* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
-#include "r600_asm.h"
-#include "r600_context.h"
+#include <stdio.h>
+#include <errno.h>
#include "util/u_memory.h"
+#include "r600_pipe.h"
#include "r600_sq.h"
#include "r600_opcodes.h"
-#include <stdio.h>
-#include <errno.h>
+#include "r600_asm.h"
static inline unsigned int r600_bc_get_num_operands(struct r600_bc_alu *alu)
{
case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SIN:
case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_COS:
return 1;
-
default: R600_ERR(
"Need instruction operand number for 0x%x.\n", alu->inst);
};
/* cf can contains only alu or only vtx or only tex */
if (bc->cf_last == NULL || bc->cf_last->inst != (type << 3) ||
bc->force_add_cf) {
- /* at most 128 slots, one add alu can add 4 slots + 4 constant worst case */
r = r600_bc_add_cf(bc);
if (r) {
free(nalu);
} else {
LIST_ADDTAIL(&nalu->bs_list, &bc->cf_last->curr_bs_head->bs_list);
}
- if (alu->last && (bc->cf_last->ndw >> 1) >= 124) {
+ /* at most 128 slots, one add alu can add 4 slots + 4 constants(2 slots)
+ * worst case */
+ if (alu->last && (bc->cf_last->ndw >> 1) >= 120) {
bc->force_add_cf = 1;
}
/* number of gpr == the last gpr used in any alu */
bc->cf_last->ndw += 2;
bc->ndw += 2;
- if (bc->use_mem_constant)
- bc->cf_last->kcache0_mode = 2;
+ bc->cf_last->kcache0_mode = 2;
/* process cur ALU instructions for bank swizzle */
if (alu->last) {
/* cf can contains only alu or only vtx or only tex */
if (bc->cf_last == NULL ||
(bc->cf_last->inst != V_SQ_CF_WORD1_SQ_CF_INST_VTX &&
- bc->cf_last->inst != V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC)) {
+ bc->cf_last->inst != V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC) ||
+ bc->force_add_cf) {
r = r600_bc_add_cf(bc);
if (r) {
free(nvtx);
/* each fetch use 4 dwords */
bc->cf_last->ndw += 4;
bc->ndw += 4;
+ if ((bc->ndw / 4) > 7)
+ bc->force_add_cf = 1;
return 0;
}
/* cf can contains only alu or only vtx or only tex */
if (bc->cf_last == NULL ||
- bc->cf_last->inst != V_SQ_CF_WORD1_SQ_CF_INST_TEX) {
+ bc->cf_last->inst != V_SQ_CF_WORD1_SQ_CF_INST_TEX ||
+ bc->force_add_cf) {
r = r600_bc_add_cf(bc);
if (r) {
free(ntex);
/* each texture fetch use 4 dwords */
bc->cf_last->ndw += 4;
bc->ndw += 4;
+ if ((bc->ndw / 4) > 7)
+ bc->force_add_cf = 1;
return 0;
}
S_SQ_VTX_WORD1_DST_SEL_Y(vtx->dst_sel_y) |
S_SQ_VTX_WORD1_DST_SEL_Z(vtx->dst_sel_z) |
S_SQ_VTX_WORD1_DST_SEL_W(vtx->dst_sel_w) |
- S_SQ_VTX_WORD1_USE_CONST_FIELDS(1) |
+ S_SQ_VTX_WORD1_USE_CONST_FIELDS(vtx->use_const_fields) |
+ S_SQ_VTX_WORD1_DATA_FORMAT(vtx->data_format) |
+ S_SQ_VTX_WORD1_NUM_FORMAT_ALL(vtx->num_format_all) |
+ S_SQ_VTX_WORD1_FORMAT_COMP_ALL(vtx->format_comp_all) |
+ S_SQ_VTX_WORD1_SRF_MODE_ALL(vtx->srf_mode_all) |
S_SQ_VTX_WORD1_GPR_DST_GPR(vtx->dst_gpr);
bc->bytecode[id++] = S_SQ_VTX_WORD2_MEGA_FETCH(1);
bc->bytecode[id++] = 0;
bc->bytecode[id++] = S_SQ_CF_ALU_WORD1_CF_INST(cf->inst >> 3) |
S_SQ_CF_ALU_WORD1_BARRIER(1) |
+ S_SQ_CF_ALU_WORD1_USES_WATERFALL(bc->chiprev == 0 ? cf->r6xx_uses_waterfall : 0) |
S_SQ_CF_ALU_WORD1_COUNT((cf->ndw / 2) - 1);
break;
case V_SQ_CF_WORD1_SQ_CF_INST_TEX: