#include "intel/common/gen_sample_positions.h"
#include "iris_batch.h"
#include "iris_context.h"
+#include "iris_defines.h"
#include "iris_pipe.h"
#include "iris_resource.h"
#include "genxml/gen_macros.h"
#include "genxml/genX_bits.h"
-#define MOCS_WB (2 << 1)
+#if GEN_GEN == 8
+#define MOCS_PTE 0x18
+#define MOCS_WB 0x78
+#else
+#define MOCS_PTE (1 << 1)
+#define MOCS_WB (2 << 1)
+#endif
+
+static uint32_t
+mocs(struct iris_bo *bo)
+{
+ return bo && bo->external ? MOCS_PTE : MOCS_WB;
+}
/**
* Statically assert that PIPE_* enums match the hardware packets.
#define iris_emit_lri(b, r, v) _iris_emit_lri(b, GENX(r##_num), v)
static void
-_iris_emit_lrr(struct iris_batch *batch, uint32_t src, uint32_t dst)
+_iris_emit_lrr(struct iris_batch *batch, uint32_t dst, uint32_t src)
{
iris_emit_cmd(batch, GENX(MI_LOAD_REGISTER_REG), lrr) {
lrr.SourceRegisterAddress = src;
* updated occasionally. See iris_binder.c for the details there.
*/
iris_emit_cmd(batch, GENX(STATE_BASE_ADDRESS), sba) {
- #if 0
- // XXX: MOCS is stupid for this.
- sba.GeneralStateMemoryObjectControlState = MOCS_WB;
- sba.StatelessDataPortAccessMemoryObjectControlState = MOCS_WB;
- sba.DynamicStateMemoryObjectControlState = MOCS_WB;
- sba.IndirectObjectMemoryObjectControlState = MOCS_WB;
- sba.InstructionMemoryObjectControlState = MOCS_WB;
- sba.BindlessSurfaceStateMemoryObjectControlState = MOCS_WB;
- #endif
+ sba.GeneralStateMOCS = MOCS_WB;
+ sba.StatelessDataPortAccessMOCS = MOCS_WB;
+ sba.DynamicStateMOCS = MOCS_WB;
+ sba.IndirectObjectMOCS = MOCS_WB;
+ sba.InstructionMOCS = MOCS_WB;
sba.GeneralStateBaseAddressModifyEnable = true;
sba.DynamicStateBaseAddressModifyEnable = true;
sba.InstructionBaseAddressModifyEnable = true;
sba.GeneralStateBufferSizeModifyEnable = true;
sba.DynamicStateBufferSizeModifyEnable = true;
+#if (GEN_GEN >= 9)
sba.BindlessSurfaceStateBaseAddressModifyEnable = true;
+ sba.BindlessSurfaceStateMOCS = MOCS_WB;
+#endif
sba.IndirectObjectBufferSizeModifyEnable = true;
sba.InstructionBuffersizeModifyEnable = true;
}
}
+static void
+iris_emit_l3_config(struct iris_batch *batch, const struct gen_l3_config *cfg,
+ bool has_slm, bool wants_dc_cache)
+{
+ uint32_t reg_val;
+ iris_pack_state(GENX(L3CNTLREG), ®_val, reg) {
+ reg.SLMEnable = has_slm;
+#if GEN_GEN == 11
+ /* WA_1406697149: Bit 9 "Error Detection Behavior Control" must be set
+ * in L3CNTLREG register. The default setting of the bit is not the
+ * desirable behavior.
+ */
+ reg.ErrorDetectionBehaviorControl = true;
+#endif
+ reg.URBAllocation = cfg->n[GEN_L3P_URB];
+ reg.ROAllocation = cfg->n[GEN_L3P_RO];
+ reg.DCAllocation = cfg->n[GEN_L3P_DC];
+ reg.AllAllocation = cfg->n[GEN_L3P_ALL];
+ }
+ iris_emit_lri(batch, L3CNTLREG, reg_val);
+}
+
+static void
+iris_emit_default_l3_config(struct iris_batch *batch,
+ const struct gen_device_info *devinfo,
+ bool compute)
+{
+ bool wants_dc_cache = true;
+ bool has_slm = compute;
+ const struct gen_l3_weights w =
+ gen_get_default_l3_weights(devinfo, wants_dc_cache, has_slm);
+ const struct gen_l3_config *cfg = gen_get_l3_config(devinfo, w);
+ iris_emit_l3_config(batch, cfg, has_slm, wants_dc_cache);
+}
+
/**
* Upload the initial GPU state for a render context.
*
emit_pipeline_select(batch, _3D);
+ iris_emit_default_l3_config(batch, devinfo, false);
+
init_state_base_address(batch);
- // XXX: INSTPM on Gen8
+#if GEN_GEN >= 9
iris_pack_state(GENX(CS_DEBUG_MODE2), ®_val, reg) {
reg.CONSTANT_BUFFERAddressOffsetDisable = true;
reg.CONSTANT_BUFFERAddressOffsetDisableMask = true;
}
iris_emit_lri(batch, CS_DEBUG_MODE2, reg_val);
+#else
+ iris_pack_state(GENX(INSTPM), ®_val, reg) {
+ reg.CONSTANT_BUFFERAddressOffsetDisable = true;
+ reg.CONSTANT_BUFFERAddressOffsetDisableMask = true;
+ }
+ iris_emit_lri(batch, INSTPM, reg_val);
+#endif
#if GEN_GEN == 9
iris_pack_state(GENX(CACHE_MODE_1), ®_val, reg) {
GEN_SAMPLE_POS_2X(pat._2xSample);
GEN_SAMPLE_POS_4X(pat._4xSample);
GEN_SAMPLE_POS_8X(pat._8xSample);
+#if GEN_GEN >= 9
GEN_SAMPLE_POS_16X(pat._16xSample);
+#endif
}
/* Use the legacy AA line coverage computation. */
iris_emit_cmd(batch, GENX(3DSTATE_WM_HZ_OP), foo);
/* No polygon stippling offsets are necessary. */
- // XXX: may need to set an offset for origin-UL framebuffers
+ /* TODO: may need to set an offset for origin-UL framebuffers */
iris_emit_cmd(batch, GENX(3DSTATE_POLY_STIPPLE_OFFSET), foo);
/* Set a static partitioning of the push constant area. */
- // XXX: this may be a bad idea...could starve the push ringbuffers...
+ /* TODO: this may be a bad idea...could starve the push ringbuffers... */
for (int i = 0; i <= MESA_SHADER_FRAGMENT; i++) {
iris_emit_cmd(batch, GENX(3DSTATE_PUSH_CONSTANT_ALLOC_VS), alloc) {
alloc._3DCommandSubOpcode = 18 + i;
emit_pipeline_select(batch, GPGPU);
- const bool has_slm = true;
- const bool wants_dc_cache = true;
-
- const struct gen_l3_weights w =
- gen_get_default_l3_weights(devinfo, wants_dc_cache, has_slm);
- const struct gen_l3_config *cfg = gen_get_l3_config(devinfo, w);
-
- uint32_t reg_val;
- iris_pack_state(GENX(L3CNTLREG), ®_val, reg) {
- reg.SLMEnable = has_slm;
-#if GEN_GEN == 11
- /* WA_1406697149: Bit 9 "Error Detection Behavior Control" must be set
- * in L3CNTLREG register. The default setting of the bit is not the
- * desirable behavior.
- */
- reg.ErrorDetectionBehaviorControl = true;
-#endif
- reg.URBAllocation = cfg->n[GEN_L3P_URB];
- reg.ROAllocation = cfg->n[GEN_L3P_RO];
- reg.DCAllocation = cfg->n[GEN_L3P_DC];
- reg.AllAllocation = cfg->n[GEN_L3P_ALL];
- }
- iris_emit_lri(batch, L3CNTLREG, reg_val);
+ iris_emit_default_l3_config(batch, devinfo, true);
init_state_base_address(batch);
}
struct iris_vertex_buffer_state {
- /** The 3DSTATE_VERTEX_BUFFERS hardware packet. */
- uint32_t vertex_buffers[1 + 33 * GENX(VERTEX_BUFFER_STATE_length)];
+ /** The VERTEX_BUFFER_STATE hardware structure. */
+ uint32_t state[GENX(VERTEX_BUFFER_STATE_length)];
/** The resource to source vertex data from. */
- struct pipe_resource *resources[33];
-
- /** The number of bound vertex buffers. */
- unsigned num_buffers;
+ struct pipe_resource *resource;
};
struct iris_depth_buffer_state {
* packets which vary by generation.
*/
struct iris_genx_state {
- /** SF_CLIP_VIEWPORT */
- uint32_t sf_cl_vp[GENX(SF_CLIP_VIEWPORT_length) * IRIS_MAX_VIEWPORTS];
+ struct iris_vertex_buffer_state vertex_buffers[33];
- struct iris_vertex_buffer_state vertex_buffers;
struct iris_depth_buffer_state depth_buffer;
uint32_t so_buffers[4 * GENX(3DSTATE_SO_BUFFER_length)];
- uint32_t streamout[4 * GENX(3DSTATE_STREAMOUT_length)];
};
/**
BRW_MAX_DRAW_BUFFERS * GENX(BLEND_STATE_ENTRY_length)];
bool alpha_to_coverage; /* for shader key */
+
+ /** Bitfield of whether blending is enabled for RT[i] - for aux resolves */
+ uint8_t blend_enables;
+
+ /** Bitfield of whether color writes are enabled for RT[i] */
+ uint8_t color_write_enables;
};
+static enum pipe_blendfactor
+fix_blendfactor(enum pipe_blendfactor f, bool alpha_to_one)
+{
+ if (alpha_to_one) {
+ if (f == PIPE_BLENDFACTOR_SRC1_ALPHA)
+ return PIPE_BLENDFACTOR_ONE;
+
+ if (f == PIPE_BLENDFACTOR_INV_SRC1_ALPHA)
+ return PIPE_BLENDFACTOR_ZERO;
+ }
+
+ return f;
+}
+
/**
* The pipe->create_blend_state() driver hook.
*
struct iris_blend_state *cso = malloc(sizeof(struct iris_blend_state));
uint32_t *blend_entry = cso->blend_state + GENX(BLEND_STATE_length);
+ cso->blend_enables = 0;
+ cso->color_write_enables = 0;
+ STATIC_ASSERT(BRW_MAX_DRAW_BUFFERS <= 8);
+
cso->alpha_to_coverage = state->alpha_to_coverage;
bool indep_alpha_blend = false;
const struct pipe_rt_blend_state *rt =
&state->rt[state->independent_blend_enable ? i : 0];
+ enum pipe_blendfactor src_rgb =
+ fix_blendfactor(rt->rgb_src_factor, state->alpha_to_one);
+ enum pipe_blendfactor src_alpha =
+ fix_blendfactor(rt->alpha_src_factor, state->alpha_to_one);
+ enum pipe_blendfactor dst_rgb =
+ fix_blendfactor(rt->rgb_dst_factor, state->alpha_to_one);
+ enum pipe_blendfactor dst_alpha =
+ fix_blendfactor(rt->alpha_dst_factor, state->alpha_to_one);
+
if (rt->rgb_func != rt->alpha_func ||
- rt->rgb_src_factor != rt->alpha_src_factor ||
- rt->rgb_dst_factor != rt->alpha_dst_factor)
+ src_rgb != src_alpha || dst_rgb != dst_alpha)
indep_alpha_blend = true;
+ if (rt->blend_enable)
+ cso->blend_enables |= 1u << i;
+
+ if (rt->colormask)
+ cso->color_write_enables |= 1u << i;
+
iris_pack_state(GENX(BLEND_STATE_ENTRY), blend_entry, be) {
be.LogicOpEnable = state->logicop_enable;
be.LogicOpFunction = state->logicop_func;
be.ColorBlendFunction = rt->rgb_func;
be.AlphaBlendFunction = rt->alpha_func;
- be.SourceBlendFactor = rt->rgb_src_factor;
- be.SourceAlphaBlendFactor = rt->alpha_src_factor;
- be.DestinationBlendFactor = rt->rgb_dst_factor;
- be.DestinationAlphaBlendFactor = rt->alpha_dst_factor;
+ be.SourceBlendFactor = src_rgb;
+ be.SourceAlphaBlendFactor = src_alpha;
+ be.DestinationBlendFactor = dst_rgb;
+ be.DestinationAlphaBlendFactor = dst_alpha;
be.WriteDisableRed = !(rt->colormask & PIPE_MASK_R);
be.WriteDisableGreen = !(rt->colormask & PIPE_MASK_G);
pb.ColorBufferBlendEnable = state->rt[0].blend_enable;
- pb.SourceBlendFactor = state->rt[0].rgb_src_factor;
- pb.SourceAlphaBlendFactor = state->rt[0].alpha_src_factor;
- pb.DestinationBlendFactor = state->rt[0].rgb_dst_factor;
- pb.DestinationAlphaBlendFactor = state->rt[0].alpha_dst_factor;
+ pb.SourceBlendFactor =
+ fix_blendfactor(state->rt[0].rgb_src_factor, state->alpha_to_one);
+ pb.SourceAlphaBlendFactor =
+ fix_blendfactor(state->rt[0].alpha_src_factor, state->alpha_to_one);
+ pb.DestinationBlendFactor =
+ fix_blendfactor(state->rt[0].rgb_dst_factor, state->alpha_to_one);
+ pb.DestinationAlphaBlendFactor =
+ fix_blendfactor(state->rt[0].alpha_dst_factor, state->alpha_to_one);
}
iris_pack_state(GENX(BLEND_STATE), cso->blend_state, bs) {
iris_bind_blend_state(struct pipe_context *ctx, void *state)
{
struct iris_context *ice = (struct iris_context *) ctx;
- ice->state.cso_blend = state;
+ struct iris_blend_state *cso = state;
+
+ ice->state.cso_blend = cso;
+ ice->state.blend_enables = cso ? cso->blend_enables : 0;
+
ice->state.dirty |= IRIS_DIRTY_PS_BLEND;
ice->state.dirty |= IRIS_DIRTY_BLEND_STATE;
ice->state.dirty |= ice->state.dirty_for_nos[IRIS_NOS_BLEND];
}
+/**
+ * Return true if the FS writes to any color outputs which are not disabled
+ * via color masking.
+ */
+static bool
+has_writeable_rt(const struct iris_blend_state *cso_blend,
+ const struct shader_info *fs_info)
+{
+ if (!fs_info)
+ return false;
+
+ unsigned rt_outputs = fs_info->outputs_written >> FRAG_RESULT_DATA0;
+
+ if (fs_info->outputs_written & BITFIELD64_BIT(FRAG_RESULT_COLOR))
+ rt_outputs = (1 << BRW_MAX_DRAW_BUFFERS) - 1;
+
+ return cso_blend->color_write_enables & rt_outputs;
+}
+
/**
* Gallium CSO for depth, stencil, and alpha testing state.
*/
bool flatshade_first; /* for stream output */
bool clamp_fragment_color; /* for shader state */
bool light_twoside; /* for shader state */
- bool rasterizer_discard; /* for 3DSTATE_STREAMOUT */
+ bool rasterizer_discard; /* for 3DSTATE_STREAMOUT and 3DSTATE_CLIP */
bool half_pixel_center; /* for 3DSTATE_MULTISAMPLE */
bool line_stipple_enable;
bool poly_stipple_enable;
struct iris_rasterizer_state *cso =
malloc(sizeof(struct iris_rasterizer_state));
-#if 0
- point_quad_rasterization -> SBE?
-
- not necessary?
- {
- poly_smooth
- force_persample_interp - ?
- bottom_edge_rule
-
- offset_units_unscaled - cap not exposed
- }
- #endif
-
- // XXX: it may make more sense just to store the pipe_rasterizer_state,
- // we're copying a lot of booleans here. But we don't need all of them...
-
cso->multisample = state->multisample;
cso->force_persample_interp = state->force_persample_interp;
cso->clip_halfz = state->clip_halfz;
state->line_smooth ? _10pixels : _05pixels;
sf.LastPixelEnable = state->line_last_pixel;
sf.LineWidth = line_width;
- sf.SmoothPointEnable = state->point_smooth || state->multisample;
+ sf.SmoothPointEnable = (state->point_smooth || state->multisample) &&
+ !state->point_quad_rasterization;
sf.PointWidthSource = state->point_size_per_vertex ? Vertex : State;
sf.PointWidth = state->point_size;
rr.GlobalDepthOffsetConstant = state->offset_units * 2;
rr.GlobalDepthOffsetScale = state->offset_scale;
rr.GlobalDepthOffsetClamp = state->offset_clamp;
- rr.SmoothPointEnable = state->point_smooth || state->multisample;
+ rr.SmoothPointEnable = state->point_smooth;
rr.AntialiasingEnable = state->line_smooth;
rr.ScissorRectangleEnable = state->scissor;
+#if GEN_GEN >= 9
rr.ViewportZNearClipTestEnable = state->depth_clip_near;
rr.ViewportZFarClipTestEnable = state->depth_clip_far;
- //rr.ConservativeRasterizationEnable = not yet supported by Gallium...
+#else
+ rr.ViewportZClipTestEnable = (state->depth_clip_near || state->depth_clip_far);
+#endif
+ /* TODO: ConservativeRasterizationEnable */
}
iris_pack_command(GENX(3DSTATE_CLIP), cso->clip, cl) {
if (cso_changed(line_stipple_enable) || cso_changed(poly_stipple_enable))
ice->state.dirty |= IRIS_DIRTY_WM;
- if (cso_changed(rasterizer_discard) || cso_changed(flatshade_first))
+ if (cso_changed(rasterizer_discard))
+ ice->state.dirty |= IRIS_DIRTY_STREAMOUT | IRIS_DIRTY_CLIP;
+
+ if (cso_changed(flatshade_first))
ice->state.dirty |= IRIS_DIRTY_STREAMOUT;
if (cso_changed(depth_clip_near) || cso_changed(depth_clip_far) ||
cso_changed(clip_halfz))
ice->state.dirty |= IRIS_DIRTY_CC_VIEWPORT;
- if (cso_changed(sprite_coord_enable) || cso_changed(light_twoside))
+ if (cso_changed(sprite_coord_enable) ||
+ cso_changed(sprite_coord_mode) ||
+ cso_changed(light_twoside))
ice->state.dirty |= IRIS_DIRTY_SBE;
}
struct iris_shader_state *shs = &ice->state.shaders[stage];
assert(start + count <= IRIS_MAX_TEXTURE_SAMPLERS);
- shs->num_samplers = MAX2(shs->num_samplers, start + count);
for (int i = 0; i < count; i++) {
shs->samplers[start + i] = states[i];
}
- // XXX: count may include NULLs
-
/* Assemble the SAMPLER_STATEs into a contiguous table that lives
* in the dynamic state memory zone, so we can point to it via the
* 3DSTATE_SAMPLER_STATE_POINTERS_* commands.
unsigned size)
{
const struct isl_format_layout *fmtl = isl_format_get_layout(format);
- const unsigned cpp = fmtl->bpb / 8;
+ const unsigned cpp = format == ISL_FORMAT_RAW ? 1 : fmtl->bpb / 8;
/* The ARB_texture_buffer_specification says:
*
.size_B = final_size,
.format = format,
.stride_B = cpp,
- .mocs = MOCS_WB);
+ .mocs = mocs(bo));
+}
+
+#define SURFACE_STATE_ALIGNMENT 64
+
+/**
+ * Allocate several contiguous SURFACE_STATE structures, one for each
+ * supported auxiliary surface mode.
+ */
+static void *
+alloc_surface_states(struct u_upload_mgr *mgr,
+ struct iris_state_ref *ref,
+ unsigned aux_usages)
+{
+ const unsigned surf_size = 4 * GENX(RENDER_SURFACE_STATE_length);
+
+ /* If this changes, update this to explicitly align pointers */
+ STATIC_ASSERT(surf_size == SURFACE_STATE_ALIGNMENT);
+
+ assert(aux_usages != 0);
+
+ void *map =
+ upload_state(mgr, ref, util_bitcount(aux_usages) * surf_size,
+ SURFACE_STATE_ALIGNMENT);
+
+ ref->offset += iris_bo_offset_from_base_address(iris_resource_bo(ref->res));
+
+ return map;
+}
+
+static void
+fill_surface_state(struct isl_device *isl_dev,
+ void *map,
+ struct iris_resource *res,
+ struct isl_view *view,
+ unsigned aux_usage)
+{
+ struct isl_surf_fill_state_info f = {
+ .surf = &res->surf,
+ .view = view,
+ .mocs = mocs(res->bo),
+ .address = res->bo->gtt_offset,
+ };
+
+ if (aux_usage != ISL_AUX_USAGE_NONE) {
+ f.aux_surf = &res->aux.surf;
+ f.aux_usage = aux_usage;
+ f.aux_address = res->aux.bo->gtt_offset + res->aux.offset;
+ // XXX: clear color
+ }
+
+ isl_surf_fill_state_s(isl_dev, map, &f);
}
/**
pipe_reference_init(&isv->base.reference, 1);
pipe_resource_reference(&isv->base.texture, tex);
- void *map = upload_state(ice->state.surface_uploader, &isv->surface_state,
- 4 * GENX(RENDER_SURFACE_STATE_length), 64);
- if (!unlikely(map))
- return NULL;
-
- struct iris_bo *state_bo = iris_resource_bo(isv->surface_state.res);
- isv->surface_state.offset += iris_bo_offset_from_base_address(state_bo);
-
if (util_format_is_depth_or_stencil(tmpl->format)) {
struct iris_resource *zres, *sres;
const struct util_format_description *desc =
isv->res = (struct iris_resource *) tex;
+ void *map = alloc_surface_states(ice->state.surface_uploader,
+ &isv->surface_state,
+ isv->res->aux.possible_usages);
+ if (!unlikely(map))
+ return NULL;
+
isl_surf_usage_flags_t usage = ISL_SURF_USAGE_TEXTURE_BIT;
if (isv->base.target == PIPE_TEXTURE_CUBE ||
isv->view.array_len =
tmpl->u.tex.last_layer - tmpl->u.tex.first_layer + 1;
- isl_surf_fill_state(&screen->isl_dev, map,
- .surf = &isv->res->surf, .view = &isv->view,
- .mocs = MOCS_WB,
- .address = isv->res->bo->gtt_offset);
- // .aux_surf =
- // .clear_color = clear_color,
+ unsigned aux_modes = isv->res->aux.possible_usages;
+ while (aux_modes) {
+ enum isl_aux_usage aux_usage = u_bit_scan(&aux_modes);
+
+ fill_surface_state(&screen->isl_dev, map, isv->res, &isv->view,
+ aux_usage);
+
+ map += SURFACE_STATE_ALIGNMENT;
+ }
} else {
fill_buffer_surface_state(&screen->isl_dev, isv->res->bo, map,
isv->view.format, tmpl->u.buf.offset,
return psurf;
- void *map = upload_state(ice->state.surface_uploader, &surf->surface_state,
- 4 * GENX(RENDER_SURFACE_STATE_length), 64);
+ void *map = alloc_surface_states(ice->state.surface_uploader,
+ &surf->surface_state,
+ res->aux.possible_usages);
if (!unlikely(map))
return NULL;
- struct iris_bo *state_bo = iris_resource_bo(surf->surface_state.res);
- surf->surface_state.offset += iris_bo_offset_from_base_address(state_bo);
+ unsigned aux_modes = res->aux.possible_usages;
+ while (aux_modes) {
+ enum isl_aux_usage aux_usage = u_bit_scan(&aux_modes);
- isl_surf_fill_state(&screen->isl_dev, map,
- .surf = &res->surf, .view = &surf->view,
- .mocs = MOCS_WB,
- .address = res->bo->gtt_offset);
- // .aux_surf =
- // .clear_color = clear_color,
+ fill_surface_state(&screen->isl_dev, map, res, &surf->view, aux_usage);
+
+ map += SURFACE_STATE_ALIGNMENT;
+ }
return psurf;
}
+#if GEN_GEN < 9
+static void
+fill_default_image_param(struct brw_image_param *param)
+{
+ memset(param, 0, sizeof(*param));
+ /* Set the swizzling shifts to all-ones to effectively disable swizzling --
+ * See emit_address_calculation() in brw_fs_surface_builder.cpp for a more
+ * detailed explanation of these parameters.
+ */
+ param->swizzling[0] = 0xff;
+ param->swizzling[1] = 0xff;
+}
+
+static void
+fill_buffer_image_param(struct brw_image_param *param,
+ enum pipe_format pfmt,
+ unsigned size)
+{
+ const unsigned cpp = util_format_get_blocksize(pfmt);
+
+ fill_default_image_param(param);
+ param->size[0] = size / cpp;
+ param->stride[0] = cpp;
+}
+#else
+#define isl_surf_fill_image_param(x, ...)
+#define fill_default_image_param(x, ...)
+#define fill_buffer_image_param(x, ...)
+#endif
+
/**
* The pipe->set_shader_images() driver hook.
*/
gl_shader_stage stage = stage_from_pipe(p_stage);
struct iris_shader_state *shs = &ice->state.shaders[stage];
- shs->num_images = MAX2(shs->num_images, start_slot + count);
+ shs->bound_image_views &= ~u_bit_consecutive(start_slot, count);
for (unsigned i = 0; i < count; i++) {
if (p_images && p_images[i].resource) {
struct iris_resource *res = (void *) img->resource;
pipe_resource_reference(&shs->image[start_slot + i].res, &res->base);
+ shs->bound_image_views |= 1 << (start_slot + i);
+
res->bind_history |= PIPE_BIND_SHADER_IMAGE;
// XXX: these are not retained forever, use a separate uploader?
void *map =
- upload_state(ice->state.surface_uploader,
- &shs->image[start_slot + i].surface_state,
- 4 * GENX(RENDER_SURFACE_STATE_length), 64);
+ alloc_surface_states(ice->state.surface_uploader,
+ &shs->image[start_slot + i].surface_state,
+ 1 << ISL_AUX_USAGE_NONE);
if (!unlikely(map)) {
pipe_resource_reference(&shs->image[start_slot + i].res, NULL);
return;
}
- struct iris_bo *surf_state_bo =
- iris_resource_bo(shs->image[start_slot + i].surface_state.res);
- shs->image[start_slot + i].surface_state.offset +=
- iris_bo_offset_from_base_address(surf_state_bo);
-
isl_surf_usage_flags_t usage = ISL_SURF_USAGE_STORAGE_BIT;
- enum isl_format isl_format =
+ enum isl_format isl_fmt =
iris_format_for_usage(devinfo, img->format, usage).fmt;
- if (img->shader_access & PIPE_IMAGE_ACCESS_READ)
- isl_format = isl_lower_storage_image_format(devinfo, isl_format);
+ bool untyped_fallback = false;
+
+ if (img->shader_access & PIPE_IMAGE_ACCESS_READ) {
+ /* On Gen8, try to use typed surfaces reads (which support a
+ * limited number of formats), and if not possible, fall back
+ * to untyped reads.
+ */
+ untyped_fallback = GEN_GEN == 8 &&
+ !isl_has_matching_typed_storage_image_format(devinfo, isl_fmt);
+
+ if (untyped_fallback)
+ isl_fmt = ISL_FORMAT_RAW;
+ else
+ isl_fmt = isl_lower_storage_image_format(devinfo, isl_fmt);
+ }
shs->image[start_slot + i].access = img->shader_access;
if (res->base.target != PIPE_BUFFER) {
struct isl_view view = {
- .format = isl_format,
+ .format = isl_fmt,
.base_level = img->u.tex.level,
.levels = 1,
.base_array_layer = img->u.tex.first_layer,
.usage = usage,
};
- isl_surf_fill_state(&screen->isl_dev, map,
- .surf = &res->surf, .view = &view,
- .mocs = MOCS_WB,
- .address = res->bo->gtt_offset);
- // .aux_surf =
- // .clear_color = clear_color,
+ if (untyped_fallback) {
+ fill_buffer_surface_state(&screen->isl_dev, res->bo, map,
+ isl_fmt, 0, res->bo->size);
+ } else {
+ /* Images don't support compression */
+ unsigned aux_modes = 1 << ISL_AUX_USAGE_NONE;
+ while (aux_modes) {
+ enum isl_aux_usage usage = u_bit_scan(&aux_modes);
+
+ fill_surface_state(&screen->isl_dev, map, res, &view, usage);
+
+ map += SURFACE_STATE_ALIGNMENT;
+ }
+ }
+
+ isl_surf_fill_image_param(&screen->isl_dev,
+ &shs->image[start_slot + i].param,
+ &res->surf, &view);
} else {
fill_buffer_surface_state(&screen->isl_dev, res->bo, map,
- isl_format, img->u.buf.offset,
+ isl_fmt, img->u.buf.offset,
img->u.buf.size);
+ fill_buffer_image_param(&shs->image[start_slot + i].param,
+ img->format, img->u.buf.size);
}
} else {
pipe_resource_reference(&shs->image[start_slot + i].res, NULL);
pipe_resource_reference(&shs->image[start_slot + i].surface_state.res,
NULL);
+ fill_default_image_param(&shs->image[start_slot + i].param);
}
}
ice->state.dirty |= IRIS_DIRTY_BINDINGS_VS << stage;
+
+ /* Broadwell also needs brw_image_params re-uploaded */
+ if (GEN_GEN < 9) {
+ ice->state.dirty |= IRIS_DIRTY_CONSTANTS_VS << stage;
+ shs->cbuf0_needs_upload = true;
+ }
}
gl_shader_stage stage = stage_from_pipe(p_stage);
struct iris_shader_state *shs = &ice->state.shaders[stage];
- unsigned i;
- for (i = 0; i < count; i++) {
+ shs->bound_sampler_views &= ~u_bit_consecutive(start, count);
+
+ for (unsigned i = 0; i < count; i++) {
pipe_sampler_view_reference((struct pipe_sampler_view **)
- &shs->textures[i], views[i]);
+ &shs->textures[start + i], views[i]);
struct iris_sampler_view *view = (void *) views[i];
- if (view)
+ if (view) {
view->res->bind_history |= PIPE_BIND_SAMPLER_VIEW;
+ shs->bound_sampler_views |= 1 << (start + i);
+ }
}
- for (; i < shs->num_textures; i++) {
- pipe_sampler_view_reference((struct pipe_sampler_view **)
- &shs->textures[i], NULL);
- }
-
- shs->num_textures = count;
ice->state.dirty |= (IRIS_DIRTY_BINDINGS_VS << stage);
}
{
struct iris_context *ice = (struct iris_context *) ctx;
memcpy(&ice->state.stencil_ref, state, sizeof(*state));
- ice->state.dirty |= IRIS_DIRTY_WM_DEPTH_STENCIL;
+ if (GEN_GEN == 8)
+ ice->state.dirty |= IRIS_DIRTY_COLOR_CALC_STATE;
+ else
+ ice->state.dirty |= IRIS_DIRTY_WM_DEPTH_STENCIL;
}
static float
return copysignf(state->scale[axis], sign) + state->translate[axis];
}
-#if 0
static void
calculate_guardband_size(uint32_t fb_width, uint32_t fb_height,
float m00, float m11, float m30, float m31,
*ymax = 0.0f;
}
}
-#endif
/**
* The pipe->set_viewport_states() driver hook.
const struct pipe_viewport_state *states)
{
struct iris_context *ice = (struct iris_context *) ctx;
- struct iris_genx_state *genx = ice->state.genx;
- uint32_t *vp_map =
- &genx->sf_cl_vp[start_slot * GENX(SF_CLIP_VIEWPORT_length)];
- for (unsigned i = 0; i < count; i++) {
- const struct pipe_viewport_state *state = &states[i];
-
- memcpy(&ice->state.viewports[start_slot + i], state, sizeof(*state));
-
- iris_pack_state(GENX(SF_CLIP_VIEWPORT), vp_map, vp) {
- vp.ViewportMatrixElementm00 = state->scale[0];
- vp.ViewportMatrixElementm11 = state->scale[1];
- vp.ViewportMatrixElementm22 = state->scale[2];
- vp.ViewportMatrixElementm30 = state->translate[0];
- vp.ViewportMatrixElementm31 = state->translate[1];
- vp.ViewportMatrixElementm32 = state->translate[2];
- /* XXX: in i965 this is computed based on the drawbuffer size,
- * but we don't have that here...
- */
- vp.XMinClipGuardband = -1.0;
- vp.XMaxClipGuardband = 1.0;
- vp.YMinClipGuardband = -1.0;
- vp.YMaxClipGuardband = 1.0;
- vp.XMinViewPort = viewport_extent(state, 0, -1.0f);
- vp.XMaxViewPort = viewport_extent(state, 0, 1.0f) - 1;
- vp.YMinViewPort = viewport_extent(state, 1, -1.0f);
- vp.YMaxViewPort = viewport_extent(state, 1, 1.0f) - 1;
- }
-
- vp_map += GENX(SF_CLIP_VIEWPORT_length);
- }
+ memcpy(&ice->state.viewports[start_slot], states, sizeof(*states) * count);
ice->state.dirty |= IRIS_DIRTY_SF_CL_VIEWPORT;
struct iris_resource *stencil_res;
unsigned samples = util_framebuffer_get_num_samples(state);
+ unsigned layers = util_framebuffer_get_num_layers(state);
if (cso->samples != samples) {
ice->state.dirty |= IRIS_DIRTY_MULTISAMPLE;
ice->state.dirty |= IRIS_DIRTY_BLEND_STATE;
}
- if ((cso->layers == 0) != (state->layers == 0)) {
+ if ((cso->layers == 0) != (layers == 0)) {
ice->state.dirty |= IRIS_DIRTY_CLIP;
}
+ if (cso->width != state->width || cso->height != state->height) {
+ ice->state.dirty |= IRIS_DIRTY_SF_CL_VIEWPORT;
+ }
+
util_copy_framebuffer_state(cso, state);
cso->samples = samples;
+ cso->layers = layers;
struct iris_depth_buffer_state *cso_z = &ice->state.genx->depth_buffer;
.swizzle = ISL_SWIZZLE_IDENTITY,
};
- struct isl_depth_stencil_hiz_emit_info info = {
- .view = &view,
- .mocs = MOCS_WB,
- };
+ struct isl_depth_stencil_hiz_emit_info info = { .view = &view };
if (cso->zsbuf) {
iris_get_depth_stencil_resources(cso->zsbuf->texture, &zres,
info.depth_surf = &zres->surf;
info.depth_address = zres->bo->gtt_offset;
- info.hiz_usage = ISL_AUX_USAGE_NONE;
+ info.mocs = mocs(zres->bo);
view.format = zres->surf.format;
+
+ if (iris_resource_level_has_hiz(zres, view.base_level)) {
+ info.hiz_usage = ISL_AUX_USAGE_HIZ;
+ info.hiz_surf = &zres->aux.surf;
+ info.hiz_address = zres->aux.bo->gtt_offset;
+ }
}
if (stencil_res) {
view.usage |= ISL_SURF_USAGE_STENCIL_BIT;
info.stencil_surf = &stencil_res->surf;
info.stencil_address = stencil_res->bo->gtt_offset;
- if (!zres)
+ if (!zres) {
view.format = stencil_res->surf.format;
+ info.mocs = mocs(stencil_res->bo);
+ }
}
}
res->bo->size - cbuf->data.offset),
.format = ISL_FORMAT_R32G32B32A32_FLOAT,
.stride_B = 1,
- .mocs = MOCS_WB)
+ .mocs = mocs(res->bo))
}
/**
uint32_t sysval = shader->system_values[i];
uint32_t value = 0;
- if (BRW_PARAM_BUILTIN_IS_CLIP_PLANE(sysval)) {
+ if (BRW_PARAM_DOMAIN(sysval) == BRW_PARAM_DOMAIN_IMAGE) {
+ unsigned img = BRW_PARAM_IMAGE_IDX(sysval);
+ unsigned offset = BRW_PARAM_IMAGE_OFFSET(sysval);
+ struct brw_image_param *param = &shs->image[img].param;
+
+ assert(offset < sizeof(struct brw_image_param));
+ value = ((uint32_t *) param)[offset];
+ } else if (sysval == BRW_PARAM_BUILTIN_ZERO) {
+ value = 0;
+ } else if (BRW_PARAM_BUILTIN_IS_CLIP_PLANE(sysval)) {
int plane = BRW_PARAM_BUILTIN_CLIP_PLANE_IDX(sysval);
int comp = BRW_PARAM_BUILTIN_CLIP_PLANE_COMP(sysval);
value = fui(ice->state.clip_planes.ucp[plane][comp]);
+ } else if (sysval == BRW_PARAM_BUILTIN_PATCH_VERTICES_IN) {
+ if (stage == MESA_SHADER_TESS_CTRL) {
+ value = ice->state.vertices_per_patch;
+ } else {
+ assert(stage == MESA_SHADER_TESS_EVAL);
+ const struct shader_info *tcs_info =
+ iris_get_shader_info(ice, MESA_SHADER_TESS_CTRL);
+ assert(tcs_info);
+
+ value = tcs_info->tess.tcs_vertices_out;
+ }
} else {
assert(!"unhandled system value");
}
res->bo->size - buffer->buffer_offset),
.format = ISL_FORMAT_RAW,
.stride_B = 1,
- .mocs = MOCS_WB);
+ .mocs = mocs(res->bo));
} else {
pipe_resource_reference(&shs->ssbo[start_slot + i], NULL);
pipe_resource_reference(&shs->ssbo_surface_state[start_slot + i].res,
free(state);
}
-static void
-iris_free_vertex_buffers(struct iris_vertex_buffer_state *cso)
-{
- for (unsigned i = 0; i < cso->num_buffers; i++)
- pipe_resource_reference(&cso->resources[i], NULL);
-}
-
/**
* The pipe->set_vertex_buffers() driver hook.
*
const struct pipe_vertex_buffer *buffers)
{
struct iris_context *ice = (struct iris_context *) ctx;
- struct iris_vertex_buffer_state *cso = &ice->state.genx->vertex_buffers;
-
- iris_free_vertex_buffers(&ice->state.genx->vertex_buffers);
-
- if (!buffers)
- count = 0;
-
- cso->num_buffers = count;
+ struct iris_genx_state *genx = ice->state.genx;
- iris_pack_command(GENX(3DSTATE_VERTEX_BUFFERS), cso->vertex_buffers, vb) {
- vb.DWordLength = 4 * MAX2(cso->num_buffers, 1) - 1;
- }
+ ice->state.bound_vertex_buffers &= ~u_bit_consecutive64(start_slot, count);
- uint32_t *vb_pack_dest = &cso->vertex_buffers[1];
+ for (unsigned i = 0; i < count; i++) {
+ const struct pipe_vertex_buffer *buffer = buffers ? &buffers[i] : NULL;
+ struct iris_vertex_buffer_state *state =
+ &genx->vertex_buffers[start_slot + i];
- if (count == 0) {
- iris_pack_state(GENX(VERTEX_BUFFER_STATE), vb_pack_dest, vb) {
- vb.VertexBufferIndex = start_slot;
- vb.NullVertexBuffer = true;
- vb.AddressModifyEnable = true;
+ if (!buffer) {
+ pipe_resource_reference(&state->resource, NULL);
+ continue;
}
- }
- for (unsigned i = 0; i < count; i++) {
- assert(!buffers[i].is_user_buffer);
+ assert(!buffer->is_user_buffer);
- pipe_resource_reference(&cso->resources[i], buffers[i].buffer.resource);
- struct iris_resource *res = (void *) cso->resources[i];
+ pipe_resource_reference(&state->resource, buffer->buffer.resource);
+ struct iris_resource *res = (void *) state->resource;
- if (res)
+ if (res) {
+ ice->state.bound_vertex_buffers |= 1ull << (start_slot + i);
res->bind_history |= PIPE_BIND_VERTEX_BUFFER;
+ }
- iris_pack_state(GENX(VERTEX_BUFFER_STATE), vb_pack_dest, vb) {
+ iris_pack_state(GENX(VERTEX_BUFFER_STATE), state->state, vb) {
vb.VertexBufferIndex = start_slot + i;
- vb.MOCS = MOCS_WB;
vb.AddressModifyEnable = true;
- vb.BufferPitch = buffers[i].stride;
+ vb.BufferPitch = buffer->stride;
if (res) {
vb.BufferSize = res->bo->size;
vb.BufferStartingAddress =
- ro_bo(NULL, res->bo->gtt_offset + buffers[i].buffer_offset);
+ ro_bo(NULL, res->bo->gtt_offset + (int) buffer->buffer_offset);
+ vb.MOCS = mocs(res->bo);
} else {
vb.NullVertexBuffer = true;
}
}
-
- vb_pack_dest += GENX(VERTEX_BUFFER_STATE_length);
}
ice->state.dirty |= IRIS_DIRTY_VERTEX_BUFFERS;
ice->state.dirty |= IRIS_DIRTY_VERTEX_ELEMENTS;
}
-/**
- * Gallium CSO for stream output (transform feedback) targets.
