* IN THE SOFTWARE.
***************************************************************************/
+#include <llvm/Config/llvm-config.h>
+
+#if LLVM_VERSION_MAJOR < 7
// llvm redefines DEBUG
#pragma push_macro("DEBUG")
#undef DEBUG
+#endif
+
+#include <rasterizer/core/state.h>
#include "JitManager.h"
+
+#if LLVM_VERSION_MAJOR < 7
#pragma pop_macro("DEBUG")
+#endif
#include "common/os.h"
#include "jit_api.h"
#include "core/state_funcs.h"
#include "gallivm/lp_bld_tgsi.h"
-#include "util/u_format.h"
+#include "util/format/u_format.h"
#include "util/u_memory.h"
#include "util/u_inlines.h"
#include "swr_state.h"
#include "swr_context.h"
+#include "gen_surf_state_llvm.h"
#include "gen_swr_context_llvm.h"
#include "swr_screen.h"
#include "swr_resource.h"
const struct pipe_blend_state *blend)
{
struct swr_blend_state *state = CALLOC_STRUCT(swr_blend_state);
+ assert(state != nullptr);
memcpy(&state->pipe, blend, sizeof(*blend));
/* set the new sampler views */
ctx->num_sampler_views[shader] = num;
for (i = 0; i < num; i++) {
- /* Note: we're using pipe_sampler_view_release() here to work around
- * a possible crash when the old view belongs to another context that
- * was already destroyed.
- */
- pipe_sampler_view_release(pipe, &ctx->sampler_views[shader][start + i]);
pipe_sampler_view_reference(&ctx->sampler_views[shader][start + i],
views[i]);
}
// soState.streamToRasterizer not used
for (uint32_t i = 0; i < stream_output->num_outputs; i++) {
+ unsigned attrib_slot = stream_output->output[i].register_index;
+ attrib_slot = swr_so_adjust_attrib(attrib_slot, swr_vs);
swr_vs->soState.streamMasks[stream_output->output[i].stream] |=
- 1 << (stream_output->output[i].register_index - 1);
+ (1 << attrib_slot);
}
for (uint32_t i = 0; i < MAX_SO_STREAMS; i++) {
swr_vs->soState.streamNumEntries[i] =
return NULL;
swr_gs->pipe.tokens = tgsi_dup_tokens(gs->tokens);
-
lp_build_tgsi_info(gs->tokens, &swr_gs->info);
-
return swr_gs;
}
-
static void
swr_bind_gs_state(struct pipe_context *pipe, void *gs)
{
swr_fence_work_delete_gs(screen->flush_fence, swr_gs);
}
+static void *
+swr_create_tcs_state(struct pipe_context *pipe,
+ const struct pipe_shader_state *tcs)
+{
+ struct swr_tess_control_shader *swr_tcs = new swr_tess_control_shader;
+ if (!swr_tcs)
+ return NULL;
+
+ swr_tcs->pipe.tokens = tgsi_dup_tokens(tcs->tokens);
+ lp_build_tgsi_info(tcs->tokens, &swr_tcs->info);
+ return swr_tcs;
+}
+
+static void
+swr_bind_tcs_state(struct pipe_context *pipe, void *tcs)
+{
+ struct swr_context *ctx = swr_context(pipe);
+
+ if (ctx->tcs == tcs)
+ return;
+
+ ctx->tcs = (swr_tess_control_shader *)tcs;
+ ctx->dirty |= SWR_NEW_TCS;
+ ctx->dirty |= SWR_NEW_TS;
+}
+
+static void
+swr_delete_tcs_state(struct pipe_context *pipe, void *tcs)
+{
+ struct swr_tess_control_shader *swr_tcs = (swr_tess_control_shader *)tcs;
+ FREE((void *)swr_tcs->pipe.tokens);
+ struct swr_screen *screen = swr_screen(pipe->screen);
+
+ /* Defer deleton of tcs state */
+ swr_fence_work_delete_tcs(screen->flush_fence, swr_tcs);
+}
+
+static void *
+swr_create_tes_state(struct pipe_context *pipe,
+ const struct pipe_shader_state *tes)
+{
+ struct swr_tess_evaluation_shader *swr_tes = new swr_tess_evaluation_shader;
+ if (!swr_tes)
+ return NULL;
+
+ swr_tes->pipe.tokens = tgsi_dup_tokens(tes->tokens);
+ lp_build_tgsi_info(tes->tokens, &swr_tes->info);
+ return swr_tes;
+}
+
+static void
+swr_bind_tes_state(struct pipe_context *pipe, void *tes)
+{
+ struct swr_context *ctx = swr_context(pipe);
+
+ if (ctx->tes == tes)
+ return;
+
+ // Save current tessellator state first
+ if (ctx->tes != nullptr) {
+ ctx->tes->ts_state = ctx->tsState;
+ }
+
+ ctx->tes = (swr_tess_evaluation_shader *)tes;
+
+ ctx->dirty |= SWR_NEW_TES;
+ ctx->dirty |= SWR_NEW_TS;
+}
+
+static void
+swr_delete_tes_state(struct pipe_context *pipe, void *tes)
+{
+ struct swr_tess_evaluation_shader *swr_tes = (swr_tess_evaluation_shader *)tes;
+ FREE((void *)swr_tes->pipe.tokens);
+ struct swr_screen *screen = swr_screen(pipe->screen);
+
+ /* Defer deleton of tes state */
+ swr_fence_work_delete_tes(screen->flush_fence, swr_tes);
+}
+
static void
swr_set_constant_buffer(struct pipe_context *pipe,
enum pipe_shader_type shader,
ctx->dirty |= SWR_NEW_FSCONSTANTS;
} else if (shader == PIPE_SHADER_GEOMETRY) {
ctx->dirty |= SWR_NEW_GSCONSTANTS;
+ } else if (shader == PIPE_SHADER_TESS_CTRL) {
