#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;
+ }
+}
+
static struct mali_rt_format
panfrost_mfbd_format(struct pipe_surface *surf)
{
/* Explode details on the format */
const struct util_format_description *desc =
- util_format_description(surf->texture->format);
+ util_format_description(surf->format);
- /* Fill in accordingly */
+ /* The swizzle for rendering is inverted from texturing */
+
+ unsigned char swizzle[4];
+ panfrost_invert_swizzle(desc->swizzle, swizzle);
+
+ /* Fill in accordingly, defaulting to 8-bit UNORM */
struct mali_rt_format fmt = {
.unk1 = 0x4000000,
.unk2 = 0x1,
.nr_channels = MALI_POSITIVE(desc->nr_channels),
- .flags = 0x444,
- .swizzle = panfrost_translate_swizzle_4(desc->swizzle),
+ .unk3 = 0x4,
+ .flags = 0x8,
+ .swizzle = panfrost_translate_swizzle_4(swizzle),
.unk4 = 0x8
};
+ if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
+ fmt.flags |= MALI_MFBD_FORMAT_SRGB;
+
+ /* sRGB handled as a dedicated flag */
+ enum pipe_format linearized = util_format_linear(surf->format);
+
+ /* If RGB, we're good to go */
+ if (util_format_is_unorm8(desc))
+ return fmt;
+
+ /* Set flags for alternative formats */
+
+ switch (linearized) {
+ case PIPE_FORMAT_B5G6R5_UNORM:
+ fmt.unk1 = 0x14000000;
+ fmt.nr_channels = MALI_POSITIVE(2);
+ fmt.unk3 |= 0x1;
+ break;
+
+ case PIPE_FORMAT_A4B4G4R4_UNORM:
+ case PIPE_FORMAT_B4G4R4A4_UNORM:
+ fmt.unk1 = 0x10000000;
+ fmt.unk3 = 0x5;
+ fmt.nr_channels = MALI_POSITIVE(1);
+ break;
+
+ case PIPE_FORMAT_R10G10B10A2_UNORM:
+ case PIPE_FORMAT_B10G10R10A2_UNORM:
+ case PIPE_FORMAT_R10G10B10X2_UNORM:
+ case PIPE_FORMAT_B10G10R10X2_UNORM:
+ fmt.unk1 = 0x08000000;
+ fmt.unk3 = 0x6;
+ fmt.nr_channels = MALI_POSITIVE(1);
+ break;
+
+ /* Generic 8-bit */
+ case PIPE_FORMAT_R8_UINT:
+ case PIPE_FORMAT_R8_SINT:
+ fmt.unk1 = 0x80000000;
+ fmt.unk3 = 0x0;
+ fmt.nr_channels = MALI_POSITIVE(1);
+ break;
+
+ /* Generic 32-bit */
+ case PIPE_FORMAT_R11G11B10_FLOAT:
+ case PIPE_FORMAT_R8G8B8A8_UINT:
+ case PIPE_FORMAT_R8G8B8A8_SINT:
+ case PIPE_FORMAT_R16G16_FLOAT:
+ case PIPE_FORMAT_R16G16_UINT:
+ case PIPE_FORMAT_R16G16_SINT:
+ case PIPE_FORMAT_R32_FLOAT:
+ case PIPE_FORMAT_R32_UINT:
+ case PIPE_FORMAT_R32_SINT:
+ case PIPE_FORMAT_R10G10B10A2_UINT:
+ fmt.unk1 = 0x88000000;
+ fmt.unk3 = 0x0;
+ fmt.nr_channels = MALI_POSITIVE(4);
+ break;
+
+ /* Generic 16-bit */
+ case PIPE_FORMAT_R8G8_UINT:
+ case PIPE_FORMAT_R8G8_SINT:
+ case PIPE_FORMAT_R16_FLOAT:
+ case PIPE_FORMAT_R16_UINT:
+ case PIPE_FORMAT_R16_SINT:
+ case PIPE_FORMAT_B5G5R5A1_UNORM:
+ fmt.unk1 = 0x84000000;
+ fmt.unk3 = 0x0;
+ fmt.nr_channels = MALI_POSITIVE(2);
+ break;
+
+ /* Generic 64-bit */
+ case PIPE_FORMAT_R32G32_FLOAT:
+ case PIPE_FORMAT_R32G32_SINT:
+ case PIPE_FORMAT_R32G32_UINT:
+ case PIPE_FORMAT_R16G16B16A16_FLOAT:
+ case PIPE_FORMAT_R16G16B16A16_SINT:
+ case PIPE_FORMAT_R16G16B16A16_UINT:
+ fmt.unk1 = 0x8c000000;
+ fmt.unk3 = 0x1;
+ fmt.nr_channels = MALI_POSITIVE(2);
+ break;
+
+ /* Generic 128-bit */
+ case PIPE_FORMAT_R32G32B32A32_FLOAT:
