#include "util/u_string.h"
#include "util/u_memory.h"
#include "util/u_inlines.h"
-#include "util/u_format.h"
+#include "util/format/u_format.h"
#include "freedreno_draw.h"
+#include "freedreno_log.h"
#include "freedreno_state.h"
#include "freedreno_resource.h"
#include "fd6_emit.h"
#include "fd6_program.h"
#include "fd6_format.h"
+#include "fd6_resource.h"
#include "fd6_zsa.h"
+#include "fd6_pack.h"
-/* some bits in common w/ a4xx: */
-#include "a4xx/fd4_draw.h"
+/**
+ * Emits the flags registers, suitable for RB_MRT_FLAG_BUFFER,
+ * RB_DEPTH_FLAG_BUFFER, SP_PS_2D_SRC_FLAGS, and RB_BLIT_FLAG_DST.
+ */
+void
+fd6_emit_flag_reference(struct fd_ringbuffer *ring, struct fd_resource *rsc,
+ int level, int layer)
+{
+ if (fd_resource_ubwc_enabled(rsc, level)) {
+ OUT_RELOC(ring, rsc->bo, fd_resource_ubwc_offset(rsc, level, layer), 0, 0);
+ OUT_RING(ring,
+ A6XX_RB_MRT_FLAG_BUFFER_PITCH_PITCH(rsc->layout.ubwc_slices[level].pitch) |
+ A6XX_RB_MRT_FLAG_BUFFER_PITCH_ARRAY_PITCH(rsc->layout.ubwc_layer_size >> 2));
+ } else {
+ OUT_RING(ring, 0x00000000); /* RB_MRT_FLAG_BUFFER[i].ADDR_LO */
+ OUT_RING(ring, 0x00000000); /* RB_MRT_FLAG_BUFFER[i].ADDR_HI */
+ OUT_RING(ring, 0x00000000);
+ }
+}
static void
emit_mrt(struct fd_ringbuffer *ring, struct pipe_framebuffer_state *pfb,
- struct fd_gmem_stateobj *gmem)
+ const struct fd_gmem_stateobj *gmem)
{
unsigned char mrt_comp[A6XX_MAX_RENDER_TARGETS] = {0};
unsigned srgb_cntl = 0;
unsigned i;
- bool layered = false;
- unsigned type = 0;
+ unsigned max_layer_index = 0;
for (i = 0; i < pfb->nr_cbufs; i++) {
- enum a6xx_color_fmt format = 0;
+ enum a6xx_format format = 0;
enum a3xx_color_swap swap = WZYX;
bool sint = false, uint = false;
struct fd_resource *rsc = NULL;
- struct fd_resource_slice *slice = NULL;
+ struct fdl_slice *slice = NULL;
uint32_t stride = 0;
- uint32_t offset, ubwc_offset;
+ uint32_t offset;
uint32_t tile_mode;
- bool ubwc_enabled;
if (!pfb->cbufs[i])
continue;
offset = fd_resource_offset(rsc, psurf->u.tex.level,
psurf->u.tex.first_layer);
- ubwc_offset = fd_resource_ubwc_offset(rsc, psurf->u.tex.level,
- psurf->u.tex.first_layer);
- ubwc_enabled = fd_resource_ubwc_enabled(rsc, psurf->u.tex.level);
-
- stride = slice->pitch * rsc->cpp * pfb->samples;
- swap = rsc->tile_mode ? WZYX : fd6_pipe2swap(pformat);
-
- if (rsc->tile_mode &&
- fd_resource_level_linear(psurf->texture, psurf->u.tex.level))
