Home | History | Annotate | Download | only in i965
      1 /*
      2  * Copyright  2016 Intel Corporation
      3  *
      4  * Permission is hereby granted, free of charge, to any person obtaining a
      5  * copy of this software and associated documentation files (the "Software"),
      6  * to deal in the Software without restriction, including without limitation
      7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
      8  * and/or sell copies of the Software, and to permit persons to whom the
      9  * Software is furnished to do so, subject to the following conditions:
     10  *
     11  * The above copyright notice and this permission notice (including the next
     12  * paragraph) shall be included in all copies or substantial portions of the
     13  * Software.
     14  *
     15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
     16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
     17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
     18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
     19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
     20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
     21  * IN THE SOFTWARE.
     22  */
     23 
     24 static inline struct blorp_address
     25 dynamic_state_address(struct blorp_batch *batch, uint32_t offset)
     26 {
     27    assert(batch->blorp->driver_ctx == batch->driver_batch);
     28    struct brw_context *brw = batch->driver_batch;
     29 
     30    return (struct blorp_address) {
     31       .buffer = brw->batch.state.bo,
     32       .offset = offset,
     33    };
     34 }
     35 
     36 static inline struct blorp_address
     37 instruction_state_address(struct blorp_batch *batch, uint32_t offset)
     38 {
     39    assert(batch->blorp->driver_ctx == batch->driver_batch);
     40    struct brw_context *brw = batch->driver_batch;
     41 
     42    return (struct blorp_address) {
     43       .buffer = brw->cache.bo,
     44       .offset = offset,
     45    };
     46 }
     47 
     48 static struct blorp_address
     49 blorp_emit_vs_state(struct blorp_batch *batch,
     50                     const struct blorp_params *params)
     51 {
     52    assert(batch->blorp->driver_ctx == batch->driver_batch);
     53    struct brw_context *brw = batch->driver_batch;
     54 
     55    uint32_t offset;
     56    blorp_emit_dynamic(batch, GENX(VS_STATE), vs, 64, &offset) {
     57       vs.Enable = false;
     58       vs.URBEntryAllocationSize = brw->urb.vsize - 1;
     59 #if GEN_GEN == 5
     60       vs.NumberofURBEntries = brw->urb.nr_vs_entries >> 2;
     61 #else
     62       vs.NumberofURBEntries = brw->urb.nr_vs_entries;
     63 #endif
     64    }
     65 
     66    return dynamic_state_address(batch, offset);
     67 }
     68 
     69 static struct blorp_address
     70 blorp_emit_sf_state(struct blorp_batch *batch,
     71                     const struct blorp_params *params)
     72 {
     73    assert(batch->blorp->driver_ctx == batch->driver_batch);
     74    struct brw_context *brw = batch->driver_batch;
     75    const struct brw_sf_prog_data *prog_data = params->sf_prog_data;
     76 
     77    uint32_t offset;
     78    blorp_emit_dynamic(batch, GENX(SF_STATE), sf, 64, &offset) {
     79 #if GEN_GEN == 4
     80       sf.KernelStartPointer =
     81          instruction_state_address(batch, params->sf_prog_kernel);
     82 #else
     83       sf.KernelStartPointer = params->sf_prog_kernel;
     84 #endif
     85       sf.GRFRegisterCount = DIV_ROUND_UP(prog_data->total_grf, 16) - 1;
     86       sf.VertexURBEntryReadLength = prog_data->urb_read_length;
     87       sf.VertexURBEntryReadOffset = BRW_SF_URB_ENTRY_READ_OFFSET;
     88       sf.DispatchGRFStartRegisterForURBData = 3;
     89 
     90       sf.URBEntryAllocationSize = brw->urb.sfsize - 1;
     91       sf.NumberofURBEntries = brw->urb.nr_sf_entries;
     92 
     93 #if GEN_GEN == 5
     94       sf.MaximumNumberofThreads = MIN2(48, brw->urb.nr_sf_entries) - 1;