+ case PIPE_FORMAT_R32G32B32A32_SINT:
+ case PIPE_FORMAT_R32G32B32A32_UINT:
+ fmt.unk1 = 0x90000000;
+ fmt.unk3 = 0x1;
+ fmt.nr_channels = MALI_POSITIVE(4);
+ break;
+
+ default:
+ unreachable("Invalid format rendering");
+ }
+
return fmt;
}
-static void
-panfrost_mfbd_enable_msaa(struct panfrost_context *ctx)
-{
- ctx->fragment_rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
-
- /* XXX */
- ctx->fragment_mfbd.unk1 |= (1 << 4) | (1 << 1);
- ctx->fragment_mfbd.rt_count_2 = 4;
-}
static void
-panfrost_mfbd_clear(struct panfrost_job *job)
+panfrost_mfbd_clear(
+ struct panfrost_job *job,
+ struct bifrost_framebuffer *fb,
+ struct bifrost_fb_extra *fbx,
+ struct bifrost_render_target *rts,
+ unsigned rt_count)
{
- struct panfrost_context *ctx = job->ctx;
- struct bifrost_render_target *buffer_color = &ctx->fragment_rts[0];
- struct bifrost_framebuffer *buffer_ds = &ctx->fragment_mfbd;
-
- if (job->clear & PIPE_CLEAR_COLOR) {
- buffer_color->clear_color_1 = job->clear_color;
- buffer_color->clear_color_2 = job->clear_color;
- buffer_color->clear_color_3 = job->clear_color;
- buffer_color->clear_color_4 = job->clear_color;
+ for (unsigned i = 0; i < rt_count; ++i) {
+ if (!(job->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];
}
if (job->clear & PIPE_CLEAR_DEPTH) {
- buffer_ds->clear_depth = job->clear_depth;
+ fb->clear_depth = job->clear_depth;
}
if (job->clear & PIPE_CLEAR_STENCIL) {
- buffer_ds->clear_stencil = job->clear_stencil;
- }
-
- if (job->clear & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
- /* Setup combined 24/8 depth/stencil */
- ctx->fragment_mfbd.unk3 |= MALI_MFBD_EXTRA;
- ctx->fragment_extra.flags = 0x405;
- ctx->fragment_extra.ds_linear.depth = ctx->depth_stencil_buffer.gpu;
- ctx->fragment_extra.ds_linear.depth_stride = ctx->pipe_framebuffer.width * 4;
+ fb->clear_stencil = job->clear_stencil;
}
}
static void
panfrost_mfbd_set_cbuf(
- struct panfrost_context *ctx,
- struct pipe_surface *surf,
- unsigned cb)
+ struct bifrost_render_target *rt,
+ struct pipe_surface *surf)
{
struct panfrost_resource *rsrc = pan_resource(surf->texture);
- signed stride =
- util_format_get_stride(surf->format, surf->texture->width0);
+ unsigned level = surf->u.tex.level;
+ unsigned first_layer = surf->u.tex.first_layer;
+ assert(surf->u.tex.last_layer == first_layer);
+ int stride = rsrc->slices[level].stride;
- ctx->fragment_rts[cb].format = panfrost_mfbd_format(surf);
+ mali_ptr base = panfrost_get_texture_address(rsrc, level, first_layer);
- /* Now, we set the layout specific pieces */
-
- if (rsrc->bo->layout == PAN_LINEAR) {
- mali_ptr framebuffer = rsrc->bo->gpu[0];
+ rt->format = panfrost_mfbd_format(surf);
- /* The default is upside down from OpenGL's perspective. */
- if (panfrost_is_scanout(ctx)) {
- framebuffer += stride * (surf->texture->height0 - 1);
- stride = -stride;
- }
+ /* Now, we set the layout specific pieces */
- /* MFBD specifies stride in tiles */
- ctx->fragment_rts[cb].framebuffer = framebuffer;
