
-    OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_INST_0, code->alu.length);
-    for (i = 0; i < code->alu.length; i++)
-        OUT_CS(code->alu.inst[i].alpha_inst);
-
-    OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, code->alu.length);
-    for (i = 0; i < code->alu.length; i++)
-        OUT_CS(code->alu.inst[i].alpha_addr);
-
-    if (code->tex.length) {
-        OUT_CS_REG_SEQ(R300_US_TEX_INST_0, code->tex.length);
-        OUT_CS_TABLE(code->tex.inst, code->tex.length);
-    }
-
-    /* Emit immediates. */
-    if (imm_count) {
-        for(i = imm_first; i < imm_end; ++i) {
-            if (constants[i].Type == RC_CONSTANT_IMMEDIATE) {
-                const float *data = constants[i].u.Immediate;
-
-                OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4);
-                OUT_CS(pack_float24(data[0]));
-                OUT_CS(pack_float24(data[1]));
-                OUT_CS(pack_float24(data[2]));
-                OUT_CS(pack_float24(data[3]));
-            }
-        }
-    }
-
-    OUT_CS_REG(R300_FG_DEPTH_SRC, fs->shader->fg_depth_src);
-    OUT_CS_REG(R300_US_W_FMT, fs->shader->us_out_w);
-    END_CS;
+    WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size);
 }
 
 void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *state)
@@ -285,67 +220,12 @@ void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, vo
     END_CS;
 }
 
-unsigned r500_get_fs_atom_size(struct r300_context *r300)
-{
-    struct r300_fragment_shader *fs = r300_fs(r300);
-    unsigned imm_count = fs->shader->immediates_count;
-    struct r500_fragment_program_code *code = &fs->shader->code.code.r500;
-
-    return 17 +
-           ((code->inst_end + 1) * 6) +
-           (imm_count ? imm_count * 7 : 0);
-}
-
 void r500_emit_fs(struct r300_context* r300, unsigned size, void *state)
 {
     struct r300_fragment_shader *fs = r300_fs(r300);
-    struct rX00_fragment_program_code* generic_code = &fs->shader->code;
-    struct r500_fragment_program_code * code = &generic_code->code.r500;
-    unsigned i;
-    unsigned imm_count = fs->shader->immediates_count;
-    unsigned imm_first = fs->shader->externals_count;
-    unsigned imm_end = generic_code->constants.Count;
-    struct rc_constant *constants = generic_code->constants.Constants;
     CS_LOCALS(r300);
 
-    BEGIN_CS(size);
-    OUT_CS_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
-    OUT_CS_REG(R500_US_PIXSIZE, code->max_temp_idx);
-    OUT_CS_REG(R500_US_CODE_RANGE,
-               R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end));
-    OUT_CS_REG(R500_US_CODE_OFFSET, 0);
-    OUT_CS_REG(R500_US_CODE_ADDR,
-               R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end));
-
-    OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR);
-    OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6);
-    for (i = 0; i <= code->inst_end; i++) {
-        OUT_CS(code->inst[i].inst0);
-        OUT_CS(code->inst[i].inst1);
-        OUT_CS(code->inst[i].inst2);
-        OUT_CS(code->inst[i].inst3);
-        OUT_CS(code->inst[i].inst4);
-        OUT_CS(code->inst[i].inst5);
-    }
-
-    /* Emit immediates. */
-    if (imm_count) {
-        for(i = imm_first; i < imm_end; ++i) {
-            if (constants[i].Type == RC_CONSTANT_IMMEDIATE) {
-                const float *data = constants[i].u.Immediate;
-
-                OUT_CS_REG(R500_GA_US_VECTOR_INDEX,
-                           R500_GA_US_VECTOR_INDEX_TYPE_CONST |
-                           (i & R500_GA_US_VECTOR_INDEX_MASK));
-                OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, 4);
-                OUT_CS_TABLE(data, 4);
-            }
-        }
-    }
-
-    OUT_CS_REG(R300_FG_DEPTH_SRC, fs->shader->fg_depth_src);
-    OUT_CS_REG(R300_US_W_FMT, fs->shader->us_out_w);
-    END_CS;
+    WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size);
 }
 
 void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *state)
diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h
index 90660883423..36a29894d01 100644
--- a/src/gallium/drivers/r300/r300_emit.h
+++ b/src/gallium/drivers/r300/r300_emit.h
@@ -29,6 +29,8 @@
 struct rX00_fragment_program_code;
 struct r300_vertex_program_code;
 
