r600g: set correct pitch/offset for depth textures in flushed state.

[mesa.git] / src / gallium / drivers / r600 / r600_asm.c

diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index b9c74e93299a0aaf92cba75aa50fe7eafe02b0c0..e910d1cc73fb2a08444b73d4886293972d7cc3a7 100644 (file)
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -246,6 +246,37 @@ int r600_bc_add_output(struct r600_bc *bc, const struct r600_bc_output *output)
 {
        int r;
 
+       if (bc->cf_last && (bc->cf_last->inst == output->inst ||
+               (bc->cf_last->inst == BC_INST(bc, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT) &&
+               output->inst == BC_INST(bc, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE))) &&
+               output->type == bc->cf_last->output.type &&
+               output->elem_size == bc->cf_last->output.elem_size &&
+               output->swizzle_x == bc->cf_last->output.swizzle_x &&
+               output->swizzle_y == bc->cf_last->output.swizzle_y &&
+               output->swizzle_z == bc->cf_last->output.swizzle_z &&
+               output->swizzle_w == bc->cf_last->output.swizzle_w &&
+               (output->burst_count + bc->cf_last->output.burst_count) <= 16) {
+
+               if ((output->gpr + output->burst_count) == bc->cf_last->output.gpr &&
+                       (output->array_base + output->burst_count) == bc->cf_last->output.array_base) {
+
+                       bc->cf_last->output.end_of_program |= output->end_of_program;
+                       bc->cf_last->output.inst = output->inst;
+                       bc->cf_last->output.gpr = output->gpr;
+                       bc->cf_last->output.array_base = output->array_base;
+                       bc->cf_last->output.burst_count += output->burst_count;
+                       return 0;
+
+               } else if (output->gpr == (bc->cf_last->output.gpr + bc->cf_last->output.burst_count) &&
+                       output->array_base == (bc->cf_last->output.array_base + bc->cf_last->output.burst_count)) {
+
+                       bc->cf_last->output.end_of_program |= output->end_of_program;
+                       bc->cf_last->output.inst = output->inst;
+                       bc->cf_last->output.burst_count += output->burst_count;
+                       return 0;
+               }
+       }
+
        r = r600_bc_add_cf(bc);
        if (r)
                return r;
@@ -355,6 +386,20 @@ static int is_alu_reduction_inst(struct r600_bc *bc, struct r600_bc_alu *alu)
        }
 }
 
+static int is_alu_cube_inst(struct r600_bc *bc, struct r600_bc_alu *alu)
+{
+       switch (bc->chiprev) {
+       case CHIPREV_R600:
+       case CHIPREV_R700:
+               return !alu->is_op3 &&
+                       alu->inst == V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CUBE;
+       case CHIPREV_EVERGREEN:
+       default:
+               return !alu->is_op3 &&
+                       alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CUBE;
+       }
+}
+
 static int is_alu_mova_inst(struct r600_bc *bc, struct r600_bc_alu *alu)
 {
        switch (bc->chiprev) {
@@ -417,8 +462,9 @@ static int is_alu_trans_unit_inst(struct r600_bc *bc, struct r600_bc_alu *alu)
        case CHIPREV_EVERGREEN:
        default:
                if (!alu->is_op3)
+                       /* Note that FLT_TO_INT* instructions are vector instructions
+                        * on Evergreen, despite what the documentation says. */
                        return alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ASHR_INT ||
-                               alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT ||
                                alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT ||
                                alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LSHL_INT ||
                                alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LSHR_INT ||
@@ -721,7 +767,8 @@ static int replace_gpr_with_pv_ps(struct r600_bc *bc,
        for (i = 0; i < 5; ++i) {
                if(prev[i] && prev[i]->dst.write && !prev[i]->dst.rel) {
                        gpr[i] = prev[i]->dst.sel;
-                       if (is_alu_reduction_inst(bc, prev[i]))
+                       /* cube writes more than PV.X */
+                       if (!is_alu_cube_inst(bc, prev[i]) && is_alu_reduction_inst(bc, prev[i]))
                                chan[i] = 0;
                        else
                                chan[i] = prev[i]->dst.chan;
@@ -1427,7 +1474,8 @@ static int r600_bc_cf_build(struct r600_bc *bc, struct r600_bc_cf *cf)
                        S_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(cf->output.elem_size) |
                        S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(cf->output.array_base) |
                        S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(cf->output.type);
-               bc->bytecode[id++] = S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(cf->output.swizzle_x) |
+               bc->bytecode[id++] = S_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(cf->output.burst_count - 1) |
+                       S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(cf->output.swizzle_x) |
                        S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Y(cf->output.swizzle_y) |
                        S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Z(cf->output.swizzle_z) |
                        S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_W(cf->output.swizzle_w) |
@@ -1709,9 +1757,9 @@ void r600_bc_dump(struct r600_bc *bc)
                        fprintf(stderr, "SWIZ_Y:%X ", cf->output.swizzle_y);
                        fprintf(stderr, "SWIZ_Z:%X ", cf->output.swizzle_z);
                        fprintf(stderr, "SWIZ_W:%X ", cf->output.swizzle_w);
-                       fprintf(stderr, "SWIZ_W:%X ", cf->output.swizzle_w);
                        fprintf(stderr, "BARRIER:%X ", cf->output.barrier);
                        fprintf(stderr, "INST:%d ", cf->output.inst);
+                       fprintf(stderr, "BURST_COUNT:%d ", cf->output.burst_count);
                        fprintf(stderr, "EOP:%X\n", cf->output.end_of_program);
                        break;
                case V_SQ_CF_WORD1_SQ_CF_INST_JUMP:
@@ -1888,7 +1936,7 @@ static void r600_vertex_data_type(enum pipe_format pformat, unsigned *format,
        }
 
        switch (desc->channel[i].type) {
-               /* Half-floats, floats, doubles */
+       /* Half-floats, floats, ints */
        case UTIL_FORMAT_TYPE_FLOAT:
                switch (desc->channel[i].size) {
                case 16:
@@ -1900,8 +1948,6 @@ static void r600_vertex_data_type(enum pipe_format pformat, unsigned *format,
                                *format = FMT_16_16_FLOAT;
                                break;
                        case 3:
-                               *format = FMT_16_16_16_FLOAT;
-                               break;
                        case 4:
                                *format = FMT_16_16_16_16_FLOAT;
                                break;
@@ -1941,8 +1987,6 @@ static void r600_vertex_data_type(enum pipe_format pformat, unsigned *format,
                                *format = FMT_8_8;
                                break;
                        case 3:
-                       //      *format = FMT_8_8_8; /* fails piglit draw-vertices test */
-                       //      break;
                        case 4:
                                *format = FMT_8_8_8_8;
                                break;
@@ -1957,8 +2001,6 @@ static void r600_vertex_data_type(enum pipe_format pformat, unsigned *format,
                                *format = FMT_16_16;
                                break;
                        case 3:
-                       //      *format = FMT_16_16_16; /* fails piglit draw-vertices test */
-                       //      break;
                        case 4:
                                *format = FMT_16_16_16_16;
                                break;