- ctx->fragment_rts[cb].framebuffer_stride = stride / 16;
- } else if (rsrc->bo->layout == PAN_AFBC) {
- ctx->fragment_rts[cb].afbc.metadata = rsrc->bo->afbc_slab.gpu;
- ctx->fragment_rts[cb].afbc.stride = 0;
- ctx->fragment_rts[cb].afbc.unk = 0x30009;
+ if (rsrc->layout == PAN_LINEAR) {
+ rt->format.block = MALI_MFBD_BLOCK_LINEAR;
+ rt->framebuffer = base;
+ rt->framebuffer_stride = stride / 16;
+ } else if (rsrc->layout == PAN_TILED) {
+ rt->format.block = MALI_MFBD_BLOCK_TILED;
+ rt->framebuffer = base;
+ rt->framebuffer_stride = stride;
+ } else if (rsrc->layout == PAN_AFBC) {
+ rt->format.block = MALI_MFBD_BLOCK_AFBC;
- ctx->fragment_rts[cb].format.flags |= MALI_MFBD_FORMAT_AFBC;
+ unsigned header_size = rsrc->slices[level].header_size;
- mali_ptr afbc_main = rsrc->bo->afbc_slab.gpu + rsrc->bo->afbc_metadata_size;
- ctx->fragment_rts[cb].framebuffer = afbc_main;
+ rt->framebuffer = base + header_size;
+ rt->afbc.metadata = base;
+ rt->afbc.stride = 0;
+ rt->afbc.unk = 0x30009;
/* TODO: Investigate shift */
- ctx->fragment_rts[cb].framebuffer_stride = stride << 1;
+ rt->framebuffer_stride = stride << 1;
} else {
- fprintf(stderr, "Invalid render layout (cbuf %d)", cb);
+ fprintf(stderr, "Invalid render layout (cbuf)");
assert(0);
}
}
-static void
-panfrost_mfbd_set_targets(struct panfrost_context *ctx)
+/* Is a format encoded like Z24S8 and therefore compatible for render? */
+
+static bool
+panfrost_is_z24s8_variant(enum pipe_format fmt)
{
- for (int cb = 0; cb < ctx->pipe_framebuffer.nr_cbufs; ++cb) {
- struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[cb];
- panfrost_mfbd_set_cbuf(ctx, surf, cb);
+ switch (fmt) {
+ case PIPE_FORMAT_Z24_UNORM_S8_UINT:
+ case PIPE_FORMAT_Z24X8_UNORM:
+ return true;
+ default:
+ return false;
}
+}
- /* Enable depth/stencil AFBC for the framebuffer (not the render target) */
- if (ctx->pipe_framebuffer.zsbuf) {
- struct panfrost_resource *rsrc = (struct panfrost_resource *) ctx->pipe_framebuffer.zsbuf->texture;
+static void
+panfrost_mfbd_set_zsbuf(
+ struct bifrost_framebuffer *fb,
+ struct bifrost_fb_extra *fbx,
+ struct pipe_surface *surf)
+{
+ struct panfrost_resource *rsrc = pan_resource(surf->texture);
- if (rsrc->bo->layout == PAN_AFBC) {
- ctx->fragment_mfbd.unk3 |= MALI_MFBD_EXTRA;
+ unsigned level = surf->u.tex.level;
+ assert(surf->u.tex.first_layer == 0);
- ctx->fragment_extra.flags =
- MALI_EXTRA_PRESENT |
- MALI_EXTRA_AFBC |
- MALI_EXTRA_AFBC_ZS |
- MALI_EXTRA_ZS |
- 0x1; /* unknown */
+ unsigned offset = rsrc->slices[level].offset;
- ctx->fragment_extra.ds_afbc.depth_stencil_afbc_metadata = rsrc->bo->afbc_slab.gpu;
- ctx->fragment_extra.ds_afbc.depth_stencil_afbc_stride = 0;
+ if (rsrc->layout == PAN_AFBC) {
+ /* The only Z/S format we can compress is Z24S8 or variants
+ * thereof (handled by the state tracker) */
+ assert(panfrost_is_z24s8_variant(surf->format));
- ctx->fragment_extra.ds_afbc.depth_stencil = rsrc->bo->afbc_slab.gpu + rsrc->bo->afbc_metadata_size;
+ mali_ptr base = rsrc->bo->gpu + offset;