- */
-struct iris_stream_output_target {
- struct pipe_stream_output_target base;
-
- uint32_t so_buffer[GENX(3DSTATE_SO_BUFFER_length)];
-
- /** Storage holding the offset where we're writing in the buffer */
- struct iris_state_ref offset;
-};
-
/**
* The pipe->create_stream_output_target() driver hook.
*
cso->base.buffer_size = buffer_size;
cso->base.context = ctx;
- upload_state(ctx->stream_uploader, &cso->offset, 4 * sizeof(uint32_t), 4);
-
- iris_pack_command(GENX(3DSTATE_SO_BUFFER), cso->so_buffer, sob) {
- sob.SurfaceBaseAddress =
- rw_bo(NULL, res->bo->gtt_offset + buffer_offset);
- sob.SOBufferEnable = true;
- sob.StreamOffsetWriteEnable = true;
- sob.StreamOutputBufferOffsetAddressEnable = true;
- sob.MOCS = MOCS_WB; // XXX: MOCS
-
- sob.SurfaceSize = MAX2(buffer_size / 4, 1) - 1;
-
- /* .SOBufferIndex, .StreamOffset, and .StreamOutputBufferOffsetAddress
- * are filled in later when we have stream IDs.
- */
- }
+ upload_state(ctx->stream_uploader, &cso->offset, sizeof(uint32_t), 4);
return &cso->base;
}
}
struct iris_stream_output_target *tgt = (void *) targets[i];
+ struct iris_resource *res = (void *) tgt->base.buffer;
/* Note that offsets[i] will either be 0, causing us to zero
* the value in the buffer, or 0xFFFFFFFF, which happens to mean
*/
assert(offsets[i] == 0 || offsets[i] == 0xFFFFFFFF);
- uint32_t dynamic[GENX(3DSTATE_SO_BUFFER_length)];
- iris_pack_state(GENX(3DSTATE_SO_BUFFER), dynamic, dyns) {
- dyns.SOBufferIndex = i;
- dyns.StreamOffset = offsets[i];
- dyns.StreamOutputBufferOffsetAddress =
- rw_bo(NULL, iris_resource_bo(tgt->offset.res)->gtt_offset + tgt->offset.offset + i * sizeof(uint32_t));
- }
-
- for (uint32_t j = 0; j < GENX(3DSTATE_SO_BUFFER_length); j++) {
- so_buffers[j] = tgt->so_buffer[j] | dynamic[j];
+ iris_pack_command(GENX(3DSTATE_SO_BUFFER), so_buffers, sob) {
+ sob.SurfaceBaseAddress =
+ rw_bo(NULL, res->bo->gtt_offset + tgt->base.buffer_offset);
+ sob.SOBufferEnable = true;
+ sob.StreamOffsetWriteEnable = true;
+ sob.StreamOutputBufferOffsetAddressEnable = true;
+ sob.MOCS = mocs(res->bo);
+
+ sob.SurfaceSize = MAX2(tgt->base.buffer_size / 4, 1) - 1;
+
+ sob.SOBufferIndex = i;
+ sob.StreamOffset = offsets[i];
+ sob.StreamOutputBufferOffsetAddress =
+ rw_bo(NULL, iris_resource_bo(tgt->offset.res)->gtt_offset +
+ tgt->offset.offset);
}
}
/* XXX: this should be generated when putting programs in place */
- // XXX: raster->sprite_coord_enable
-
for (int fs_attr = 0; fs_attr < VARYING_SLOT_MAX; fs_attr++) {
const int input_index = wm_prog_data->urb_setup[fs_attr];
if (input_index < 0 || input_index >= 16)
sbe.ForceVertexURBEntryReadLength = true;
sbe.ConstantInterpolationEnable = wm_prog_data->flat_inputs;
sbe.PointSpriteTextureCoordinateEnable = sprite_coord_overrides;
-
+#if GEN_GEN >= 9
for (int i = 0; i < 32; i++) {
sbe.AttributeActiveComponentFormat[i] = ACTIVE_COMPONENT_XYZW;
}
+#endif
}
iris_emit_sbe_swiz(batch, ice, urb_read_offset, sprite_coord_overrides);
iris_populate_fs_key(const struct iris_context *ice,
struct brw_wm_prog_key *key)
{
- /* XXX: dirty flags? */
const struct pipe_framebuffer_state *fb = &ice->state.framebuffer;
const struct iris_depth_stencil_alpha_state *zsa = ice->state.cso_zsa;
const struct iris_rasterizer_state *rast = ice->state.cso_rast;
key->coherent_fb_fetch = true;
- // XXX: uint64_t input_slots_valid; - for >16 inputs
-
- // XXX: key->force_dual_color_blend for unigine
- // XXX: respect hint for high_quality_derivatives:1;
+ /* TODO: support key->force_dual_color_blend for Unigine */
+ /* TODO: Respect glHint for key->high_quality_derivatives */
}
static void
{
}
-#if 0
- // XXX: these need to go in INIT_THREAD_DISPATCH_FIELDS
- pkt.SamplerCount = \
- DIV_ROUND_UP(CLAMP(stage_state->sampler_count, 0, 16), 4); \
-
-#endif
-
static uint64_t
KSP(const struct iris_compiled_shader *shader)
{
return iris_bo_offset_from_base_address(res->bo) + shader->assembly.offset;
}
-// Gen11 workaround table #2056 WABTPPrefetchDisable suggests to disable
-// prefetching of binding tables in A0 and B0 steppings. XXX: Revisit
-// this WA on C0 stepping.
+/* Gen11 workaround table #2056 WABTPPrefetchDisable suggests to disable
+ * prefetching of binding tables in A0 and B0 steppings. XXX: Revisit
+ * this WA on C0 stepping.
+ *
+ * TODO: Fill out SamplerCount for prefetching?
+ */
#define INIT_THREAD_DISPATCH_FIELDS(pkt, prefix, stage) \
pkt.KernelStartPointer = KSP(shader); \
pkt.Enable = true; \
\
if (prog_data->total_scratch) { \
- uint32_t scratch_addr = \
+ struct iris_bo *bo = \
iris_get_scratch_space(ice, prog_data->total_scratch, stage); \
+ uint32_t scratch_addr = bo->gtt_offset; \
pkt.PerThreadScratchSpace = ffs(prog_data->total_scratch) - 11; \
pkt.ScratchSpaceBasePointer = rw_bo(NULL, scratch_addr); \
}
iris_pack_command(GENX(3DSTATE_PS), ps_state, ps) {
ps.VectorMaskEnable = true;
- //ps.SamplerCount = ...
// XXX: WABTPPrefetchDisable, see above, drop at C0
ps.BindingTableEntryCount = GEN_GEN == 11 ? 0 :
prog_data->binding_table.size_bytes / 4;
ps.FloatingPointMode = prog_data->use_alt_mode;
ps.MaximumNumberofThreadsPerPSD = 64 - (GEN_GEN == 8 ? 2 : 1);
- ps.PushConstantEnable = shader->num_system_values > 0 ||
- prog_data->ubo_ranges[0].length > 0;
+ ps.PushConstantEnable = prog_data->ubo_ranges[0].length > 0;
/* From the documentation for this packet:
* "If the PS kernel does not need the Position XY Offsets to
KSP(shader) + brw_wm_prog_data_prog_offset(wm_prog_data, ps, 2);
if (prog_data->total_scratch) {
- uint32_t scratch_addr =
+ struct iris_bo *bo =
iris_get_scratch_space(ice, prog_data->total_scratch,
MESA_SHADER_FRAGMENT);
+ uint32_t scratch_addr = bo->gtt_offset;
ps.PerThreadScratchSpace = ffs(prog_data->total_scratch) - 11;
ps.ScratchSpaceBasePointer = rw_bo(NULL, scratch_addr);
}
iris_pack_command(GENX(3DSTATE_PS_EXTRA), psx_state, psx) {
psx.PixelShaderValid = true;
psx.PixelShaderComputedDepthMode = wm_prog_data->computed_depth_mode;
- // XXX: alpha test / alpha to coverage :/
- psx.PixelShaderKillsPixel = wm_prog_data->uses_kill ||
- wm_prog_data->uses_omask;
+ psx.PixelShaderKillsPixel = wm_prog_data->uses_kill;
psx.AttributeEnable = wm_prog_data->num_varying_inputs != 0;
psx.PixelShaderUsesSourceDepth = wm_prog_data->uses_src_depth;
psx.PixelShaderUsesSourceW = wm_prog_data->uses_src_w;
psx.PixelShaderIsPerSample = wm_prog_data->persample_dispatch;
+ psx.oMaskPresenttoRenderTarget = wm_prog_data->uses_omask;
+#if GEN_GEN >= 9
if (wm_prog_data->uses_sample_mask) {
/* TODO: conservative rasterization */
if (wm_prog_data->post_depth_coverage)
psx.InputCoverageMaskState = ICMS_NORMAL;
}
- psx.oMaskPresenttoRenderTarget = wm_prog_data->uses_omask;
psx.PixelShaderPullsBary = wm_prog_data->pulls_bary;
psx.PixelShaderComputesStencil = wm_prog_data->computed_stencil;
-
+#else
+ psx.PixelShaderUsesInputCoverageMask = wm_prog_data->uses_sample_mask;
+#endif
// XXX: UAV bit
}
}
return ice->state.null_fb.offset;
}
+static uint32_t
+surf_state_offset_for_aux(struct iris_resource *res,
+ enum isl_aux_usage aux_usage)
+{
+ return SURFACE_STATE_ALIGNMENT *
+ util_bitcount(res->aux.possible_usages & ((1 << aux_usage) - 1));
+}
+
/**
* Add a surface to the validation list, as well as the buffer containing
* the corresponding SURFACE_STATE.