     95 #else
     96       sf.MaximumNumberofThreads = MIN2(24, brw->urb.nr_sf_entries) - 1;
     97 #endif
     98 
     99       sf.ViewportTransformEnable = false;
    100 
    101       sf.CullMode = CULLMODE_NONE;
    102    }
    103 
    104    return dynamic_state_address(batch, offset);
    105 }
    106 
    107 static struct blorp_address
    108 blorp_emit_wm_state(struct blorp_batch *batch,
    109                     const struct blorp_params *params)
    110 {
    111    const struct brw_wm_prog_data *prog_data = params->wm_prog_data;
    112 
    113    uint32_t offset;
    114    blorp_emit_dynamic(batch, GENX(WM_STATE), wm, 64, &offset) {
    115       if (params->src.enabled) {
    116          /* Iron Lake can't do sampler prefetch */
    117          wm.SamplerCount = (GEN_GEN != 5);
    118          wm.BindingTableEntryCount = 2;
    119          uint32_t sampler = blorp_emit_sampler_state(batch, params);
    120          wm.SamplerStatePointer = dynamic_state_address(batch, sampler);
    121       }
    122 
    123       if (prog_data) {
    124          wm.DispatchGRFStartRegisterForConstantSetupData0 =
    125             prog_data->base.dispatch_grf_start_reg;
    126          wm.SetupURBEntryReadLength = prog_data->num_varying_inputs * 2;
    127          wm.SetupURBEntryReadOffset = 0;
    128 
    129          wm.DepthCoefficientURBReadOffset = 1;
    130          wm.PixelShaderKillsPixel = prog_data->uses_kill;
    131          wm.ThreadDispatchEnable = true;
    132          wm.EarlyDepthTestEnable = true;
    133 
    134          wm._8PixelDispatchEnable = prog_data->dispatch_8;
    135          wm._16PixelDispatchEnable = prog_data->dispatch_16;
    136 
    137 #if GEN_GEN == 4
    138          wm.KernelStartPointer0 =
    139             instruction_state_address(batch, params->wm_prog_kernel);
    140          wm.GRFRegisterCount0 = prog_data->reg_blocks_0;
    141 #else
    142          wm.KernelStartPointer0 = params->wm_prog_kernel;
    143          wm.GRFRegisterCount0 = prog_data->reg_blocks_0;
    144          wm.KernelStartPointer2 =
    145             params->wm_prog_kernel + prog_data->prog_offset_2;
    146          wm.GRFRegisterCount2 = prog_data->reg_blocks_2;
    147 #endif
    148       }
    149 
    150       wm.MaximumNumberofThreads =
    151          batch->blorp->compiler->devinfo->max_wm_threads - 1;
    152    }
    153 
    154    return dynamic_state_address(batch, offset);
    155 }
    156 
    157 static struct blorp_address
    158 blorp_emit_color_calc_state(struct blorp_batch *batch,
    159                             const struct blorp_params *params)
    160 {
    161    uint32_t cc_viewport = blorp_emit_cc_viewport(batch, params);
    162 
    163    uint32_t offset;
    164    blorp_emit_dynamic(batch, GENX(COLOR_CALC_STATE), cc, 64, &offset) {
    165       cc.CCViewportStatePointer = dynamic_state_address(batch, cc_viewport);
    166    }
    167 
    168    return dynamic_state_address(batch, offset);
    169 }
    170 
    171 static void
    172 blorp_emit_pipeline(struct blorp_batch *batch,
    173                     const struct blorp_params *params)
    174 {
    175    assert(batch->blorp->driver_ctx == batch->driver_batch);
    176    struct brw_context *brw = batch->driver_batch;
    177 
    178    emit_urb_config(batch, params);
    179 
    180    blorp_emit(batch, GENX(3DSTATE_PIPELINED_POINTERS), pp) {
    181       pp.PointertoVSState = blorp_emit_vs_state(batch, params);
    182       pp.GSEnable = false;
    183       pp.ClipEnable = false;
    184       pp.PointertoSFState = blorp_emit_sf_state(batch, params);
    185       pp.PointertoWMState = blorp_emit_wm_state(batch, params);
    186       pp.PointertoColorCalcState = blorp_emit_color_calc_state(batch, params);
    187    }
    188 
    189    brw_upload_urb_fence(brw);
    190 
    191    blorp_emit(batch, GENX(CS_URB_STATE), curb);
    192    blorp_emit(batch, GENX(CONSTANT_BUFFER), curb);
    193 }
    194