*
*/
+#include "pan_bo.h"
#include "pan_context.h"
#include "pan_util.h"
#include "pan_format.h"
-#include "util/u_format.h"
-
-static void
-panfrost_invert_swizzle(const unsigned char *in, unsigned char *out)
-{
- /* First, default to all zeroes to prevent uninitialized junk */
-
- for (unsigned c = 0; c < 4; ++c)
- out[c] = PIPE_SWIZZLE_0;
-
- /* Now "do" what the swizzle says */
-
- for (unsigned c = 0; c < 4; ++c) {
- unsigned char i = in[c];
-
- /* Who cares? */
- if (i < PIPE_SWIZZLE_X || i > PIPE_SWIZZLE_W)
- continue;
-
- /* Invert */
- unsigned idx = i - PIPE_SWIZZLE_X;
- out[idx] = PIPE_SWIZZLE_X + c;
- }
-}
+#include "util/format/u_format.h"
static struct mali_rt_format
panfrost_mfbd_format(struct pipe_surface *surf)
.unk3 = 0x4,
.flags = 0x8,
.swizzle = panfrost_translate_swizzle_4(swizzle),
- .unk4 = 0x8
+ .no_preload = true
};
if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
static void
panfrost_mfbd_clear(
- struct panfrost_job *job,
+ struct panfrost_batch *batch,
struct bifrost_framebuffer *fb,
struct bifrost_fb_extra *fbx,
struct bifrost_render_target *rts,
unsigned rt_count)
{
for (unsigned i = 0; i < rt_count; ++i) {
- if (!(job->clear & (PIPE_CLEAR_COLOR0 << i)))
+ if (!(batch->clear & (PIPE_CLEAR_COLOR0 << i)))
continue;
- rts[i].clear_color_1 = job->clear_color[i][0];
- rts[i].clear_color_2 = job->clear_color[i][1];
- rts[i].clear_color_3 = job->clear_color[i][2];
- rts[i].clear_color_4 = job->clear_color[i][3];
+ rts[i].clear_color_1 = batch->clear_color[i][0];
+ rts[i].clear_color_2 = batch->clear_color[i][1];
+ rts[i].clear_color_3 = batch->clear_color[i][2];
+ rts[i].clear_color_4 = batch->clear_color[i][3];
}
- if (job->clear & PIPE_CLEAR_DEPTH) {
- fb->clear_depth = job->clear_depth;
+ if (batch->clear & PIPE_CLEAR_DEPTH) {
+ fb->clear_depth = batch->clear_depth;
}
- if (job->clear & PIPE_CLEAR_STENCIL) {
- fb->clear_stencil = job->clear_stencil;
+ if (batch->clear & PIPE_CLEAR_STENCIL) {
+ fb->clear_stencil = batch->clear_stencil;
}
}
/* Now, we set the layout specific pieces */
if (rsrc->layout == PAN_LINEAR) {
- rt->format.block = MALI_MFBD_BLOCK_LINEAR;
+ rt->format.block = MALI_BLOCK_LINEAR;
rt->framebuffer = base;
rt->framebuffer_stride = stride / 16;
} else if (rsrc->layout == PAN_TILED) {
- rt->format.block = MALI_MFBD_BLOCK_TILED;
+ rt->format.block = MALI_BLOCK_TILED;
rt->framebuffer = base;
rt->framebuffer_stride = stride;
} else if (rsrc->layout == PAN_AFBC) {
- rt->format.block = MALI_MFBD_BLOCK_AFBC;
+ rt->format.block = MALI_BLOCK_AFBC;
unsigned header_size = rsrc->slices[level].header_size;
}
}
-/* Is a format encoded like Z24S8 and therefore compatible for render? */
-
-static bool
-panfrost_is_z24s8_variant(enum pipe_format fmt)
-{
- switch (fmt) {
- case PIPE_FORMAT_Z24_UNORM_S8_UINT:
- case PIPE_FORMAT_Z24X8_UNORM:
- return true;
- default:
- return false;
- }
-}
-
static void
panfrost_mfbd_set_zsbuf(
struct bifrost_framebuffer *fb,
fbx->flags |= 0x1;
} else if (surf->format == PIPE_FORMAT_Z32_UNORM) {
/* default flags (0 in bottom place) */
+ } else if (surf->format == PIPE_FORMAT_Z32_FLOAT) {
+ fbx->flags |= 0xA;
+ fb->mfbd_flags ^= 0x100;
+ fb->mfbd_flags |= 0x200;
+ } else if (surf->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) {
+ fbx->flags |= 0x1000A;
+ fb->mfbd_flags ^= 0x100;
+ fb->mfbd_flags |= 0x201;
+
+ struct panfrost_resource *stencil = rsrc->separate_stencil;
+ struct panfrost_slice stencil_slice = stencil->slices[level];
+
+ fbx->ds_linear.stencil = stencil->bo->gpu + stencil_slice.offset;
+ fbx->ds_linear.stencil_stride = stencil_slice.stride;
}
} else {
}
static mali_ptr