static uint32_t
use_surface(struct iris_batch *batch,
struct pipe_surface *p_surf,
- bool writeable)
+ bool writeable,
+ enum isl_aux_usage aux_usage)
{
struct iris_surface *surf = (void *) p_surf;
+ struct iris_resource *res = (void *) p_surf->texture;
iris_use_pinned_bo(batch, iris_resource_bo(p_surf->texture), writeable);
iris_use_pinned_bo(batch, iris_resource_bo(surf->surface_state.res), false);
- return surf->surface_state.offset;
+ if (res->aux.bo)
+ iris_use_pinned_bo(batch, res->aux.bo, writeable);
+
+ return surf->surface_state.offset +
+ surf_state_offset_for_aux(res, aux_usage);
}
static uint32_t
-use_sampler_view(struct iris_batch *batch, struct iris_sampler_view *isv)
+use_sampler_view(struct iris_context *ice,
+ struct iris_batch *batch,
+ struct iris_sampler_view *isv)
{
+ // XXX: ASTC hacks
+ enum isl_aux_usage aux_usage =
+ iris_resource_texture_aux_usage(ice, isv->res, isv->view.format, 0);
+
iris_use_pinned_bo(batch, isv->res->bo, false);
iris_use_pinned_bo(batch, iris_resource_bo(isv->surface_state.res), false);
- return isv->surface_state.offset;
+ if (isv->res->aux.bo)
+ iris_use_pinned_bo(batch, isv->res->aux.bo, false);
+
+ return isv->surface_state.offset +
+ surf_state_offset_for_aux(isv->res, aux_usage);
}
static uint32_t
if (!shs->image[i].res)
return use_null_surface(batch, ice);
+ struct iris_resource *res = (void *) shs->image[i].res;
struct iris_state_ref *surf_state = &shs->image[i].surface_state;
+ bool write = shs->image[i].access & PIPE_IMAGE_ACCESS_WRITE;
- iris_use_pinned_bo(batch, iris_resource_bo(shs->image[i].res),
- shs->image[i].access & PIPE_IMAGE_ACCESS_WRITE);
+ iris_use_pinned_bo(batch, res->bo, write);
iris_use_pinned_bo(batch, iris_resource_bo(surf_state->res), false);
+ if (res->aux.bo)
+ iris_use_pinned_bo(batch, res->aux.bo, write);
+
return surf_state->offset;
}
/* Note that cso_fb->nr_cbufs == fs_key->nr_color_regions. */
if (cso_fb->nr_cbufs) {
for (unsigned i = 0; i < cso_fb->nr_cbufs; i++) {
- uint32_t addr =
- cso_fb->cbufs[i] ? use_surface(batch, cso_fb->cbufs[i], true)
- : use_null_fb_surface(batch, ice);
+ uint32_t addr;
+ if (cso_fb->cbufs[i]) {
+ addr = use_surface(batch, cso_fb->cbufs[i], true,
+ ice->state.draw_aux_usage[i]);
+ } else {
+ addr = use_null_fb_surface(batch, ice);
+ }
push_bt_entry(addr);
}
} else {
}
}
- bt_assert(texture_start, info->num_textures > 0);
+ unsigned num_textures = util_last_bit(info->textures_used);
+
+ bt_assert(texture_start, num_textures > 0);
- for (int i = 0; i < info->num_textures; i++) {
+ for (int i = 0; i < num_textures; i++) {
struct iris_sampler_view *view = shs->textures[i];
- uint32_t addr = view ? use_sampler_view(batch, view)
+ uint32_t addr = view ? use_sampler_view(ice, batch, view)
: use_null_surface(batch, ice);
push_bt_entry(addr);
}
push_bt_entry(addr);
}
- const int num_ubos = iris_get_shader_num_ubos(ice, stage);
+ bt_assert(ubo_start, shader->num_cbufs > 0);
- bt_assert(ubo_start, num_ubos > 0);
-
- for (int i = 0; i < num_ubos; i++) {
+ for (int i = 0; i < shader->num_cbufs; i++) {
uint32_t addr = use_const_buffer(batch, ice, &shs->constbuf[i]);
push_bt_entry(addr);
}
}
#if 0
- // XXX: not implemented yet
+ /* XXX: YUV surfaces not implemented yet */
bt_assert(plane_start[1], ...);
bt_assert(plane_start[2], ...);
#endif
struct iris_batch *batch,
const struct pipe_draw_info *draw)
{
- // XXX: whack IRIS_SHADER_DIRTY_BINDING_TABLE on new batch
+ struct iris_genx_state *genx = ice->state.genx;
const uint64_t clean = ~ice->state.dirty;
iris_use_optional_res(batch, ice->state.last_res.scissor, false);
}
+ if (ice->state.streamout_active && (clean & IRIS_DIRTY_SO_BUFFERS)) {
+ for (int i = 0; i < 4; i++) {
+ struct iris_stream_output_target *tgt =
+ (void *) ice->state.so_target[i];
+ if (tgt) {
+ iris_use_pinned_bo(batch, iris_resource_bo(tgt->base.buffer),
+ true);
+ iris_use_pinned_bo(batch, iris_resource_bo(tgt->offset.res),
+ true);
+ }
+ }
+ }
+
for (int stage = 0; stage <= MESA_SHADER_FRAGMENT; stage++) {
if (!(clean & (IRIS_DIRTY_CONSTANTS_VS << stage)))
continue;
for (int stage = 0; stage <= MESA_SHADER_FRAGMENT; stage++) {
if (clean & (IRIS_DIRTY_VS << stage)) {
struct iris_compiled_shader *shader = ice->shaders.prog[stage];
+
if (shader) {
struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
iris_use_pinned_bo(batch, bo, false);
- }
- // XXX: scratch buffer
+ struct brw_stage_prog_data *prog_data = shader->prog_data;
+
+ if (prog_data->total_scratch > 0) {
+ struct iris_bo *bo =
+ iris_get_scratch_space(ice, prog_data->total_scratch, stage);
+ iris_use_pinned_bo(batch, bo, true);
+ }
+ }
}
}
struct iris_resource *zres, *sres;
iris_get_depth_stencil_resources(cso_fb->zsbuf->texture,
&zres, &sres);
- // XXX: might not be writable...
- if (zres)
- iris_use_pinned_bo(batch, zres->bo, true);
- if (sres)
- iris_use_pinned_bo(batch, sres->bo, true);
+ if (zres) {
+ iris_cache_flush_for_depth(batch, zres->bo);
+
+ iris_use_pinned_bo(batch, zres->bo,
+ ice->state.depth_writes_enabled);
+ if (zres->aux.bo) {
+ iris_use_pinned_bo(batch, zres->aux.bo,
+ ice->state.depth_writes_enabled);
+ }
+ }
+
+ if (sres) {
+ iris_cache_flush_for_depth(batch, sres->bo);
+
+ iris_use_pinned_bo(batch, sres->bo,
+ ice->state.stencil_writes_enabled);
+ }
}
}
}
if (clean & IRIS_DIRTY_VERTEX_BUFFERS) {
- struct iris_vertex_buffer_state *cso = &ice->state.genx->vertex_buffers;
- for (unsigned i = 0; i < cso->num_buffers; i++) {
- struct iris_resource *res = (void *) cso->resources[i];
- iris_use_pinned_bo(batch, res->bo, false);
+ uint64_t bound = ice->state.bound_vertex_buffers;
+ while (bound) {
+ const int i = u_bit_scan64(&bound);
+ struct pipe_resource *res = genx->vertex_buffers[i].resource;
+ iris_use_pinned_bo(batch, iris_resource_bo(res), false);
}
}
}
if (clean & IRIS_DIRTY_CS) {
struct iris_compiled_shader *shader = ice->shaders.prog[stage];
+
if (shader) {
struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
iris_use_pinned_bo(batch, bo, false);
- }
- // XXX: scratch buffer
+ struct brw_stage_prog_data *prog_data = shader->prog_data;
+
+ if (prog_data->total_scratch > 0) {
+ struct iris_bo *bo =
+ iris_get_scratch_space(ice, prog_data->total_scratch, stage);
+ iris_use_pinned_bo(batch, bo, true);
+ }
+ }
}
}
flush_for_state_base_change(batch);
iris_emit_cmd(batch, GENX(STATE_BASE_ADDRESS), sba) {
- // XXX: sba.SurfaceStateMemoryObjectControlState = MOCS_WB;
+ sba.SurfaceStateMOCS = MOCS_WB;
sba.SurfaceStateBaseAddressModifyEnable = true;
sba.SurfaceStateBaseAddress = ro_bo(binder->bo, 0);
}
}
if (dirty & IRIS_DIRTY_SF_CL_VIEWPORT) {
+ struct pipe_framebuffer_state *cso_fb = &ice->state.framebuffer;
+ uint32_t sf_cl_vp_address;
+ uint32_t *vp_map =
+ stream_state(batch, ice->state.dynamic_uploader,
+ &ice->state.last_res.sf_cl_vp,
+ 4 * ice->state.num_viewports *
+ GENX(SF_CLIP_VIEWPORT_length), 64, &sf_cl_vp_address);
+
+ for (unsigned i = 0; i < ice->state.num_viewports; i++) {
+ const struct pipe_viewport_state *state = &ice->state.viewports[i];
+ float gb_xmin, gb_xmax, gb_ymin, gb_ymax;
+
+ float vp_xmin = viewport_extent(state, 0, -1.0f);
+ float vp_xmax = viewport_extent(state, 0, 1.0f);
+ float vp_ymin = viewport_extent(state, 1, -1.0f);
+ float vp_ymax = viewport_extent(state, 1, 1.0f);
+
+ calculate_guardband_size(cso_fb->width, cso_fb->height,
+ state->scale[0], state->scale[1],
+ state->translate[0], state->translate[1],
+ &gb_xmin, &gb_xmax, &gb_ymin, &gb_ymax);
+
+ iris_pack_state(GENX(SF_CLIP_VIEWPORT), vp_map, vp) {
+ vp.ViewportMatrixElementm00 = state->scale[0];
+ vp.ViewportMatrixElementm11 = state->scale[1];
+ vp.ViewportMatrixElementm22 = state->scale[2];
+ vp.ViewportMatrixElementm30 = state->translate[0];
+ vp.ViewportMatrixElementm31 = state->translate[1];
+ vp.ViewportMatrixElementm32 = state->translate[2];
+ vp.XMinClipGuardband = gb_xmin;
+ vp.XMaxClipGuardband = gb_xmax;
+ vp.YMinClipGuardband = gb_ymin;
+ vp.YMaxClipGuardband = gb_ymax;
+ vp.XMinViewPort = MAX2(vp_xmin, 0);
+ vp.XMaxViewPort = MIN2(vp_xmax, cso_fb->width) - 1;
+ vp.YMinViewPort = MAX2(vp_ymin, 0);
+ vp.YMaxViewPort = MIN2(vp_ymax, cso_fb->height) - 1;
+ }
+
+ vp_map += GENX(SF_CLIP_VIEWPORT_length);
+ }
+
iris_emit_cmd(batch, GENX(3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP), ptr) {
- ptr.SFClipViewportPointer =
- emit_state(batch, ice->state.dynamic_uploader,
- &ice->state.last_res.sf_cl_vp,
- genx->sf_cl_vp, 4 * GENX(SF_CLIP_VIEWPORT_length) *
- ice->state.num_viewports, 64);
+ ptr.SFClipViewportPointer = sf_cl_vp_address;
}
}
- /* XXX: L3 State */
-
- // XXX: this is only flagged at setup, we assume a static configuration
if (dirty & IRIS_DIRTY_URB) {
iris_upload_urb_config(ice, batch);
}
struct pipe_framebuffer_state *cso_fb = &ice->state.framebuffer;
struct iris_depth_stencil_alpha_state *cso_zsa = ice->state.cso_zsa;
const int header_dwords = GENX(BLEND_STATE_length);
- const int rt_dwords = cso_fb->nr_cbufs * GENX(BLEND_STATE_ENTRY_length);
+
+ /* Always write at least one BLEND_STATE - the final RT message will
+ * reference BLEND_STATE[0] even if there aren't color writes. There
+ * may still be alpha testing, computed depth, and so on.