+uint32_t pack_float24(float f);
+
 void r300_emit_aos(struct r300_context* r300, int offset, boolean indexed);
 
 void r300_emit_blend_state(struct r300_context* r300,
@@ -43,16 +45,12 @@ void r300_emit_clip_state(struct r300_context* r300,
 void r300_emit_dsa_state(struct r300_context* r300,
                          unsigned size, void* state);
 
-unsigned r300_get_fs_atom_size(struct r300_context *r300);
-
 void r300_emit_fs(struct r300_context* r300, unsigned size, void *state);
 
 void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *state);
 
 void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state);
 
-unsigned r500_get_fs_atom_size(struct r300_context *r300);
-
 void r500_emit_fs(struct r300_context* r300, unsigned size, void *state);
 
 void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *state);
diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c
index a434808046c..e585394304e 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -28,7 +28,9 @@
 #include "tgsi/tgsi_dump.h"
 #include "tgsi/tgsi_ureg.h"
 
+#include "r300_cb.h"
 #include "r300_context.h"
+#include "r300_emit.h"
 #include "r300_screen.h"
 #include "r300_fs.h"
 #include "r300_reg.h"
@@ -229,6 +231,117 @@ static void r300_dummy_fragment_shader(
     ureg_destroy(ureg);
 }
 
+static void r300_emit_fs_code_to_buffer(
+    struct r300_context *r300,
+    struct r300_fragment_shader_code *shader)
+{
+    struct rX00_fragment_program_code *generic_code = &shader->code;
+    unsigned imm_count = shader->immediates_count;
+    unsigned imm_first = shader->externals_count;
+    unsigned imm_end = generic_code->constants.Count;
+    struct rc_constant *constants = generic_code->constants.Constants;
+    unsigned i;
+    CB_LOCALS;
+
+    if (r300->screen->caps.is_r500) {
+        struct r500_fragment_program_code *code = &generic_code->code.r500;
+
+        shader->cb_code_size = 17 +
+                               ((code->inst_end + 1) * 6) +
+                               imm_count * 7;
+
+        NEW_CB(shader->cb_code, shader->cb_code_size);
+        OUT_CB_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
+        OUT_CB_REG(R500_US_PIXSIZE, code->max_temp_idx);
+        OUT_CB_REG(R500_US_CODE_RANGE,
+                   R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end));
+        OUT_CB_REG(R500_US_CODE_OFFSET, 0);
+        OUT_CB_REG(R500_US_CODE_ADDR,
+                   R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end));
+
+        OUT_CB_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR);
+        OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6);
+        for (i = 0; i <= code->inst_end; i++) {
+            OUT_CB(code->inst[i].inst0);
+            OUT_CB(code->inst[i].inst1);
+            OUT_CB(code->inst[i].inst2);
+            OUT_CB(code->inst[i].inst3);
+            OUT_CB(code->inst[i].inst4);
+            OUT_CB(code->inst[i].inst5);
+        }
+
+        /* Emit immediates. */
+        if (imm_count) {
+            for(i = imm_first; i < imm_end; ++i) {
+                if (constants[i].Type == RC_CONSTANT_IMMEDIATE) {
+                    const float *data = constants[i].u.Immediate;
+
+                    OUT_CB_REG(R500_GA_US_VECTOR_INDEX,
+                               R500_GA_US_VECTOR_INDEX_TYPE_CONST |
+                               (i & R500_GA_US_VECTOR_INDEX_MASK));
+                    OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, 4);
+                    OUT_CB_TABLE(data, 4);
+                }
+            }
+        }
+    } else { /* r300 */
+        struct r300_fragment_program_code *code = &generic_code->code.r300;
+
+        shader->cb_code_size = 19 +
+                               code->alu.length * 4 +
+                               (code->tex.length ? (1 + code->tex.length) : 0) +
+                               imm_count * 5;
+
+        NEW_CB(shader->cb_code, shader->cb_code_size);
+        OUT_CB_REG(R300_US_CONFIG, code->config);
+        OUT_CB_REG(R300_US_PIXSIZE, code->pixsize);
+        OUT_CB_REG(R300_US_CODE_OFFSET, code->code_offset);
+
+        OUT_CB_REG_SEQ(R300_US_CODE_ADDR_0, 4);
+        OUT_CB_TABLE(code->code_addr, 4);
+
+        OUT_CB_REG_SEQ(R300_US_ALU_RGB_INST_0, code->alu.length);
+        for (i = 0; i < code->alu.length; i++)
+            OUT_CB(code->alu.inst[i].rgb_inst);
+
+        OUT_CB_REG_SEQ(R300_US_ALU_RGB_ADDR_0, code->alu.length);
+        for (i = 0; i < code->alu.length; i++)
+            OUT_CB(code->alu.inst[i].rgb_addr);
+
+        OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_INST_0, code->alu.length);
+        for (i = 0; i < code->alu.length; i++)
+            OUT_CB(code->alu.inst[i].alpha_inst);
+
+        OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, code->alu.length);
+        for (i = 0; i < code->alu.length; i++)
+            OUT_CB(code->alu.inst[i].alpha_addr);
+
+        if (code->tex.length) {
+            OUT_CB_REG_SEQ(R300_US_TEX_INST_0, code->tex.length);
+            OUT_CB_TABLE(code->tex.inst, code->tex.length);
+        }
+
+        /* Emit immediates. */
+        if (imm_count) {
+            for(i = imm_first; i < imm_end; ++i) {
+                if (constants[i].Type == RC_CONSTANT_IMMEDIATE) {
+                    const float *data = constants[i].u.Immediate;
+
+                    OUT_CB_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4);
+                    OUT_CB(pack_float24(data[0]));
+                    OUT_CB(pack_float24(data[1]));
+                    OUT_CB(pack_float24(data[2]));
+                    OUT_CB(pack_float24(data[3]));
+                }
+            }
+        }
+    }
+
+    OUT_CB_REG(R300_FG_DEPTH_SRC, shader->fg_depth_src);
+    OUT_CB_REG(R300_US_W_FMT, shader->us_out_w);
+    END_CB;
+}
+
 static void r300_translate_fragment_shader(
     struct r300_context* r300,
     struct r300_fragment_shader_code* shader,
@@ -338,6 +451,9 @@ static void r300_translate_fragment_shader(
 