- tile_mode = TILE6_LINEAR;
- else
- tile_mode = rsc->tile_mode;
-
- if (psurf->u.tex.first_layer < psurf->u.tex.last_layer) {
- layered = true;
- if (psurf->texture->target == PIPE_TEXTURE_2D_ARRAY && psurf->texture->nr_samples > 0)
- type = LAYER_MULTISAMPLE_ARRAY;
- else if (psurf->texture->target == PIPE_TEXTURE_2D_ARRAY)
- type = LAYER_2D_ARRAY;
- else if (psurf->texture->target == PIPE_TEXTURE_CUBE)
- type = LAYER_CUBEMAP;
- else if (psurf->texture->target == PIPE_TEXTURE_3D)
- type = LAYER_3D;
-
- stride /= pfb->samples;
- }
- debug_assert((offset + slice->size0) <= fd_bo_size(rsc->bo));
+ stride = slice->pitch;
+ swap = fd6_resource_swap(rsc, pformat);
- OUT_PKT4(ring, REG_A6XX_RB_MRT_BUF_INFO(i), 6);
- OUT_RING(ring, A6XX_RB_MRT_BUF_INFO_COLOR_FORMAT(format) |
- A6XX_RB_MRT_BUF_INFO_COLOR_TILE_MODE(tile_mode) |
- A6XX_RB_MRT_BUF_INFO_COLOR_SWAP(swap));
- OUT_RING(ring, A6XX_RB_MRT_PITCH(stride));
- OUT_RING(ring, A6XX_RB_MRT_ARRAY_PITCH(slice->size0));
- OUT_RELOCW(ring, rsc->bo, offset, 0, 0); /* BASE_LO/HI */
- OUT_RING(ring, base); /* RB_MRT[i].BASE_GMEM */
- OUT_PKT4(ring, REG_A6XX_SP_FS_MRT_REG(i), 1);
- OUT_RING(ring, A6XX_SP_FS_MRT_REG_COLOR_FORMAT(format) |
- COND(sint, A6XX_SP_FS_MRT_REG_COLOR_SINT) |
- COND(uint, A6XX_SP_FS_MRT_REG_COLOR_UINT));
+ tile_mode = fd_resource_tile_mode(psurf->texture, psurf->u.tex.level);
+ max_layer_index = psurf->u.tex.last_layer - psurf->u.tex.first_layer;
- OUT_PKT4(ring, REG_A6XX_RB_MRT_FLAG_BUFFER(i), 3);
- if (ubwc_enabled) {
- OUT_RELOCW(ring, rsc->bo, ubwc_offset, 0, 0); /* BASE_LO/HI */
- OUT_RING(ring, A6XX_RB_MRT_FLAG_BUFFER_PITCH_PITCH(rsc->ubwc_pitch) |
- A6XX_RB_MRT_FLAG_BUFFER_PITCH_ARRAY_PITCH(rsc->ubwc_size));
- } else {
- OUT_RING(ring, 0x00000000); /* RB_MRT_FLAG_BUFFER[i].ADDR_LO */
- OUT_RING(ring, 0x00000000); /* RB_MRT_FLAG_BUFFER[i].ADDR_HI */
- OUT_RING(ring, 0x00000000);
- }
- }
+ debug_assert((offset + slice->size0) <= fd_bo_size(rsc->bo));
- OUT_PKT4(ring, REG_A6XX_RB_SRGB_CNTL, 1);
- OUT_RING(ring, srgb_cntl);
+ OUT_REG(ring,
+ A6XX_RB_MRT_BUF_INFO(i,
+ .color_format = format,
+ .color_tile_mode = tile_mode,
+ .color_swap = swap),
+ A6XX_RB_MRT_PITCH(i, .a6xx_rb_mrt_pitch = stride),
+ A6XX_RB_MRT_ARRAY_PITCH(i, .a6xx_rb_mrt_array_pitch = slice->size0),
+ A6XX_RB_MRT_BASE(i, .bo = rsc->bo, .bo_offset = offset),