+ ctx->dirty |= SWR_NEW_TCSCONSTANTS;
+ } else if (shader == PIPE_SHADER_TESS_EVAL) {
+ ctx->dirty |= SWR_NEW_TESCONSTANTS;
}
-
if (cb && cb->user_buffer) {
pipe_resource_reference(&constants, NULL);
}
assert(num_elements <= PIPE_MAX_ATTRIBS);
velems = new swr_vertex_element_state;
if (velems) {
- memset(&velems->fsState, 0, sizeof(velems->fsState));
+ memset((void*)&velems->fsState, 0, sizeof(velems->fsState));
velems->fsState.bVertexIDOffsetEnable = true;
velems->fsState.numAttribs = num_elements;
for (unsigned i = 0; i < num_elements; i++) {
? ComponentControl::StoreSrc
: ComponentControl::Store1Fp;
velems->fsState.layout[i].ComponentPacking = ComponentEnable::XYZW;
- velems->fsState.layout[i].InstanceDataStepRate =
+ velems->fsState.layout[i].InstanceAdvancementState =
attribs[i].instance_divisor;
/* Calculate the pitch of each stream */
static void
swr_set_scissor_states(struct pipe_context *pipe,
unsigned start_slot,
- unsigned num_viewports,
- const struct pipe_scissor_state *scissor)
+ unsigned num_scissors,
+ const struct pipe_scissor_state *scissors)
{
struct swr_context *ctx = swr_context(pipe);
- ctx->scissor = *scissor;
- ctx->swr_scissor.xmin = scissor->minx;
- ctx->swr_scissor.xmax = scissor->maxx;
- ctx->swr_scissor.ymin = scissor->miny;
- ctx->swr_scissor.ymax = scissor->maxy;
+ memcpy(ctx->scissors + start_slot, scissors,
+ sizeof(struct pipe_scissor_state) * num_scissors);
+
+ for (unsigned i = 0; i < num_scissors; i++) {
+ auto idx = start_slot + i;
+ ctx->swr_scissors[idx].xmin = scissors[idx].minx;
+ ctx->swr_scissors[idx].xmax = scissors[idx].maxx;
+ ctx->swr_scissors[idx].ymin = scissors[idx].miny;
+ ctx->swr_scissors[idx].ymax = scissors[idx].maxy;
+ }
ctx->dirty |= SWR_NEW_SCISSOR;
}
{
struct swr_context *ctx = swr_context(pipe);
- ctx->viewport = *vpt;
+ memcpy(ctx->viewports + start_slot, vpt, sizeof(struct pipe_viewport_state) * num_viewports);
ctx->dirty |= SWR_NEW_VIEWPORT;
}
{
struct swr_context *ctx = swr_context(pipe);
- boolean changed = !util_framebuffer_state_equal(&ctx->framebuffer, fb);
+ bool changed = !util_framebuffer_state_equal(&ctx->framebuffer, fb);
assert(fb->width <= KNOB_GUARDBAND_WIDTH);
assert(fb->height <= KNOB_GUARDBAND_HEIGHT);
/* VBO vertex buffers */
for (uint32_t i = 0; i < ctx->num_vertex_buffers; i++) {
struct pipe_vertex_buffer *vb = &ctx->vertex_buffer[i];
- if (!vb->is_user_buffer)
+ if (!vb->is_user_buffer && vb->buffer.resource)
swr_resource_read(vb->buffer.resource);
}
jit_tex->width = res->width0;
jit_tex->height = res->height0;
- jit_tex->base_ptr = swr->pBaseAddress;
+ jit_tex->base_ptr = (uint8_t*)swr->xpBaseAddress;
+ jit_tex->num_samples = swr->numSamples;
+ jit_tex->sample_stride = 0;
if (view->target != PIPE_BUFFER) {
jit_tex->first_level = view->u.tex.first_level;
jit_tex->last_level = view->u.tex.last_level;
num_constants = pDC->num_constantsGS;
scratch = &ctx->scratch->gs_constants;
break;
+ case PIPE_SHADER_TESS_CTRL:
+ constant = pDC->constantTCS;
+ num_constants = pDC->num_constantsTCS;
+ scratch = &ctx->scratch->tcs_constants;
+ break;
+ case PIPE_SHADER_TESS_EVAL:
+ constant = pDC->constantTES;
+ num_constants = pDC->num_constantsTES;
+ scratch = &ctx->scratch->tes_constants;
+ break;
default:
- debug_printf("Unsupported shader type constants\n");
+ assert(0 && "Unsupported shader type constants");
return;
}
struct SWR_SURFACE_STATE *rt = &pDC->renderTargets[attachment];
/* Do nothing if the render target hasn't changed */
- if ((!sf || !sf->texture) && rt->pBaseAddress == nullptr)
+ if ((!sf || !sf->texture) && (void*)(rt->xpBaseAddress) == nullptr)
return false;
/* Deal with disabling RT up front */
const SWR_SURFACE_STATE *swr_surface = &swr->swr;
SWR_FORMAT fmt = mesa_to_swr_format(sf->format);
- if (attachment == SWR_ATTACHMENT_STENCIL && swr->secondary.pBaseAddress) {
+ if (attachment == SWR_ATTACHMENT_STENCIL && swr->secondary.xpBaseAddress) {
swr_surface = &swr->secondary;
fmt = swr_surface->format;
}
- if (rt->pBaseAddress == swr_surface->pBaseAddress &&
+ if (rt->xpBaseAddress == swr_surface->xpBaseAddress &&
rt->format == fmt &&
rt->lod == sf->u.tex.level &&
rt->arrayIndex == sf->u.tex.first_layer)
bool need_fence = false;
/* StoreTile for changed target */
- if (rt->pBaseAddress) {
+ if (rt->xpBaseAddress) {
/* If changing attachment to a new target, mark tiles as