+ */
+ const int rt_dwords =
+ MAX2(cso_fb->nr_cbufs, 1) * GENX(BLEND_STATE_ENTRY_length);
+
uint32_t blend_offset;
uint32_t *blend_map =
stream_state(batch, ice->state.dynamic_uploader,
if (dirty & IRIS_DIRTY_COLOR_CALC_STATE) {
struct iris_depth_stencil_alpha_state *cso = ice->state.cso_zsa;
+#if GEN_GEN == 8
+ struct pipe_stencil_ref *p_stencil_refs = &ice->state.stencil_ref;
+#endif
uint32_t cc_offset;
void *cc_map =
stream_state(batch, ice->state.dynamic_uploader,
cc.BlendConstantColorGreen = ice->state.blend_color.color[1];
cc.BlendConstantColorBlue = ice->state.blend_color.color[2];
cc.BlendConstantColorAlpha = ice->state.blend_color.color[3];
+#if GEN_GEN == 8
+ cc.StencilReferenceValue = p_stencil_refs->ref_value[0];
+ cc.BackfaceStencilReferenceValue = p_stencil_refs->ref_value[1];
+#endif
}
iris_emit_cmd(batch, GENX(3DSTATE_CC_STATE_POINTERS), ptr) {
ptr.ColorCalcStatePointer = cc_offset;
if (dirty & IRIS_DIRTY_SAMPLE_MASK) {
iris_emit_cmd(batch, GENX(3DSTATE_SAMPLE_MASK), ms) {
- ms.SampleMask = MAX2(ice->state.sample_mask, 1);
+ ms.SampleMask = ice->state.sample_mask;
}
}
struct iris_rasterizer_state *cso_rast = ice->state.cso_rast;
struct pipe_framebuffer_state *cso_fb = &ice->state.framebuffer;
- bool reject = cso_rast->rasterizer_discard &&
- ice->state.prims_generated_query_active;
-
uint32_t dynamic_clip[GENX(3DSTATE_CLIP_length)];
iris_pack_command(GENX(3DSTATE_CLIP), &dynamic_clip, cl) {
cl.StatisticsEnable = ice->state.statistics_counters_enabled;
- cl.ClipMode = reject ? CLIPMODE_REJECT_ALL : CLIPMODE_NORMAL;
+ cl.ClipMode = cso_rast->rasterizer_discard ? CLIPMODE_REJECT_ALL
+ : CLIPMODE_NORMAL;
if (wm_prog_data->barycentric_interp_modes &
BRW_BARYCENTRIC_NONPERSPECTIVE_BITS)
cl.NonPerspectiveBarycentricEnable = true;
}
- /* XXX: FS program updates needs to flag IRIS_DIRTY_WM */
if (dirty & IRIS_DIRTY_WM) {
struct iris_rasterizer_state *cso = ice->state.cso_rast;
uint32_t dynamic_wm[GENX(3DSTATE_WM_length)];
wm.EarlyDepthStencilControl = EDSC_PREPS;
else if (wm_prog_data->has_side_effects)
wm.EarlyDepthStencilControl = EDSC_PSEXEC;
+
+ /* We could skip this bit if color writes are enabled. */
+ if (wm_prog_data->has_side_effects || wm_prog_data->uses_kill)
+ wm.ForceThreadDispatchEnable = ForceON;
}
iris_emit_merge(batch, cso->wm, dynamic_wm, ARRAY_SIZE(cso->wm));
}
if (dirty & IRIS_DIRTY_PS_BLEND) {
struct iris_blend_state *cso_blend = ice->state.cso_blend;
struct iris_depth_stencil_alpha_state *cso_zsa = ice->state.cso_zsa;
+ const struct shader_info *fs_info =
+ iris_get_shader_info(ice, MESA_SHADER_FRAGMENT);
+
uint32_t dynamic_pb[GENX(3DSTATE_PS_BLEND_length)];
iris_pack_command(GENX(3DSTATE_PS_BLEND), &dynamic_pb, pb) {
- pb.HasWriteableRT = true; // XXX: comes from somewhere :(
+ pb.HasWriteableRT = has_writeable_rt(cso_blend, fs_info);
pb.AlphaTestEnable = cso_zsa->alpha.enabled;
}
if (dirty & IRIS_DIRTY_WM_DEPTH_STENCIL) {
struct iris_depth_stencil_alpha_state *cso = ice->state.cso_zsa;
+#if GEN_GEN >= 9
struct pipe_stencil_ref *p_stencil_refs = &ice->state.stencil_ref;
-
uint32_t stencil_refs[GENX(3DSTATE_WM_DEPTH_STENCIL_length)];
iris_pack_command(GENX(3DSTATE_WM_DEPTH_STENCIL), &stencil_refs, wmds) {
wmds.StencilReferenceValue = p_stencil_refs->ref_value[0];
wmds.BackfaceStencilReferenceValue = p_stencil_refs->ref_value[1];
}
iris_emit_merge(batch, cso->wmds, stencil_refs, ARRAY_SIZE(cso->wmds));
+#else
+ iris_batch_emit(batch, cso->wmds, sizeof(cso->wmds));
+#endif
}
if (dirty & IRIS_DIRTY_SCISSOR_RECT) {
struct iris_resource *zres, *sres;
iris_get_depth_stencil_resources(cso_fb->zsbuf->texture,
&zres, &sres);
- // XXX: might not be writable...