+ A6XX_RB_MRT_BASE_GMEM(i, .unknown = base));
- OUT_PKT4(ring, REG_A6XX_SP_SRGB_CNTL, 1);
- OUT_RING(ring, srgb_cntl);
+ OUT_REG(ring,
+ A6XX_SP_FS_MRT_REG(i, .color_format = format,
+ .color_sint = sint, .color_uint = uint));
- OUT_PKT4(ring, REG_A6XX_RB_RENDER_COMPONENTS, 1);
- OUT_RING(ring, A6XX_RB_RENDER_COMPONENTS_RT0(mrt_comp[0]) |
- A6XX_RB_RENDER_COMPONENTS_RT1(mrt_comp[1]) |
- A6XX_RB_RENDER_COMPONENTS_RT2(mrt_comp[2]) |
- A6XX_RB_RENDER_COMPONENTS_RT3(mrt_comp[3]) |
- A6XX_RB_RENDER_COMPONENTS_RT4(mrt_comp[4]) |
- A6XX_RB_RENDER_COMPONENTS_RT5(mrt_comp[5]) |
- A6XX_RB_RENDER_COMPONENTS_RT6(mrt_comp[6]) |
- A6XX_RB_RENDER_COMPONENTS_RT7(mrt_comp[7]));
+ OUT_PKT4(ring, REG_A6XX_RB_MRT_FLAG_BUFFER(i), 3);
+ fd6_emit_flag_reference(ring, rsc,
+ psurf->u.tex.level, psurf->u.tex.first_layer);
+ }
- OUT_PKT4(ring, REG_A6XX_SP_FS_RENDER_COMPONENTS, 1);
- OUT_RING(ring,
- A6XX_SP_FS_RENDER_COMPONENTS_RT0(mrt_comp[0]) |
- A6XX_SP_FS_RENDER_COMPONENTS_RT1(mrt_comp[1]) |
- A6XX_SP_FS_RENDER_COMPONENTS_RT2(mrt_comp[2]) |
- A6XX_SP_FS_RENDER_COMPONENTS_RT3(mrt_comp[3]) |
- A6XX_SP_FS_RENDER_COMPONENTS_RT4(mrt_comp[4]) |
- A6XX_SP_FS_RENDER_COMPONENTS_RT5(mrt_comp[5]) |
- A6XX_SP_FS_RENDER_COMPONENTS_RT6(mrt_comp[6]) |
- A6XX_SP_FS_RENDER_COMPONENTS_RT7(mrt_comp[7]));
-
- OUT_PKT4(ring, REG_A6XX_GRAS_LAYER_CNTL, 1);
- OUT_RING(ring, COND(layered, A6XX_GRAS_LAYER_CNTL_LAYERED |
- A6XX_GRAS_LAYER_CNTL_TYPE(type)));
+ OUT_REG(ring, A6XX_RB_SRGB_CNTL(.dword = srgb_cntl));
+ OUT_REG(ring, A6XX_SP_SRGB_CNTL(.dword = srgb_cntl));
+
+ OUT_REG(ring, A6XX_RB_RENDER_COMPONENTS(
+ .rt0 = mrt_comp[0],
+ .rt1 = mrt_comp[1],
+ .rt2 = mrt_comp[2],
+ .rt3 = mrt_comp[3],
+ .rt4 = mrt_comp[4],
+ .rt5 = mrt_comp[5],
+ .rt6 = mrt_comp[6],
+ .rt7 = mrt_comp[7]));
+
+ OUT_REG(ring, A6XX_SP_FS_RENDER_COMPONENTS(
+ .rt0 = mrt_comp[0],
+ .rt1 = mrt_comp[1],
+ .rt2 = mrt_comp[2],
+ .rt3 = mrt_comp[3],
+ .rt4 = mrt_comp[4],
+ .rt5 = mrt_comp[5],
+ .rt6 = mrt_comp[6],
+ .rt7 = mrt_comp[7]));
+
+ OUT_REG(ring, A6XX_GRAS_MAX_LAYER_INDEX(max_layer_index));
}
static void
emit_zs(struct fd_ringbuffer *ring, struct pipe_surface *zsbuf,
- struct fd_gmem_stateobj *gmem)
+ const struct fd_gmem_stateobj *gmem)
{
if (zsbuf) {
struct fd_resource *rsc = fd_resource(zsbuf->texture);
enum a6xx_depth_format fmt = fd6_pipe2depth(zsbuf->format);