* INVALID so they are reloaded from surface. */
swr_store_render_target(&ctx->pipe, attachment, SWR_TILE_INVALID);
need_fence = true;
+ } else {
+ /* if no previous attachment, invalidate tiles that may be marked
+ * RESOLVED because of an old attachment */
+ swr_invalidate_render_target(&ctx->pipe, attachment, sf->width, sf->height);
+ /* no need to set fence here */
}
/* Make new attachment */
return need_fence;
}
+/*
+ * for cases where resources are shared between contexts, invalidate
+ * this ctx's resource. so it can be fetched fresh. Old ctx's resource
+ * is already stored during a flush
+ */
+static inline void
+swr_invalidate_buffers_after_ctx_change(struct pipe_context *pipe)
+{
+ struct swr_context *ctx = swr_context(pipe);
+
+ for (uint32_t i = 0; i < ctx->framebuffer.nr_cbufs; i++) {
+ struct pipe_surface *cb = ctx->framebuffer.cbufs[i];
+ if (cb) {
+ struct swr_resource *res = swr_resource(cb->texture);
+ if (res->curr_pipe != pipe) {
+ /* if curr_pipe is NULL (first use), status should not be WRITE */
+ assert(res->curr_pipe || !(res->status & SWR_RESOURCE_WRITE));
+ if (res->status & SWR_RESOURCE_WRITE) {
+ swr_invalidate_render_target(pipe, i, cb->width, cb->height);
+ }
+ }
+ res->curr_pipe = pipe;
+ }
+ }
+ if (ctx->framebuffer.zsbuf) {
+ struct pipe_surface *zb = ctx->framebuffer.zsbuf;
+ if (zb) {
+ struct swr_resource *res = swr_resource(zb->texture);
+ if (res->curr_pipe != pipe) {
+ /* if curr_pipe is NULL (first use), status should not be WRITE */
+ assert(res->curr_pipe || !(res->status & SWR_RESOURCE_WRITE));
+ if (res->status & SWR_RESOURCE_WRITE) {
+ swr_invalidate_render_target(pipe, SWR_ATTACHMENT_DEPTH, zb->width, zb->height);
+ swr_invalidate_render_target(pipe, SWR_ATTACHMENT_STENCIL, zb->width, zb->height);
+ }
+ }
+ res->curr_pipe = pipe;
+ }
+ }
+}
+
static inline void
swr_user_vbuf_range(const struct pipe_draw_info *info,
const struct swr_vertex_element_state *velems,
{
/* FIXME: The size is too large - we don't access the full extra stride. */
unsigned elems;
+ unsigned elem_pitch = vb->stride + velems->stream_pitch[i];
if (velems->instanced_bufs & (1U << i)) {
elems = info->instance_count / velems->min_instance_div[i] + 1;
*totelems = info->start_instance + elems;
*base = info->start_instance * vb->stride;
- *size = elems * vb->stride;
+ *size = elems * elem_pitch;
} else if (vb->stride) {
elems = info->max_index - info->min_index + 1;
- *totelems = info->max_index + 1;
- *base = info->min_index * vb->stride;
- *size = elems * vb->stride;
+ *totelems = (info->max_index + info->index_bias) + 1;
+ *base = (info->min_index + info->index_bias) * vb->stride;
+ *size = elems * elem_pitch;
} else {
*totelems = 1;
*base = 0;
sizeof(ctx->poly_stipple.pipe.stipple));
}
+
+static struct tgsi_shader_info *
+swr_get_last_fe(const struct swr_context *ctx)
+{
+ tgsi_shader_info *pLastFE = &ctx->vs->info.base;
+
+ if (ctx->gs) {
+ pLastFE = &ctx->gs->info.base;
+ }
+ else if (ctx->tes) {
+ pLastFE = &ctx->tes->info.base;
+ }
+ else if (ctx->tcs) {
+ pLastFE = &ctx->tcs->info.base;
+ }
+ return pLastFE;
+}
+
+
void
swr_update_derived(struct pipe_context *pipe,
const struct pipe_draw_info *p_draw_info)
}
/* Update screen->pipe to current pipe context. */
- if (screen->pipe != pipe)
- screen->pipe = pipe;
+ screen->pipe = pipe;
/* Any state that requires dirty flags to be re-triggered sets this mask */
/* For example, user_buffer vertex and index buffers. */
unsigned post_update_dirty_flags = 0;
+ /* bring resources that changed context up-to-date */
+ swr_invalidate_buffers_after_ctx_change(pipe);
+
/* Render Targets */
if (ctx->dirty & SWR_NEW_FRAMEBUFFER) {
struct pipe_framebuffer_state *fb = &ctx->framebuffer;
/* Raster state */
if (ctx->dirty & (SWR_NEW_RASTERIZER |
SWR_NEW_VS | // clipping
+ SWR_NEW_TES |
+ SWR_NEW_TCS |
SWR_NEW_FRAMEBUFFER)) {
pipe_rasterizer_state *rasterizer = ctx->rasterizer;
pipe_framebuffer_state *fb = &ctx->framebuffer;
if (zb && swr_resource(zb->texture)->has_depth)
rastState->depthFormat = swr_resource(zb->texture)->swr.format;
- rastState->depthClipEnable = rasterizer->depth_clip;
+ rastState->depthClipEnable = rasterizer->depth_clip_near;
+ rastState->clipEnable = rasterizer->depth_clip_near | rasterizer->depth_clip_far;
rastState->clipHalfZ = rasterizer->clip_halfz;
- rastState->clipDistanceMask =
- ctx->vs->info.base.num_written_clipdistance ?
- ctx->vs->info.base.clipdist_writemask & rasterizer->clip_plane_enable :
- rasterizer->clip_plane_enable;
-
- rastState->cullDistanceMask =
- ctx->vs->info.base.culldist_writemask << ctx->vs->info.base.num_written_clipdistance;
-
- SwrSetRastState(ctx->swrContext, rastState);
- }
-
- /* Scissor */
- if (ctx->dirty & SWR_NEW_SCISSOR) {
- SwrSetScissorRects(ctx->swrContext, 1, &ctx->swr_scissor);
+ ctx->api.pfnSwrSetRastState(ctx->swrContext, rastState);
}
/* Viewport */
if (ctx->dirty & (SWR_NEW_VIEWPORT | SWR_NEW_FRAMEBUFFER
| SWR_NEW_RASTERIZER)) {
- pipe_viewport_state *state = &ctx->viewport;
+ pipe_viewport_state *state = &ctx->viewports[0];
pipe_framebuffer_state *fb = &ctx->framebuffer;
pipe_rasterizer_state *rasterizer = ctx->rasterizer;
- SWR_VIEWPORT *vp = &ctx->derived.vp;
+ SWR_VIEWPORT *vp = &ctx->derived.vp[0];
SWR_VIEWPORT_MATRICES *vpm = &ctx->derived.vpm;
- vp->x = state->translate[0] - state->scale[0];
- vp->width = 2 * state->scale[0];
- vp->y = state->translate[1] - fabs(state->scale[1]);
- vp->height = 2 * fabs(state->scale[1]);
- util_viewport_zmin_zmax(state, rasterizer->clip_halfz,
- &vp->minZ, &vp->maxZ);
-
- vpm->m00[0] = state->scale[0];
- vpm->m11[0] = state->scale[1];
- vpm->m22[0] = state->scale[2];
- vpm->m30[0] = state->translate[0];
- vpm->m31[0] = state->translate[1];
- vpm->m32[0] = state->translate[2];
-
- /* Now that the matrix is calculated, clip the view coords to screen
- * size. OpenGL allows for -ve x,y in the viewport. */
- if (vp->x < 0.0f) {
- vp->width += vp->x;
- vp->x = 0.0f;
- }
- if (vp->y < 0.0f) {
- vp->height += vp->y;
- vp->y = 0.0f;
- }
- vp->width = std::min(vp->width, (float)fb->width - vp->x);
- vp->height = std::min(vp->height, (float)fb->height - vp->y);
+ for (unsigned i = 0; i < KNOB_NUM_VIEWPORTS_SCISSORS; i++) {
+ vp->x = state->translate[0] - state->scale[0];
+ vp->width = 2 * state->scale[0];
+ vp->y = state->translate[1] - fabs(state->scale[1]);
+ vp->height = 2 * fabs(state->scale[1]);
+ util_viewport_zmin_zmax(state, rasterizer->clip_halfz,
+ &vp->minZ, &vp->maxZ);
- SwrSetViewports(ctx->swrContext, 1, vp, vpm);
- }
+ if (rasterizer->depth_clip_near) {
+ vp->minZ = 0.0f;
+ }
- /* Set vertex & index buffers */
- /* (using draw info if called by swr_draw_vbo) */
- /* TODO: This is always true, because the index buffer comes from
- * pipe_draw_info.
- */
- if (1 || ctx->dirty & SWR_NEW_VERTEX) {
- uint32_t scratch_total;
- uint8_t *scratch = NULL;
+ if (rasterizer->depth_clip_far) {
+ vp->maxZ = 1.0f;
+ }
- /* If being called by swr_draw_vbo, copy draw details */
- struct pipe_draw_info info = {0};
- if (p_draw_info)
- info = *p_draw_info;
+ vpm->m00[i] = state->scale[0];
+ vpm->m11[i] = state->scale[1];
+ vpm->m22[i] = state->scale[2];
+ vpm->m30[i] = state->translate[0];
+ vpm->m31[i] = state->translate[1];
+ vpm->m32[i] = state->translate[2];
+
+ /* Now that the matrix is calculated, clip the view coords to screen
+ * size. OpenGL allows for -ve x,y in the viewport. */
+ if (vp->x < 0.0f) {
+ vp->width += vp->x;
+ vp->x = 0.0f;
+ }
+ if (vp->y < 0.0f) {
+ vp->height += vp->y;
+ vp->y = 0.0f;
+ }
+ vp->width = std::min(vp->width, (float) fb->width - vp->x);
+ vp->height = std::min(vp->height, (float) fb->height - vp->y);
- /* We must get all the scratch space in one go */
- scratch_total = 0;
- for (UINT i = 0; i < ctx->num_vertex_buffers; i++) {
- struct pipe_vertex_buffer *vb = &ctx->vertex_buffer[i];
+ vp++;
+ state++;
+ }
+ ctx->api.pfnSwrSetViewports(ctx->swrContext, KNOB_NUM_VIEWPORTS_SCISSORS,
+ &ctx->derived.vp[0], &ctx->derived.vpm);
+ }
- if (!vb->is_user_buffer)
- continue;
+ /* When called from swr_clear (p_draw_info = null), render targets,
+ * rasterState and viewports (dependent on render targets) are the only
+ * necessary validation. Defer remaining validation by setting
+ * post_update_dirty_flags and clear all dirty flags. BackendState is
+ * still unconditionally validated below */
+ if (!p_draw_info) {
+ post_update_dirty_flags = ctx->dirty & ~(SWR_NEW_FRAMEBUFFER |
