diff options
Diffstat (limited to 'src/mesa/drivers/dri/i965/brw_wm_fp.c')
| -rw-r--r-- | src/mesa/drivers/dri/i965/brw_wm_fp.c | 51 |
1 files changed, 45 insertions, 6 deletions
diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c index 606baab967..5dc076a825 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_fp.c +++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c @@ -226,9 +226,42 @@ static struct prog_instruction * emit_op(struct brw_wm_compile *c, 0, 0, 0, /* tex unit, target, shadow */ src0, src1, src2); } - +/* Many Mesa opcodes produce the same value across all the result channels. + * We'd rather not have to support that splatting in the opcode implementations, + * and brw_wm_pass*.c wants to optimize them out by shuffling references around + * anyway. We can easily get both by emitting the opcode to one channel, and + * then MOVing it to the others, which brw_wm_pass*.c already understands. + */ +static struct prog_instruction *emit_scalar_insn(struct brw_wm_compile *c, + const struct prog_instruction *inst0) +{ + struct prog_instruction *inst; + unsigned int dst_chan; + unsigned int other_channel_mask; + + if (inst0->DstReg.WriteMask == 0) + return NULL; + + dst_chan = _mesa_ffs(inst0->DstReg.WriteMask) - 1; + inst = get_fp_inst(c); + *inst = *inst0; + inst->DstReg.WriteMask = 1 << dst_chan; + + other_channel_mask = inst0->DstReg.WriteMask & ~(1 << dst_chan); + if (other_channel_mask != 0) { + inst = emit_op(c, + OPCODE_MOV, + dst_mask(inst0->DstReg, other_channel_mask), + 0, + src_swizzle1(src_reg_from_dst(inst0->DstReg), dst_chan), + src_undef(), + src_undef()); + } + return inst; +} + /*********************************************************************** * Special instructions for interpolation and other tasks @@ -690,7 +723,7 @@ static void precalc_tex( struct brw_wm_compile *c, /* tmp0 = 1 / tmp1 */ emit_op(c, OPCODE_RCP, - tmp0, + dst_mask(tmp0, WRITEMASK_X), 0, tmp1src, src_undef(), @@ -701,7 +734,7 @@ static void precalc_tex( struct brw_wm_compile *c, tmpcoord, 0, src0, - tmp0src, + src_swizzle1(tmp0src, SWIZZLE_X), src_undef()); release_temp(c, tmp0); @@ -724,7 +757,11 @@ static void precalc_tex( struct brw_wm_compile *c, tmpcoord, 0, inst->SrcReg[0], - scale, + src_swizzle(scale, + SWIZZLE_X, + SWIZZLE_Y, + SWIZZLE_ONE, + SWIZZLE_ONE), src_undef()); coord = src_reg_from_dst(tmpcoord); @@ -1141,9 +1178,11 @@ void brw_wm_pass_fp( struct brw_wm_compile *c ) break; case OPCODE_PRINT: break; - default: - emit_insn(c, inst); + if (brw_wm_is_scalar_result(inst->Opcode)) + emit_scalar_insn(c, inst); + else + emit_insn(c, inst); break; } } |