- struct fd_resource_slice *slice = fd_resource_slice(rsc, 0);
- uint32_t stride = slice->pitch * rsc->cpp;
+ struct fdl_slice *slice = fd_resource_slice(rsc, 0);
+ uint32_t stride = slice->pitch;
uint32_t size = slice->size0;
uint32_t base = gmem ? gmem->zsbuf_base[0] : 0;
uint32_t offset = fd_resource_offset(rsc, zsbuf->u.tex.level,
zsbuf->u.tex.first_layer);
- uint32_t ubwc_offset = fd_resource_ubwc_offset(rsc, zsbuf->u.tex.level,
- zsbuf->u.tex.first_layer);
- bool ubwc_enabled = fd_resource_ubwc_enabled(rsc, zsbuf->u.tex.level);
-
- OUT_PKT4(ring, REG_A6XX_RB_DEPTH_BUFFER_INFO, 6);
- OUT_RING(ring, A6XX_RB_DEPTH_BUFFER_INFO_DEPTH_FORMAT(fmt));
- OUT_RING(ring, A6XX_RB_DEPTH_BUFFER_PITCH(stride));
- OUT_RING(ring, A6XX_RB_DEPTH_BUFFER_ARRAY_PITCH(size));
- OUT_RELOCW(ring, rsc->bo, offset, 0, 0); /* RB_DEPTH_BUFFER_BASE_LO/HI */
- OUT_RING(ring, base); /* RB_DEPTH_BUFFER_BASE_GMEM */
+ OUT_REG(ring,
+ A6XX_RB_DEPTH_BUFFER_INFO(.depth_format = fmt),
+ A6XX_RB_DEPTH_BUFFER_PITCH(.a6xx_rb_depth_buffer_pitch = stride),
+ A6XX_RB_DEPTH_BUFFER_ARRAY_PITCH(.a6xx_rb_depth_buffer_array_pitch = size),
+ A6XX_RB_DEPTH_BUFFER_BASE(.bo = rsc->bo, .bo_offset = offset),
+ A6XX_RB_DEPTH_BUFFER_BASE_GMEM(.dword = base));
- OUT_PKT4(ring, REG_A6XX_GRAS_SU_DEPTH_BUFFER_INFO, 1);
- OUT_RING(ring, A6XX_GRAS_SU_DEPTH_BUFFER_INFO_DEPTH_FORMAT(fmt));
+ OUT_REG(ring, A6XX_GRAS_SU_DEPTH_BUFFER_INFO(.depth_format = fmt));
OUT_PKT4(ring, REG_A6XX_RB_DEPTH_FLAG_BUFFER_BASE_LO, 3);
- if (ubwc_enabled) {
- OUT_RELOCW(ring, rsc->bo, ubwc_offset, 0, 0); /* BASE_LO/HI */
- OUT_RING(ring, A6XX_RB_DEPTH_FLAG_BUFFER_PITCH_PITCH(rsc->ubwc_pitch) |
- A6XX_RB_DEPTH_FLAG_BUFFER_PITCH_ARRAY_PITCH(rsc->ubwc_size));
- } else {
- OUT_RING(ring, 0x00000000); /* RB_DEPTH_FLAG_BUFFER_BASE_LO */
- OUT_RING(ring, 0x00000000); /* RB_DEPTH_FLAG_BUFFER_BASE_HI */
- OUT_RING(ring, 0x00000000); /* RB_DEPTH_FLAG_BUFFER_PITCH */
- }
+ fd6_emit_flag_reference(ring, rsc,
+ zsbuf->u.tex.level, zsbuf->u.tex.first_layer);
if (rsc->lrz) {
- OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_BUFFER_BASE_LO, 5);
- OUT_RELOCW(ring, rsc->lrz, 0, 0, 0);
- OUT_RING(ring, A6XX_GRAS_LRZ_BUFFER_PITCH_PITCH(rsc->lrz_pitch));
- //OUT_RELOCW(ring, rsc->lrz, 0, 0, 0); /* GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_LO/HI */
- // XXX a6xx seems to use a different buffer here.. not sure what for..
- OUT_RING(ring, 0x00000000);
- OUT_RING(ring, 0x00000000);
+ OUT_REG(ring,
+ A6XX_GRAS_LRZ_BUFFER_BASE(.bo = rsc->lrz),
+ A6XX_GRAS_LRZ_BUFFER_PITCH(.pitch = rsc->lrz_pitch),
+ // XXX a6xx seems to use a different buffer here.. not sure what for..
+ A6XX_GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_LO(0),
+ A6XX_GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_HI(0));
} else {
OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_BUFFER_BASE_LO, 5);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(UNK_25));
if (rsc->stencil) {
- struct fd_resource_slice *slice = fd_resource_slice(rsc->stencil, 0);
- stride = slice->pitch * rsc->stencil->cpp;
+ struct fdl_slice *slice = fd_resource_slice(rsc->stencil, 0);
+ stride = slice->pitch;
size = slice->size0;
uint32_t base = gmem ? gmem->zsbuf_base[1] : 0;
- OUT_PKT4(ring, REG_A6XX_RB_STENCIL_INFO, 6);
- OUT_RING(ring, A6XX_RB_STENCIL_INFO_SEPARATE_STENCIL);
- OUT_RING(ring, A6XX_RB_STENCIL_BUFFER_PITCH(stride));
- OUT_RING(ring, A6XX_RB_STENCIL_BUFFER_ARRAY_PITCH(size));
- OUT_RELOCW(ring, rsc->stencil->bo, 0, 0, 0); /* RB_STENCIL_BASE_LO/HI */
- OUT_RING(ring, base); /* RB_STENCIL_BASE_LO */
+ OUT_REG(ring,
+ A6XX_RB_STENCIL_INFO(.separate_stencil = true),
+ A6XX_RB_STENCIL_BUFFER_PITCH(.a6xx_rb_stencil_buffer_pitch = stride),
+ A6XX_RB_STENCIL_BUFFER_ARRAY_PITCH(.a6xx_rb_stencil_buffer_array_pitch = size),
+ A6XX_RB_STENCIL_BUFFER_BASE(.bo = rsc->stencil->bo),
+ A6XX_RB_STENCIL_BUFFER_BASE_GMEM(.dword = base));
} else {
- OUT_PKT4(ring, REG_A6XX_RB_STENCIL_INFO, 1);
- OUT_RING(ring, 0x00000000); /* RB_STENCIL_INFO */
+ OUT_REG(ring, A6XX_RB_STENCIL_INFO(0));
}
} else {
OUT_PKT4(ring, REG_A6XX_RB_DEPTH_BUFFER_INFO, 6);
OUT_RING(ring, 0x00000000); /* RB_DEPTH_BUFFER_BASE_HI */
OUT_RING(ring, 0x00000000); /* RB_DEPTH_BUFFER_BASE_GMEM */
- OUT_PKT4(ring, REG_A6XX_GRAS_SU_DEPTH_BUFFER_INFO, 1);
- OUT_RING(ring, A6XX_GRAS_SU_DEPTH_BUFFER_INFO_DEPTH_FORMAT(DEPTH6_NONE));
+ OUT_REG(ring, A6XX_GRAS_SU_DEPTH_BUFFER_INFO(.depth_format = DEPTH6_NONE));
OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_BUFFER_BASE_LO, 5);
OUT_RING(ring, 0x00000000); /* RB_DEPTH_FLAG_BUFFER_BASE_LO */
OUT_RING(ring, 0x00000000); /* GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_LO */
OUT_RING(ring, 0x00000000); /* GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_HI */
- OUT_PKT4(ring, REG_A6XX_RB_STENCIL_INFO, 1);
- OUT_RING(ring, 0x00000000); /* RB_STENCIL_INFO */
+ OUT_REG(ring, A6XX_RB_STENCIL_INFO(0));
}
}
static bool
use_hw_binning(struct fd_batch *batch)
{
- struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
// TODO figure out hw limits for binning
- return fd_binning_enabled && ((gmem->nbins_x * gmem->nbins_y) > 2) &&
+ return fd_binning_enabled && ((gmem->nbins_x * gmem->nbins_y) >= 2) &&
(batch->num_draws > 0);
}
static void
patch_fb_read(struct fd_batch *batch)
{
- struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
for (unsigned i = 0; i < fd_patch_num_elements(&batch->fb_read_patches); i++) {
struct fd_cs_patch *patch = fd_patch_element(&batch->fb_read_patches, i);
cntl |= A6XX_RB_RENDER_CNTL_BINNING;
OUT_PKT7(ring, CP_REG_WRITE, 3);
- OUT_RING(ring, 0x2);
+ OUT_RING(ring, CP_REG_WRITE_0_TRACKER(TRACK_RENDER_CNTL));
OUT_RING(ring, REG_A6XX_RB_RENDER_CNTL);
OUT_RING(ring, cntl |
COND(depth_ubwc_enable, A6XX_RB_RENDER_CNTL_FLAG_DEPTH) |
A6XX_RB_RENDER_CNTL_FLAG_MRTS(mrts_ubwc_enable));
}
-#define VSC_DATA_SIZE(pitch) ((pitch) * 32 + 0x100) /* extra size to store VSC_SIZE */
-#define VSC_DATA2_SIZE(pitch) ((pitch) * 32)
+/* extra size to store VSC_DRAW_STRM_SIZE: */
+#define VSC_DRAW_STRM_SIZE(pitch) ((pitch) * 32 + 0x100)
+#define VSC_PRIM_STRM_SIZE(pitch) ((pitch) * 32)
static void
update_vsc_pipe(struct fd_batch *batch)
{
struct fd_context *ctx = batch->ctx;
struct fd6_context *fd6_ctx = fd6_context(ctx);
- struct fd_gmem_stateobj *gmem = &ctx->gmem;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
struct fd_ringbuffer *ring = batch->gmem;
int i;
+ if (batch->draw_strm_bits/8 > fd6_ctx->vsc_draw_strm_pitch) {
+ if (fd6_ctx->vsc_draw_strm)
+ fd_bo_del(fd6_ctx->vsc_draw_strm);
+ fd6_ctx->vsc_draw_strm = NULL;
+ /* Note: probably only need to align to 0x40, but aligning stronger
+ * reduces the odds that we will have to realloc again on the next
+ * frame:
+ */
+ fd6_ctx->vsc_draw_strm_pitch = align(batch->draw_strm_bits/8, 0x4000);
+ debug_printf("pre-resize VSC_DRAW_STRM_PITCH to: 0x%x\n",
+ fd6_ctx->vsc_draw_strm_pitch);
+ }
- if (!fd6_ctx->vsc_data) {
- fd6_ctx->vsc_data = fd_bo_new(ctx->screen->dev,
- VSC_DATA_SIZE(fd6_ctx->vsc_data_pitch),
- DRM_FREEDRENO_GEM_TYPE_KMEM, "vsc_data");
+ if (batch->prim_strm_bits/8 > fd6_ctx->vsc_prim_strm_pitch) {