+ SWR_NEW_RASTERIZER |
+ SWR_NEW_VIEWPORT);
+ ctx->dirty = 0;
+ }
- uint32_t elems, base, size;
- swr_user_vbuf_range(&info, ctx->velems, vb, i, &elems, &base, &size);
- scratch_total += AlignUp(size, 4);
- }
+ /* Scissor */
+ if (ctx->dirty & SWR_NEW_SCISSOR) {
+ ctx->api.pfnSwrSetScissorRects(ctx->swrContext, KNOB_NUM_VIEWPORTS_SCISSORS, ctx->swr_scissors);
+ }
- if (scratch_total) {
- scratch = (uint8_t *)swr_copy_to_scratch_space(
- ctx, &ctx->scratch->vertex_buffer, NULL, scratch_total);
- }
+ /* Set vertex & index buffers */
+ if (ctx->dirty & SWR_NEW_VERTEX) {
+ const struct pipe_draw_info &info = *p_draw_info;
/* vertex buffers */
SWR_VERTEX_BUFFER_STATE swrVertexBuffers[PIPE_MAX_ATTRIBS];
for (UINT i = 0; i < ctx->num_vertex_buffers; i++) {
- uint32_t size, pitch, elems, partial_inbounds;
- uint32_t min_vertex_index;
+ uint32_t size = 0, pitch = 0, elems = 0, partial_inbounds = 0;
+ uint32_t min_vertex_index = 0;
const uint8_t *p_data;
struct pipe_vertex_buffer *vb = &ctx->vertex_buffer[i];
pitch = vb->stride;
- if (!vb->is_user_buffer) {
+ if (vb->is_user_buffer) {
+ /* Client buffer
+ * client memory is one-time use, re-trigger SWR_NEW_VERTEX to
+ * revalidate on each draw */
+ post_update_dirty_flags |= SWR_NEW_VERTEX;
+
+ uint32_t base;
+ swr_user_vbuf_range(&info, ctx->velems, vb, i, &elems, &base, &size);
+ partial_inbounds = 0;
+ min_vertex_index = info.min_index + info.index_bias;
+
+ size = AlignUp(size, 4);
+ /* If size of client memory copy is too large, don't copy. The
+ * draw will access user-buffer directly and then block. This is
+ * faster than queuing many large client draws. */
+ if (size >= screen->client_copy_limit) {
+ post_update_dirty_flags |= SWR_BLOCK_CLIENT_DRAW;
+ p_data = (const uint8_t *) vb->buffer.user;
+ } else {
+ /* Copy only needed vertices to scratch space */
+ const void *ptr = (const uint8_t *) vb->buffer.user + base;
+ ptr = (uint8_t *)swr_copy_to_scratch_space(
+ ctx, &ctx->scratch->vertex_buffer, ptr, size);
+ p_data = (const uint8_t *)ptr - base;
+ }
+ } else if (vb->buffer.resource) {
/* VBO */
if (!pitch) {
/* If pitch=0 (ie vb->stride), buffer contains a single
}
p_data = swr_resource_data(vb->buffer.resource) + vb->buffer_offset;
- } else {
- /* Client buffer
- * client memory is one-time use, re-trigger SWR_NEW_VERTEX to
- * revalidate on each draw */
- post_update_dirty_flags |= SWR_NEW_VERTEX;
-
- uint32_t base;
- swr_user_vbuf_range(&info, ctx->velems, vb, i, &elems, &base, &size);
- partial_inbounds = 0;
- min_vertex_index = info.min_index;
-
- /* Copy only needed vertices to scratch space */
- size = AlignUp(size, 4);
- const void *ptr = (const uint8_t *) vb->buffer.user + base;
- memcpy(scratch, ptr, size);
- ptr = scratch;
- scratch += size;
- p_data = (const uint8_t *)ptr - base;
- }
+ } else
+ p_data = NULL;
swrVertexBuffers[i] = {0};
swrVertexBuffers[i].index = i;
swrVertexBuffers[i].pitch = pitch;
- swrVertexBuffers[i].pData = p_data;
+ swrVertexBuffers[i].xpData = (gfxptr_t) p_data;
swrVertexBuffers[i].size = size;
swrVertexBuffers[i].minVertex = min_vertex_index;
swrVertexBuffers[i].maxVertex = elems;
swrVertexBuffers[i].partialInboundsSize = partial_inbounds;
}
- SwrSetVertexBuffers(
+ ctx->api.pfnSwrSetVertexBuffers(
ctx->swrContext, ctx->num_vertex_buffers, swrVertexBuffers);
/* index buffer, if required (info passed in by swr_draw_vbo) */
const uint8_t *p_data;
uint32_t size, pitch;
- pitch = p_draw_info->index_size ? p_draw_info->index_size : sizeof(uint32_t);
+ pitch = info.index_size ? info.index_size : sizeof(uint32_t);
index_type = swr_convert_index_type(pitch);
if (!info.has_user_indices) {
post_update_dirty_flags |= SWR_NEW_VERTEX;
size = info.count * pitch;
- size = AlignUp(size, 4);
- /* Copy indices to scratch space */
- const void *ptr = info.index.user;
- ptr = swr_copy_to_scratch_space(
- ctx, &ctx->scratch->index_buffer, ptr, size);
- p_data = (const uint8_t *)ptr;
+ size = AlignUp(size, 4);
+ /* If size of client memory copy is too large, don't copy. The