+ unsigned header_size = rsrc->slices[level].header_size;
- ctx->fragment_extra.ds_afbc.zero1 = 0x10009;
- ctx->fragment_extra.ds_afbc.padding = 0x1000;
+ fb->mfbd_flags |= MALI_MFBD_EXTRA;
- ctx->fragment_mfbd.unk3 |= MALI_MFBD_DEPTH_WRITE;
- }
- }
+ fbx->flags =
+ MALI_EXTRA_PRESENT |
+ MALI_EXTRA_AFBC |
+ MALI_EXTRA_AFBC_ZS |
+ MALI_EXTRA_ZS |
+ 0x1; /* unknown */
- /* For the special case of a depth-only FBO, we need to attach a dummy render target */
+ fbx->ds_afbc.depth_stencil = base + header_size;
+ fbx->ds_afbc.depth_stencil_afbc_metadata = base;
+ fbx->ds_afbc.depth_stencil_afbc_stride = 0;
- if (ctx->pipe_framebuffer.nr_cbufs == 0) {
- struct mali_rt_format null_rt = {
- .unk1 = 0x4000000,
- .unk4 = 0x8
- };
+ fbx->ds_afbc.zero1 = 0x10009;
+ fbx->ds_afbc.padding = 0x1000;
+ } else if (rsrc->layout == PAN_LINEAR) {
+ /* TODO: Z32F(S8) support, which is always linear */
+
+ int stride = rsrc->slices[level].stride;
+
+ fb->mfbd_flags |= MALI_MFBD_EXTRA;
+ fbx->flags |= MALI_EXTRA_PRESENT | MALI_EXTRA_ZS;
+
+ fbx->ds_linear.depth = rsrc->bo->gpu + offset;
+ fbx->ds_linear.depth_stride = stride;
+
+ if (panfrost_is_z24s8_variant(surf->format)) {
+ fbx->flags |= 0x1;
+ } else if (surf->format == PIPE_FORMAT_Z32_UNORM) {
+ /* default flags (0 in bottom place) */
+ }
- ctx->fragment_rts[0].format = null_rt;
- ctx->fragment_rts[0].framebuffer = 0;
- ctx->fragment_rts[0].framebuffer_stride = 0;
+ } else {
+ assert(0);
}
}
/* Helper for sequential uploads used for MFBD */
#define UPLOAD(dest, offset, src, max) { \
- size_t sz = sizeof(src); \
- memcpy(dest.cpu + offset, &src, sz); \
+ size_t sz = sizeof(*src); \
+ memcpy(dest.cpu + offset, src, sz); \
assert((offset + sz) <= max); \
offset += sz; \
}
static mali_ptr
-panfrost_mfbd_upload(struct panfrost_context *ctx)
+panfrost_mfbd_upload(
+ struct panfrost_context *ctx,
+ struct bifrost_framebuffer *fb,
+ struct bifrost_fb_extra *fbx,
+ struct bifrost_render_target *rts,
+ unsigned cbufs)
{
off_t offset = 0;
- /* We always upload at least one (dummy) cbuf */
- unsigned cbufs = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
-
/* There may be extra data stuck in the middle */
- bool has_extra = ctx->fragment_mfbd.unk3 & MALI_MFBD_EXTRA;
+ bool has_extra = fb->mfbd_flags & MALI_MFBD_EXTRA;
/* Compute total size for transfer */
/* Do the transfer */
- UPLOAD(m_f_trans, offset, ctx->fragment_mfbd, total_sz);
+ UPLOAD(m_f_trans, offset, fb, total_sz);
if (has_extra)
- UPLOAD(m_f_trans, offset, ctx->fragment_extra, total_sz);
+ UPLOAD(m_f_trans, offset, fbx, total_sz);
for (unsigned c = 0; c < cbufs; ++c) {
- UPLOAD(m_f_trans, offset, ctx->fragment_rts[c], total_sz);
+ UPLOAD(m_f_trans, offset, &rts[c], total_sz);
}
- /* Return pointer suitable for the fragment seciton */
+ /* Return pointer suitable for the fragment section */
return m_f_trans.gpu | MALI_MFBD | (has_extra ? 2 : 0);
}
/* Creates an MFBD for the FRAGMENT section of the bound framebuffer */
mali_ptr