+ if (fd6_ctx->vsc_prim_strm)
+ fd_bo_del(fd6_ctx->vsc_prim_strm);
+ fd6_ctx->vsc_prim_strm = NULL;
+ fd6_ctx->vsc_prim_strm_pitch = align(batch->prim_strm_bits/8, 0x4000);
+ debug_printf("pre-resize VSC_PRIM_STRM_PITCH to: 0x%x\n",
+ fd6_ctx->vsc_prim_strm_pitch);
}
- if (!fd6_ctx->vsc_data2) {
- fd6_ctx->vsc_data2 = fd_bo_new(ctx->screen->dev,
- VSC_DATA2_SIZE(fd6_ctx->vsc_data2_pitch),
- DRM_FREEDRENO_GEM_TYPE_KMEM, "vsc_data2");
+ if (!fd6_ctx->vsc_draw_strm) {
+ fd6_ctx->vsc_draw_strm = fd_bo_new(ctx->screen->dev,
+ VSC_DRAW_STRM_SIZE(fd6_ctx->vsc_draw_strm_pitch),
+ DRM_FREEDRENO_GEM_TYPE_KMEM, "vsc_draw_strm");
}
- OUT_PKT4(ring, REG_A6XX_VSC_BIN_SIZE, 3);
- OUT_RING(ring, A6XX_VSC_BIN_SIZE_WIDTH(gmem->bin_w) |
- A6XX_VSC_BIN_SIZE_HEIGHT(gmem->bin_h));
- OUT_RELOCW(ring, fd6_ctx->vsc_data,
- 32 * fd6_ctx->vsc_data_pitch, 0, 0); /* VSC_SIZE_ADDRESS_LO/HI */
+ if (!fd6_ctx->vsc_prim_strm) {
+ fd6_ctx->vsc_prim_strm = fd_bo_new(ctx->screen->dev,
+ VSC_PRIM_STRM_SIZE(fd6_ctx->vsc_prim_strm_pitch),
+ DRM_FREEDRENO_GEM_TYPE_KMEM, "vsc_prim_strm");
+ }
- OUT_PKT4(ring, REG_A6XX_VSC_BIN_COUNT, 1);
- OUT_RING(ring, A6XX_VSC_BIN_COUNT_NX(gmem->nbins_x) |
- A6XX_VSC_BIN_COUNT_NY(gmem->nbins_y));
+ OUT_REG(ring,
+ A6XX_VSC_BIN_SIZE(.width = gmem->bin_w, .height = gmem->bin_h),
+ A6XX_VSC_DRAW_STRM_SIZE_ADDRESS(
+ .bo = fd6_ctx->vsc_draw_strm,
+ .bo_offset = 32 * fd6_ctx->vsc_draw_strm_pitch));
+
+ OUT_REG(ring, A6XX_VSC_BIN_COUNT(.nx = gmem->nbins_x,
+ .ny = gmem->nbins_y));
OUT_PKT4(ring, REG_A6XX_VSC_PIPE_CONFIG_REG(0), 32);
for (i = 0; i < 32; i++) {
- struct fd_vsc_pipe *pipe = &ctx->vsc_pipe[i];
+ const struct fd_vsc_pipe *pipe = &gmem->vsc_pipe[i];
OUT_RING(ring, A6XX_VSC_PIPE_CONFIG_REG_X(pipe->x) |
A6XX_VSC_PIPE_CONFIG_REG_Y(pipe->y) |
A6XX_VSC_PIPE_CONFIG_REG_W(pipe->w) |
A6XX_VSC_PIPE_CONFIG_REG_H(pipe->h));
}
- OUT_PKT4(ring, REG_A6XX_VSC_PIPE_DATA2_ADDRESS_LO, 4);
- OUT_RELOCW(ring, fd6_ctx->vsc_data2, 0, 0, 0);
- OUT_RING(ring, fd6_ctx->vsc_data2_pitch);
- OUT_RING(ring, fd_bo_size(fd6_ctx->vsc_data2));
+ OUT_REG(ring,
+ A6XX_VSC_PRIM_STRM_ADDRESS(.bo = fd6_ctx->vsc_prim_strm),
+ A6XX_VSC_PRIM_STRM_PITCH(.dword = fd6_ctx->vsc_prim_strm_pitch),
+ A6XX_VSC_PRIM_STRM_ARRAY_PITCH(.dword = fd_bo_size(fd6_ctx->vsc_prim_strm)));
- OUT_PKT4(ring, REG_A6XX_VSC_PIPE_DATA_ADDRESS_LO, 4);