+ * draw will access user-buffer directly and then block. This is
+ * faster than queuing many large client draws. */
+ if (size >= screen->client_copy_limit) {
+ post_update_dirty_flags |= SWR_BLOCK_CLIENT_DRAW;
+ p_data = (const uint8_t *) info.index.user;
+ } else {
+ /* Copy indices to scratch space */
+ const void *ptr = info.index.user;
+ ptr = swr_copy_to_scratch_space(
+ ctx, &ctx->scratch->index_buffer, ptr, size);
+ p_data = (const uint8_t *)ptr;
+ }
}
SWR_INDEX_BUFFER_STATE swrIndexBuffer;
- swrIndexBuffer.format = swr_convert_index_type(p_draw_info->index_size);
- swrIndexBuffer.pIndices = p_data;
+ swrIndexBuffer.format = swr_convert_index_type(info.index_size);
+ swrIndexBuffer.xpIndices = (gfxptr_t) p_data;
swrIndexBuffer.size = size;
- SwrSetIndexBuffer(ctx->swrContext, &swrIndexBuffer);
+ ctx->api.pfnSwrSetIndexBuffer(ctx->swrContext, &swrIndexBuffer);
}
struct swr_vertex_element_state *velems = ctx->velems;
/* GeometryShader */
if (ctx->dirty & (SWR_NEW_GS |
SWR_NEW_VS |
+ SWR_NEW_TCS |
+ SWR_NEW_TES |
SWR_NEW_SAMPLER |
SWR_NEW_SAMPLER_VIEW)) {
if (ctx->gs) {
} else {
func = swr_compile_gs(ctx, key);
}
- SwrSetGsFunc(ctx->swrContext, func);
+ ctx->api.pfnSwrSetGsFunc(ctx->swrContext, func);
/* JIT sampler state */
if (ctx->dirty & SWR_NEW_SAMPLER) {
ctx->swrDC.texturesGS);
}
- SwrSetGsState(ctx->swrContext, &ctx->gs->gsState);
+ ctx->api.pfnSwrSetGsState(ctx->swrContext, &ctx->gs->gsState);
} else {
SWR_GS_STATE state = { 0 };
- SwrSetGsState(ctx->swrContext, &state);
- SwrSetGsFunc(ctx->swrContext, NULL);
+ ctx->api.pfnSwrSetGsState(ctx->swrContext, &state);
+ ctx->api.pfnSwrSetGsFunc(ctx->swrContext, NULL);
}
}
- /* VertexShader */
- if (ctx->dirty & (SWR_NEW_VS |
- SWR_NEW_RASTERIZER | // for clip planes
+ // We may need to restore tessellation state
+ // This restored state may be however overwritten
+ // during shader compilation
+ if (ctx->dirty & SWR_NEW_TS) {
+ if (ctx->tes != nullptr) {
+ ctx->tsState = ctx->tes->ts_state;
+ ctx->api.pfnSwrSetTsState(ctx->swrContext, &ctx->tsState);
+ } else {
+ SWR_TS_STATE state = { 0 };
+ ctx->api.pfnSwrSetTsState(ctx->swrContext, &state);
+ }
+ }
+
+ // Tessellation Evaluation Shader
+ // Compile TES first, because TCS is optional
+ if (ctx->dirty & (SWR_NEW_GS |
+ SWR_NEW_VS |
+ SWR_NEW_TCS |
+ SWR_NEW_TES |
SWR_NEW_SAMPLER |
- SWR_NEW_SAMPLER_VIEW |
- SWR_NEW_FRAMEBUFFER)) {
+ SWR_NEW_SAMPLER_VIEW)) {
+ if (ctx->tes) {
+ swr_jit_tes_key key;
+ swr_generate_tes_key(key, ctx, ctx->tes);
+
+ auto search = ctx->tes->map.find(key);
+ PFN_TES_FUNC func;
+ if (search != ctx->tes->map.end()) {
+ func = search->second->shader;
+ } else {
+ func = swr_compile_tes(ctx, key);
+ }
+
+ ctx->api.pfnSwrSetDsFunc(ctx->swrContext, func);
+
+ /* JIT sampler state */
+ if (ctx->dirty & SWR_NEW_SAMPLER) {
+ swr_update_sampler_state(ctx,
+ PIPE_SHADER_TESS_EVAL,
+ key.nr_samplers,
+ ctx->swrDC.samplersTES);
+ }
+
+ /* JIT sampler view state */
+ if (ctx->dirty & (SWR_NEW_SAMPLER_VIEW | SWR_NEW_FRAMEBUFFER)) {
+ swr_update_texture_state(ctx,
+ PIPE_SHADER_TESS_EVAL,
+ key.nr_sampler_views,
+ ctx->swrDC.texturesTES);
+ }
+
+ // Update tessellation state in case it's been updated
+ ctx->api.pfnSwrSetTsState(ctx->swrContext, &ctx->tsState);
+ } else {
+ ctx->api.pfnSwrSetDsFunc(ctx->swrContext, NULL);
+ }
+ }
+
+ /* Tessellation Control Shader */
+ if (ctx->dirty & (SWR_NEW_GS |
+ SWR_NEW_VS |
+ SWR_NEW_TCS |
+ SWR_NEW_TES |
+ SWR_NEW_SAMPLER |
+ SWR_NEW_SAMPLER_VIEW)) {
+ if (ctx->tcs) {
+ ctx->tcs->vertices_per_patch = p_draw_info->vertices_per_patch;
+
+ swr_jit_tcs_key key;
+ swr_generate_tcs_key(key, ctx, ctx->tcs);
+
+ auto search = ctx->tcs->map.find(key);
+ PFN_TCS_FUNC func;
+ if (search != ctx->tcs->map.end()) {
+ func = search->second->shader;
+ } else {
+ func = swr_compile_tcs(ctx, key);
+ }
+
+ ctx->api.pfnSwrSetHsFunc(ctx->swrContext, func);
+
+ /* JIT sampler state */
+ if (ctx->dirty & SWR_NEW_SAMPLER) {
+ swr_update_sampler_state(ctx,
+ PIPE_SHADER_TESS_CTRL,
+ key.nr_samplers,
+ ctx->swrDC.samplersTCS);
+ }
+
+ /* JIT sampler view state */