- if (zres)
- iris_use_pinned_bo(batch, zres->bo, true);
- if (sres)
- iris_use_pinned_bo(batch, sres->bo, true);
+ if (zres) {
+ iris_use_pinned_bo(batch, zres->bo,
+ ice->state.depth_writes_enabled);
+ if (zres->aux.bo) {
+ iris_use_pinned_bo(batch, zres->aux.bo,
+ ice->state.depth_writes_enabled);
+ }
+ }
+
+ if (sres) {
+ iris_use_pinned_bo(batch, sres->bo,
+ ice->state.stencil_writes_enabled);
+ }
}
}
}
if (dirty & IRIS_DIRTY_VERTEX_BUFFERS) {
- struct iris_vertex_buffer_state *cso = &ice->state.genx->vertex_buffers;
- const unsigned vb_dwords = GENX(VERTEX_BUFFER_STATE_length);
+ int count = util_bitcount64(ice->state.bound_vertex_buffers);
- if (cso->num_buffers > 0) {
+ if (count) {
/* The VF cache designers cut corners, and made the cache key's
* <VertexBufferIndex, Memory Address> tuple only consider the bottom
* 32 bits of the address. If you have two vertex buffers which get
*/
unsigned flush_flags = 0;
- for (unsigned i = 0; i < cso->num_buffers; i++) {
+ uint64_t bound = ice->state.bound_vertex_buffers;
+ while (bound) {
+ const int i = u_bit_scan64(&bound);
uint16_t high_bits = 0;
- struct iris_resource *res = (void *) cso->resources[i];
+ struct iris_resource *res =
+ (void *) genx->vertex_buffers[i].resource;
if (res) {
iris_use_pinned_bo(batch, res->bo, false);
high_bits = res->bo->gtt_offset >> 32ull;
if (high_bits != ice->state.last_vbo_high_bits[i]) {
- flush_flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE;
+ flush_flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE |
+ PIPE_CONTROL_CS_STALL;
ice->state.last_vbo_high_bits[i] = high_bits;
}
if (flush_flags)
iris_emit_pipe_control_flush(batch, flush_flags);
- iris_batch_emit(batch, cso->vertex_buffers, sizeof(uint32_t) *
- (1 + vb_dwords * cso->num_buffers));
+ const unsigned vb_dwords = GENX(VERTEX_BUFFER_STATE_length);
+
+ uint32_t *map =
+ iris_get_command_space(batch, 4 * (1 + vb_dwords * count));
+ _iris_pack_command(batch, GENX(3DSTATE_VERTEX_BUFFERS), map, vb) {
+ vb.DWordLength = (vb_dwords * count + 1) - 2;
+ }
+ map += 1;
+
+ bound = ice->state.bound_vertex_buffers;
+ while (bound) {
+ const int i = u_bit_scan64(&bound);
+ memcpy(map, genx->vertex_buffers[i].state,
+ sizeof(uint32_t) * vb_dwords);
+ map += vb_dwords;
+ }
}
}
}
}
- // XXX: Gen8 - PMA fix
+ /* TODO: Gen8 PMA fix */
}
static void
iris_emit_cmd(batch, GENX(3DSTATE_INDEX_BUFFER), ib) {
ib.IndexFormat = draw->index_size >> 1;
- ib.MOCS = MOCS_WB;
+ ib.MOCS = mocs(bo);
ib.BufferSize = bo->size;
ib.BufferStartingAddress = ro_bo(bo, offset);
}
/* The VF cache key only uses 32-bits, see vertex buffer comment above */
uint16_t high_bits = bo->gtt_offset >> 32ull;
if (high_bits != ice->state.last_index_bo_high_bits) {
- iris_emit_pipe_control_flush(batch, PIPE_CONTROL_VF_CACHE_INVALIDATE);
+ iris_emit_pipe_control_flush(batch, PIPE_CONTROL_VF_CACHE_INVALIDATE |
+ PIPE_CONTROL_CS_STALL);
ice->state.last_index_bo_high_bits = high_bits;
}
}
lri.DataDWord = 0;
}
}
+ } else if (draw->count_from_stream_output) {
+ struct iris_stream_output_target *so =
+ (void *) draw->count_from_stream_output;
+
+ /* XXX: Replace with actual cache tracking */
+ iris_emit_pipe_control_flush(batch, PIPE_CONTROL_CS_STALL);
+
+ iris_emit_cmd(batch, GENX(MI_LOAD_REGISTER_MEM), lrm) {
+ lrm.RegisterAddress = CS_GPR(0);
+ lrm.MemoryAddress =
+ ro_bo(iris_resource_bo(so->offset.res), so->offset.offset);
+ }
+ iris_math_div32_gpr0(ice, batch, so->stride);
+ _iris_emit_lrr(batch, _3DPRIM_VERTEX_COUNT, CS_GPR(0));
+
+ _iris_emit_lri(batch, _3DPRIM_START_VERTEX, 0);
+ _iris_emit_lri(batch, _3DPRIM_BASE_VERTEX, 0);
+ _iris_emit_lri(batch, _3DPRIM_START_INSTANCE, 0);
+ _iris_emit_lri(batch, _3DPRIM_INSTANCE_COUNT, draw->instance_count);
}
iris_emit_cmd(batch, GENX(3DPRIMITIVE), prim) {
- prim.StartInstanceLocation = draw->start_instance;
- prim.InstanceCount = draw->instance_count;
- prim.VertexCountPerInstance = draw->count;
prim.VertexAccessType = draw->index_size > 0 ? RANDOM : SEQUENTIAL;
prim.PredicateEnable =
ice->state.predicate == IRIS_PREDICATE_STATE_USE_BIT;
- // XXX: this is probably bonkers.
- prim.StartVertexLocation = draw->start;
+ if (draw->indirect || draw->count_from_stream_output) {
+ prim.IndirectParameterEnable = true;
+ } else {
+ prim.StartInstanceLocation = draw->start_instance;
+ prim.InstanceCount = draw->instance_count;
+ prim.VertexCountPerInstance = draw->count;
- prim.IndirectParameterEnable = draw->indirect != NULL;
+ // XXX: this is probably bonkers.
+ prim.StartVertexLocation = draw->start;
- if (draw->index_size) {
- prim.BaseVertexLocation += draw->index_bias;
- } else {
- prim.StartVertexLocation += draw->index_bias;
- }
+ if (draw->index_size) {
+ prim.BaseVertexLocation += draw->index_bias;
+ } else {
+ prim.StartVertexLocation += draw->index_bias;
+ }
- //prim.BaseVertexLocation = ...;
+ //prim.BaseVertexLocation = ...;
+ }
}
}
struct brw_stage_prog_data *prog_data = shader->prog_data;
struct brw_cs_prog_data *cs_prog_data = (void *) prog_data;
+ /* Always pin the binder. If we're emitting new binding table pointers,
+ * we need it. If not, we're probably inheriting old tables via the
+ * context, and need it anyway. Since true zero-bindings cases are
+ * practically non-existent, just pin it and avoid last_res tracking.
+ */
+ iris_use_pinned_bo(batch, ice->state.binder.bo, false);
+
if ((dirty & IRIS_DIRTY_CONSTANTS_CS) && shs->cbuf0_needs_upload)
upload_uniforms(ice, MESA_SHADER_COMPUTE);
iris_emit_cmd(batch, GENX(MEDIA_VFE_STATE), vfe) {
if (prog_data->total_scratch) {
- uint32_t scratch_addr =
+ struct iris_bo *bo =
iris_get_scratch_space(ice, prog_data->total_scratch,
MESA_SHADER_COMPUTE);
vfe.PerThreadScratchSpace = ffs(prog_data->total_scratch) - 11;
- vfe.ScratchSpaceBasePointer = rw_bo(NULL, scratch_addr);
+ vfe.ScratchSpaceBasePointer = rw_bo(bo, 0);
}
vfe.MaximumNumberofThreads =
vfe.ResetGatewayTimer =
Resettingrelativetimerandlatchingtheglobaltimestamp;
#endif
-
+#if GEN_GEN == 8
+ vfe.BypassGatewayControl = true;
+#endif
vfe.NumberofURBEntries = 2;
vfe.URBEntryAllocationSize = 2;
- // XXX: Use Indirect Payload Storage?
vfe.CURBEAllocationSize =
ALIGN(cs_prog_data->push.per_thread.regs * cs_prog_data->threads +
cs_prog_data->push.cross_thread.regs, 2);
}
}
- // XXX: hack iris_set_constant_buffers to upload these thread counts
- // XXX: along with regular uniforms for compute shaders, somehow.
-
+ /* TODO: Combine subgroup-id with cbuf0 so we can push regular uniforms */
uint32_t curbe_data_offset = 0;
- // TODO: Move subgroup-id into uniforms ubo so we can push uniforms
assert(cs_prog_data->push.cross_thread.dwords == 0 &&
cs_prog_data->push.per_thread.dwords == 1 &&
cs_prog_data->base.param[0] == BRW_PARAM_BUILTIN_SUBGROUP_ID);
static void
iris_destroy_state(struct iris_context *ice)
{
- iris_free_vertex_buffers(&ice->state.genx->vertex_buffers);
+ struct iris_genx_state *genx = ice->state.genx;
+
+ uint64_t bound_vbs = ice->state.bound_vertex_buffers;
+ while (bound_vbs) {
+ const int i = u_bit_scan64(&bound_vbs);
+ pipe_resource_reference(&genx->vertex_buffers[i].resource, NULL);
+ }
+ free(ice->state.genx);
- // XXX: unreference resources/surfaces.
for (unsigned i = 0; i < ice->state.framebuffer.nr_cbufs; i++) {
pipe_surface_reference(&ice->state.framebuffer.cbufs[i], NULL);
}
for (int stage = 0; stage < MESA_SHADER_STAGES; stage++) {
struct iris_shader_state *shs = &ice->state.shaders[stage];
pipe_resource_reference(&shs->sampler_table.res, NULL);
+ for (int i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+ pipe_resource_reference(&shs->constbuf[i].data.res, NULL);
+ pipe_resource_reference(&shs->constbuf[i].surface_state.res, NULL);
+ }
+ for (int i = 0; i < PIPE_MAX_SHADER_IMAGES; i++) {
+ pipe_resource_reference(&shs->image[i].res, NULL);
+ pipe_resource_reference(&shs->image[i].surface_state.res, NULL);
+ }
+ for (int i = 0; i < PIPE_MAX_SHADER_BUFFERS; i++) {
+ pipe_resource_reference(&shs->ssbo[i], NULL);
+ pipe_resource_reference(&shs->ssbo_surface_state[i].res, NULL);
+ }
+ for (int i = 0; i < IRIS_MAX_TEXTURE_SAMPLERS; i++) {
+ pipe_sampler_view_reference((struct pipe_sampler_view **)
+ &shs->textures[i], NULL);
+ }
}
- free(ice->state.genx);
+ pipe_resource_reference(&ice->state.grid_size.res, NULL);
+ pipe_resource_reference(&ice->state.grid_surf_state.res, NULL);
+
+ pipe_resource_reference(&ice->state.null_fb.res, NULL);
pipe_resource_reference(&ice->state.unbound_tex.res, NULL);
pipe_resource_reference(&ice->state.last_res.cc_vp, NULL);
/* ------------------------------------------------------------------- */
static void
-iris_load_register_reg32(struct iris_batch *batch, uint32_t src,
- uint32_t dst)
+iris_load_register_reg32(struct iris_batch *batch, uint32_t dst,
+ uint32_t src)
{
- _iris_emit_lrr(batch, src, dst);
+ _iris_emit_lrr(batch, dst, src);
}
static void
-iris_load_register_reg64(struct iris_batch *batch, uint32_t src,
- uint32_t dst)
+iris_load_register_reg64(struct iris_batch *batch, uint32_t dst,
+ uint32_t src)
{
- _iris_emit_lrr(batch, src, dst);
- _iris_emit_lrr(batch, src + 4, dst + 4);
+ _iris_emit_lrr(batch, dst, src);
+ _iris_emit_lrr(batch, dst + 4, src + 4);
}
static void
projects / mesa.git / blobdiff