     /* And, finally... */
     rc_destroy(&compiler.Base);
+
+    /* Build the command buffer. */
+    r300_emit_fs_code_to_buffer(r300, shader);
 }
 
 boolean r300_pick_fragment_shader(struct r300_context* r300)
diff --git a/src/gallium/drivers/r300/r300_fs.h b/src/gallium/drivers/r300/r300_fs.h
index 1cc43553595..51bfa88c5ef 100644
--- a/src/gallium/drivers/r300/r300_fs.h
+++ b/src/gallium/drivers/r300/r300_fs.h
@@ -50,6 +50,9 @@ struct r300_fragment_shader_code {
     struct r300_fragment_program_external_state compare_state;
     struct rX00_fragment_program_code code;
 
+    unsigned cb_code_size;
+    uint32_t *cb_code;
+
     struct r300_fragment_shader_code* next;
 };
 
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 8176bf4f98a..4892a1d7a34 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -759,13 +759,12 @@ void r300_mark_fs_code_dirty(struct r300_context *r300)
     r300->fs.dirty = TRUE;
     r300->fs_rc_constant_state.dirty = TRUE;
     r300->fs_constants.dirty = TRUE;
+    r300->fs.size = fs->shader->cb_code_size;
 
     if (r300->screen->caps.is_r500) {
-        r300->fs.size = r500_get_fs_atom_size(r300);
         r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7;
         r300->fs_constants.size = fs->shader->externals_count * 4 + 3;
     } else {
-        r300->fs.size = r300_get_fs_atom_size(r300);
         r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5;
         r300->fs_constants.size = fs->shader->externals_count * 4 + 1;
     }
@@ -799,6 +798,7 @@ static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
         tmp = ptr;
         ptr = ptr->next;
         rc_constants_destroy(&tmp->code.constants);
+        FREE(tmp->cb_code);
         FREE(tmp);
     }
     FREE((void*)fs->state.tokens);