+ if (ctx->dirty & (SWR_NEW_SAMPLER_VIEW | SWR_NEW_FRAMEBUFFER)) {
+ swr_update_texture_state(ctx,
+ PIPE_SHADER_TESS_CTRL,
+ key.nr_sampler_views,
+ ctx->swrDC.texturesTCS);
+ }
+
+ // Update tessellation state in case it's been updated
+ ctx->api.pfnSwrSetTsState(ctx->swrContext, &ctx->tsState);
+ } else {
+ ctx->api.pfnSwrSetHsFunc(ctx->swrContext, NULL);
+ }
+ }
+
+ /* VertexShader */
+ if (ctx->dirty
+ & (SWR_NEW_VS | SWR_NEW_RASTERIZER | // for clip planes
+ SWR_NEW_SAMPLER | SWR_NEW_SAMPLER_VIEW | SWR_NEW_FRAMEBUFFER)) {
swr_jit_vs_key key;
swr_generate_vs_key(key, ctx, ctx->vs);
auto search = ctx->vs->map.find(key);
} else {
func = swr_compile_vs(ctx, key);
}
- SwrSetVertexFunc(ctx->swrContext, func);
+ ctx->api.pfnSwrSetVertexFunc(ctx->swrContext, func);
/* JIT sampler state */
if (ctx->dirty & SWR_NEW_SAMPLER) {
- swr_update_sampler_state(ctx,
- PIPE_SHADER_VERTEX,
- key.nr_samplers,
- ctx->swrDC.samplersVS);
+ swr_update_sampler_state(
+ ctx, PIPE_SHADER_VERTEX, key.nr_samplers, ctx->swrDC.samplersVS);
}
/* JIT sampler view state */
if (ctx->dirty & (SWR_NEW_FS |
SWR_NEW_VS |
SWR_NEW_GS |
+ SWR_NEW_TES |
+ SWR_NEW_TCS |
SWR_NEW_RASTERIZER |
SWR_NEW_SAMPLER |
SWR_NEW_SAMPLER_VIEW |
psState.writesODepth = ctx->fs->info.base.writes_z;
psState.usesSourceDepth = ctx->fs->info.base.reads_z;
psState.shadingRate = SWR_SHADING_RATE_PIXEL;
- psState.numRenderTargets = ctx->framebuffer.nr_cbufs;
+ psState.renderTargetMask = (1 << ctx->framebuffer.nr_cbufs) - 1;
psState.posOffset = SWR_PS_POSITION_SAMPLE_NONE;
uint32_t barycentricsMask = 0;
#if 0
psState.barycentricsMask = barycentricsMask;
psState.usesUAV = false; // XXX
psState.forceEarlyZ = false;
- SwrSetPixelShaderState(ctx->swrContext, &psState);
+ ctx->api.pfnSwrSetPixelShaderState(ctx->swrContext, &psState);
/* JIT sampler state */
if (ctx->dirty & (SWR_NEW_SAMPLER |
swr_update_constants(ctx, PIPE_SHADER_GEOMETRY);
}
+ /* Tessellation Control Shader Constants */
+ if (ctx->dirty & SWR_NEW_TCSCONSTANTS) {
+ swr_update_constants(ctx, PIPE_SHADER_TESS_CTRL);
+ }
+
+ /* Tessellation Evaluation Shader Constants */
+ if (ctx->dirty & SWR_NEW_TESCONSTANTS) {
+ swr_update_constants(ctx, PIPE_SHADER_TESS_EVAL);
+ }
+
/* Depth/stencil state */
if (ctx->dirty & (SWR_NEW_DEPTH_STENCIL_ALPHA | SWR_NEW_FRAMEBUFFER)) {
struct pipe_depth_state *depth = &(ctx->depth_stencil->depth);
depthStencilState.depthTestEnable = depth->enabled;
depthStencilState.depthTestFunc = swr_convert_depth_func(depth->func);
depthStencilState.depthWriteEnable = depth->writemask;
- SwrSetDepthStencilState(ctx->swrContext, &depthStencilState);
+ ctx->api.pfnSwrSetDepthStencilState(ctx->swrContext, &depthStencilState);
depthBoundsState.depthBoundsTestEnable = depth->bounds_test;
depthBoundsState.depthBoundsTestMinValue = depth->bounds_min;
depthBoundsState.depthBoundsTestMaxValue = depth->bounds_max;
- SwrSetDepthBoundsState(ctx->swrContext, &depthBoundsState);
+ ctx->api.pfnSwrSetDepthBoundsState(ctx->swrContext, &depthBoundsState);
}
/* Blend State */
blendState.renderTarget[0].writeDisableGreen = 1;
blendState.renderTarget[0].writeDisableBlue = 1;
blendState.renderTarget[0].writeDisableAlpha = 1;
- SwrSetBlendFunc(ctx->swrContext, 0, NULL);
+ ctx->api.pfnSwrSetBlendFunc(ctx->swrContext, 0, NULL);
}
else
for (int target = 0;
if (compileState.blendState.blendEnable == false &&
compileState.blendState.logicOpEnable == false &&
ctx->depth_stencil->alpha.enabled == 0) {
- SwrSetBlendFunc(ctx->swrContext, target, NULL);
+ ctx->api.pfnSwrSetBlendFunc(ctx->swrContext, target, NULL);
continue;
}
compileState.alphaTestFormat = ALPHA_TEST_FLOAT32; // xxx
compileState.Canonicalize();
-
+
PFN_BLEND_JIT_FUNC func = NULL;
auto search = ctx->blendJIT->find(compileState);
if (search != ctx->blendJIT->end()) {
ctx->blendJIT->insert(std::make_pair(compileState, func));
}
- SwrSetBlendFunc(ctx->swrContext, target, func);
+ ctx->api.pfnSwrSetBlendFunc(ctx->swrContext, target, func);
}
- SwrSetBlendState(ctx->swrContext, &blendState);