-panfrost_mfbd_upload(
- struct panfrost_context *ctx,
+panfrost_mfbd_upload(struct panfrost_batch *batch,
struct bifrost_framebuffer *fb,
struct bifrost_fb_extra *fbx,
struct bifrost_render_target *rts,
- unsigned cbufs)
+ unsigned rt_count)
{
off_t offset = 0;
size_t total_sz =
sizeof(struct bifrost_framebuffer) +
(has_extra ? sizeof(struct bifrost_fb_extra) : 0) +
- sizeof(struct bifrost_render_target) * cbufs;
+ sizeof(struct bifrost_render_target) * 4;
struct panfrost_transfer m_f_trans =
- panfrost_allocate_transient(ctx, total_sz);
+ panfrost_allocate_transient(batch, total_sz);
/* Do the transfer */
if (has_extra)
UPLOAD(m_f_trans, offset, fbx, total_sz);
- for (unsigned c = 0; c < cbufs; ++c) {
+ for (unsigned c = 0; c < 4; ++c) {
UPLOAD(m_f_trans, offset, &rts[c], total_sz);
}
/* Return pointer suitable for the fragment section */
- return m_f_trans.gpu | MALI_MFBD | (has_extra ? 2 : 0);
+ unsigned tag =
+ MALI_MFBD |
+ (has_extra ? MALI_MFBD_TAG_EXTRA : 0) |
+ (MALI_POSITIVE(rt_count) << 2);
+
+ return m_f_trans.gpu | tag;
}
#undef UPLOAD
+static struct bifrost_framebuffer
+panfrost_emit_mfbd(struct panfrost_batch *batch, unsigned vertex_count)
+{
+ struct panfrost_context *ctx = batch->ctx;
+ struct pipe_context *gallium = (struct pipe_context *) ctx;
+ struct panfrost_screen *screen = pan_screen(gallium->screen);
+
+ unsigned width = batch->key.width;
+ unsigned height = batch->key.height;
+
+ unsigned shift = panfrost_get_stack_shift(batch->stack_size);
+
+ struct bifrost_framebuffer framebuffer = {
+ .width1 = MALI_POSITIVE(width),
+ .height1 = MALI_POSITIVE(height),
+ .width2 = MALI_POSITIVE(width),
+ .height2 = MALI_POSITIVE(height),
+
+ .unk1 = 0x1080,
+
+ .rt_count_1 = MALI_POSITIVE(batch->key.nr_cbufs),
+ .rt_count_2 = 4,
+
+ .unknown2 = 0x1f,
+ .tiler = panfrost_emit_midg_tiler(batch, vertex_count),
+
+ .stack_shift = shift,
+ .unk0 = 0x1e,
+ .scratchpad = panfrost_batch_get_scratchpad(batch, shift, screen->thread_tls_alloc, screen->core_count)->gpu
+ };
+
+ return framebuffer;
+}
+
+void
+panfrost_attach_mfbd(struct panfrost_batch *batch, unsigned vertex_count)
+{
+ struct bifrost_framebuffer mfbd =
+ panfrost_emit_mfbd(batch, vertex_count);
+
+ memcpy(batch->framebuffer.cpu, &mfbd, sizeof(mfbd));
+}
+
/* Creates an MFBD for the FRAGMENT section of the bound framebuffer */
mali_ptr
-panfrost_mfbd_fragment(struct panfrost_context *ctx, bool has_draws)
+panfrost_mfbd_fragment(struct panfrost_batch *batch, bool has_draws)
{
- struct panfrost_job *job = panfrost_get_job_for_fbo(ctx);
+ struct bifrost_framebuffer fb = panfrost_emit_mfbd(batch, has_draws);
+ struct bifrost_fb_extra fbx = {0};
+ struct bifrost_render_target rts[4] = {0};
+
+ /* We always upload at least one dummy GL_NONE render target */
- struct bifrost_framebuffer fb = panfrost_emit_mfbd(ctx, has_draws);
- struct bifrost_fb_extra fbx = {};
- struct bifrost_render_target rts[4] = {};
+ unsigned rt_descriptors = MAX2(batch->key.nr_cbufs, 1);
- /* XXX: MRT case */
- fb.rt_count_2 = 1;
+ fb.rt_count_1 = MALI_POSITIVE(rt_descriptors);
+ fb.rt_count_2 = rt_descriptors;
fb.mfbd_flags = 0x100;
/* TODO: MRT clear */
- panfrost_mfbd_clear(job, &fb, &fbx, rts, fb.rt_count_2);
+ panfrost_mfbd_clear(batch, &fb, &fbx, rts, fb.rt_count_2);
- for (int cb = 0; cb < ctx->pipe_framebuffer.nr_cbufs; ++cb) {
- struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[cb];
- unsigned bpp = util_format_get_blocksize(surf->format);
- panfrost_mfbd_set_cbuf(&rts[cb], surf);
+ /* Upload either the render target or a dummy GL_NONE target */
- /* What is this? Looks like some extension of the bpp field.