-panfrost_mfbd_fragment(struct panfrost_context *ctx)
+panfrost_mfbd_fragment(struct panfrost_context *ctx, bool has_draws)
{
struct panfrost_job *job = panfrost_get_job_for_fbo(ctx);
- struct bifrost_framebuffer fb = panfrost_emit_mfbd(ctx);
+ struct bifrost_framebuffer fb = panfrost_emit_mfbd(ctx, has_draws);
+ struct bifrost_fb_extra fbx = {};
+ struct bifrost_render_target rts[4] = {};
/* XXX: MRT case */
fb.rt_count_2 = 1;
- fb.unk3 = 0x100;
+ fb.mfbd_flags = 0x100;
- struct bifrost_render_target rt = {};
+ /* TODO: MRT clear */
+ panfrost_mfbd_clear(job, &fb, &fbx, rts, fb.rt_count_2);
- memcpy(&ctx->fragment_rts[0], &rt, sizeof(rt));
- memset(&ctx->fragment_extra, 0, sizeof(ctx->fragment_extra));
- memcpy(&ctx->fragment_mfbd, &fb, sizeof(fb));
+ 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_clear(job);
- panfrost_mfbd_set_targets(ctx);
+ panfrost_mfbd_set_cbuf(&rts[cb], surf);
- if (job->msaa)
- panfrost_mfbd_enable_msaa(ctx);
+ /* 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);
+ }
- if (ctx->pipe_framebuffer.nr_cbufs == 1) {
- struct panfrost_resource *rsrc = (struct panfrost_resource *) ctx->pipe_framebuffer.cbufs[0]->texture;
+ if (ctx->pipe_framebuffer.zsbuf) {
+ panfrost_mfbd_set_zsbuf(&fb, &fbx, ctx->pipe_framebuffer.zsbuf);
+ }
- if (rsrc->bo->has_checksum) {
- int stride = util_format_get_stride(rsrc->base.format, rsrc->base.width0);
+ /* For the special case of a depth-only FBO, we need to attach a dummy render target */
- ctx->fragment_mfbd.unk3 |= MALI_MFBD_EXTRA;
- ctx->fragment_extra.flags |= MALI_EXTRA_PRESENT;
- ctx->fragment_extra.checksum_stride = rsrc->bo->checksum_stride;
- ctx->fragment_extra.checksum = rsrc->bo->gpu[0] + stride * rsrc->base.height0;
+ if (ctx->pipe_framebuffer.nr_cbufs == 0) {
+ struct mali_rt_format null_rt = {
+ .unk1 = 0x4000000,
+ .unk4 = 0x8
+ };
+
+ rts[0].format = null_rt;
+ rts[0].framebuffer = 0;
+ rts[0].framebuffer_stride = 0;
+ }
+
+ /* When scanning out, the depth buffer is immediately invalidated, so
+ * we don't need to waste bandwidth writing it out. This can improve
+ * performance substantially (Z32_UNORM 1080p @ 60fps is 475 MB/s of
+ * memory bandwidth!).
+ *
+ * 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;
+ }
+
+ /* Actualize the requirements */
+
+ if (job->requirements & PAN_REQ_MSAA) {
+ rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
+
+ /* XXX */
+ fb.unk1 |= (1 << 4) | (1 << 1);
+ fb.rt_count_2 = 4;
+ }
+
+ if (job->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];
+ struct panfrost_resource *rsrc = pan_resource(surf->texture);
+ struct panfrost_bo *bo = rsrc->bo;
+
+ if (rsrc->checksummed) {
+ unsigned level = surf->u.tex.level;
+ struct panfrost_slice *slice = &rsrc->slices[level];
+
+ fb.mfbd_flags |= MALI_MFBD_EXTRA;
+ fbx.flags |= MALI_EXTRA_PRESENT;
+ fbx.checksum_stride = slice->checksum_stride;
+ fbx.checksum = bo->gpu + slice->checksum_offset;
}
}
- return panfrost_mfbd_upload(ctx);
+ /* 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);
}