+ ctx->api.pfnSwrSetBlendState(ctx->swrContext, &blendState);
}
if (ctx->dirty & SWR_NEW_STIPPLE) {
swr_update_poly_stipple(ctx);
}
- if (ctx->dirty & (SWR_NEW_VS | SWR_NEW_SO | SWR_NEW_RASTERIZER)) {
+ if (ctx->dirty & (SWR_NEW_VS | SWR_NEW_TCS | SWR_NEW_TES | SWR_NEW_SO | SWR_NEW_RASTERIZER)) {
ctx->vs->soState.rasterizerDisable =
ctx->rasterizer->rasterizer_discard;
- SwrSetSoState(ctx->swrContext, &ctx->vs->soState);
+ ctx->api.pfnSwrSetSoState(ctx->swrContext, &ctx->vs->soState);
pipe_stream_output_info *stream_output = &ctx->vs->pipe.stream_output;
- for (uint32_t i = 0; i < ctx->num_so_targets; i++) {
+ for (uint32_t i = 0; i < MAX_SO_STREAMS; i++) {
SWR_STREAMOUT_BUFFER buffer = {0};
- if (!ctx->so_targets[i])
- continue;
- buffer.enable = true;
- buffer.pBuffer =
- (uint32_t *)(swr_resource_data(ctx->so_targets[i]->buffer) +
- ctx->so_targets[i]->buffer_offset);
- buffer.bufferSize = ctx->so_targets[i]->buffer_size >> 2;
- buffer.pitch = stream_output->stride[i];
- buffer.streamOffset = 0;
-
- SwrSetSoBuffers(ctx->swrContext, &buffer, i);
+ if (ctx->so_targets[i]) {
+ buffer.enable = true;
+ buffer.pBuffer =
+ (gfxptr_t)(swr_resource_data(ctx->so_targets[i]->buffer) +
+ ctx->so_targets[i]->buffer_offset);
+ buffer.bufferSize = ctx->so_targets[i]->buffer_size >> 2;
+ buffer.pitch = stream_output->stride[i];
+ buffer.streamOffset = 0;
+ }
+
+ ctx->api.pfnSwrSetSoBuffers(ctx->swrContext, &buffer, i);
}
}
+
if (ctx->dirty & (SWR_NEW_CLIP | SWR_NEW_RASTERIZER | SWR_NEW_VS)) {
// shader exporting clip distances overrides all user clip planes
if (ctx->rasterizer->clip_plane_enable &&
- !ctx->vs->info.base.num_written_clipdistance)
+ !swr_get_last_fe(ctx)->num_written_clipdistance)
{
swr_draw_context *pDC = &ctx->swrDC;
memcpy(pDC->userClipPlanes,
SWR_BACKEND_STATE backendState = {0};
if (ctx->gs) {
backendState.numAttributes = ctx->gs->info.base.num_outputs - 1;
- } else {
+ } else
+ if (ctx->tes) {
+ backendState.numAttributes = ctx->tes->info.base.num_outputs - 1;
+ // no case for TCS, because if TCS is active, TES must be active
+ // as well - pipeline stages after tessellation does not support patches
+ } else {
backendState.numAttributes = ctx->vs->info.base.num_outputs - 1;
if (ctx->fs->info.base.uses_primid) {
backendState.numAttributes++;
(ctx->rasterizer->flatshade ? ctx->fs->flatConstantMask : 0);
backendState.pointSpriteTexCoordMask = ctx->fs->pointSpriteMask;
- struct tgsi_shader_info *pLastFE =
- ctx->gs ?
- &ctx->gs->info.base :
- &ctx->vs->info.base;
+ struct tgsi_shader_info *pLastFE = swr_get_last_fe(ctx);
+
backendState.readRenderTargetArrayIndex = pLastFE->writes_layer;
backendState.readViewportArrayIndex = pLastFE->writes_viewport_index;
+ backendState.vertexAttribOffset = VERTEX_ATTRIB_START_SLOT; // TODO: optimize
+
+ backendState.clipDistanceMask =
+ pLastFE->num_written_clipdistance ?
+ pLastFE->clipdist_writemask & ctx->rasterizer->clip_plane_enable :
+ ctx->rasterizer->clip_plane_enable;
+
+ backendState.cullDistanceMask =
+ pLastFE->culldist_writemask << pLastFE->num_written_clipdistance;
- SwrSetBackendState(ctx->swrContext, &backendState);
+ // Assume old layout of SGV, POSITION, CLIPCULL, ATTRIB
+ backendState.vertexClipCullOffset = backendState.vertexAttribOffset - 2;
+
+ ctx->api.pfnSwrSetBackendState(ctx->swrContext, &backendState);
/* Ensure that any in-progress attachment change StoreTiles finish */
if (swr_is_fence_pending(screen->flush_fence))
}
swr->num_so_targets = num_targets;
+ swr->swrDC.soPrims = &swr->so_primCounter;
swr->dirty |= SWR_NEW_SO;
}
pipe->bind_gs_state = swr_bind_gs_state;
pipe->delete_gs_state = swr_delete_gs_state;
+ pipe->create_tcs_state = swr_create_tcs_state;
+ pipe->bind_tcs_state = swr_bind_tcs_state;
+ pipe->delete_tcs_state = swr_delete_tcs_state;
+
+ pipe->create_tes_state = swr_create_tes_state;
+ pipe->bind_tes_state = swr_bind_tes_state;
+ pipe->delete_tes_state = swr_delete_tes_state;
+
pipe->set_constant_buffer = swr_set_constant_buffer;
pipe->create_vertex_elements_state = swr_create_vertex_elements_state;