- * Maybe it establishes how much internal tilebuffer space is
- * reserved? */
- fb.rt_count_2 = MAX2(fb.rt_count_2, ALIGN_POT(bpp, 4) / 4);
- }
+ for (int cb = 0; cb < rt_descriptors; ++cb) {
+ struct pipe_surface *surf = batch->key.cbufs[cb];
- if (ctx->pipe_framebuffer.zsbuf) {
- panfrost_mfbd_set_zsbuf(&fb, &fbx, ctx->pipe_framebuffer.zsbuf);
- }
+ if (surf) {
+ panfrost_mfbd_set_cbuf(&rts[cb], surf);
- /* For the special case of a depth-only FBO, we need to attach a dummy render target */
+ /* What is this? Looks like some extension of the bpp
+ * field. Maybe it establishes how much internal
+ * tilebuffer space is reserved? */
- if (ctx->pipe_framebuffer.nr_cbufs == 0) {
- struct mali_rt_format null_rt = {
- .unk1 = 0x4000000,
- .unk4 = 0x8
- };
+ unsigned bpp = util_format_get_blocksize(surf->format);
+ fb.rt_count_2 = MAX2(fb.rt_count_2, ALIGN_POT(bpp, 4) / 4);
+ } else {
+ struct mali_rt_format null_rt = {
+ .unk1 = 0x4000000,
+ .no_preload = true
+ };
- rts[0].format = null_rt;
- rts[0].framebuffer = 0;
- rts[0].framebuffer_stride = 0;
+ rts[cb].format = null_rt;
+ rts[cb].framebuffer = 0;
+ rts[cb].framebuffer_stride = 0;
+ }
+
+ /* TODO: Break out the field */
+ rts[cb].format.unk1 |= (cb * 0x400);
+ }
+
+ if (batch->key.zsbuf) {
+ panfrost_mfbd_set_zsbuf(&fb, &fbx, batch->key.zsbuf);
}
/* When scanning out, the depth buffer is immediately invalidated, so
* The exception is ReadPixels, but this is not supported on GLES so we
* can safely ignore it. */
- if (panfrost_is_scanout(ctx)) {
- job->requirements &= ~PAN_REQ_DEPTH_WRITE;
- }
+ if (panfrost_batch_is_scanout(batch))
+ batch->requirements &= ~PAN_REQ_DEPTH_WRITE;
/* Actualize the requirements */
- if (job->requirements & PAN_REQ_MSAA) {
+ if (batch->requirements & PAN_REQ_MSAA) {
rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
/* XXX */
fb.rt_count_2 = 4;
}
- if (job->requirements & PAN_REQ_DEPTH_WRITE)
+ if (batch->requirements & PAN_REQ_DEPTH_WRITE)
fb.mfbd_flags |= MALI_MFBD_DEPTH_WRITE;
/* Checksumming only works with a single render target */
- if (ctx->pipe_framebuffer.nr_cbufs == 1) {
- struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[0];
+ if (batch->key.nr_cbufs == 1) {
+ struct pipe_surface *surf = batch->key.cbufs[0];
struct panfrost_resource *rsrc = pan_resource(surf->texture);
struct panfrost_bo *bo = rsrc->bo;
}
}
- /* We always upload at least one (dummy) cbuf */
- unsigned cbufs = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
-
- return panfrost_mfbd_upload(ctx, &fb, &fbx, rts, cbufs);
+ return panfrost_mfbd_upload(batch, &fb, &fbx, rts, rt_descriptors);
}
projects / mesa.git / blobdiff