- OUT_RELOCW(ring, fd6_ctx->vsc_data, 0, 0, 0);
- OUT_RING(ring, fd6_ctx->vsc_data_pitch);
- OUT_RING(ring, fd_bo_size(fd6_ctx->vsc_data));
+ OUT_REG(ring,
+ A6XX_VSC_DRAW_STRM_ADDRESS(.bo = fd6_ctx->vsc_draw_strm),
+ A6XX_VSC_DRAW_STRM_PITCH(.dword = fd6_ctx->vsc_draw_strm_pitch),
+ A6XX_VSC_DRAW_STRM_ARRAY_PITCH(.dword = fd_bo_size(fd6_ctx->vsc_draw_strm)));
}
/* TODO we probably have more than 8 scratch regs.. although the first
#define OVERFLOW_FLAG_REG REG_A6XX_CP_SCRATCH_REG(0)
/*
- * If overflow is detected, either 0x1 (VSC_DATA overflow) or 0x3
- * (VSC_DATA2 overflow) plus the size of the overflowed buffer is
+ * If overflow is detected, either 0x1 (VSC_DRAW_STRM overflow) or 0x3
+ * (VSC_PRIM_STRM overflow) plus the size of the overflowed buffer is
* written to control->vsc_overflow. This allows the CPU to
* detect which buffer overflowed (and, since the current size is
* encoded as well, this protects against already-submitted but
emit_vsc_overflow_test(struct fd_batch *batch)
{
struct fd_ringbuffer *ring = batch->gmem;
- struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
struct fd6_context *fd6_ctx = fd6_context(batch->ctx);
- debug_assert((fd6_ctx->vsc_data_pitch & 0x3) == 0);
- debug_assert((fd6_ctx->vsc_data2_pitch & 0x3) == 0);
+ debug_assert((fd6_ctx->vsc_draw_strm_pitch & 0x3) == 0);
+ debug_assert((fd6_ctx->vsc_prim_strm_pitch & 0x3) == 0);
/* Clear vsc_scratch: */
OUT_PKT7(ring, CP_MEM_WRITE, 3);
- OUT_RELOCW(ring, control_ptr(fd6_ctx, vsc_scratch));
+ OUT_RELOC(ring, control_ptr(fd6_ctx, vsc_scratch));
OUT_RING(ring, 0x0);
/* Check for overflow, write vsc_scratch if detected: */
OUT_PKT7(ring, CP_COND_WRITE5, 8);
OUT_RING(ring, CP_COND_WRITE5_0_FUNCTION(WRITE_GE) |
CP_COND_WRITE5_0_WRITE_MEMORY);
- OUT_RING(ring, CP_COND_WRITE5_1_POLL_ADDR_LO(REG_A6XX_VSC_SIZE_REG(i)));
+ OUT_RING(ring, CP_COND_WRITE5_1_POLL_ADDR_LO(REG_A6XX_VSC_DRAW_STRM_SIZE_REG(i)));
OUT_RING(ring, CP_COND_WRITE5_2_POLL_ADDR_HI(0));
- OUT_RING(ring, CP_COND_WRITE5_3_REF(fd6_ctx->vsc_data_pitch));
+ OUT_RING(ring, CP_COND_WRITE5_3_REF(fd6_ctx->vsc_draw_strm_pitch));
OUT_RING(ring, CP_COND_WRITE5_4_MASK(~0));
- OUT_RELOCW(ring, control_ptr(fd6_ctx, vsc_scratch)); /* WRITE_ADDR_LO/HI */
- OUT_RING(ring, CP_COND_WRITE5_7_WRITE_DATA(1 + fd6_ctx->vsc_data_pitch));
+ OUT_RELOC(ring, control_ptr(fd6_ctx, vsc_scratch)); /* WRITE_ADDR_LO/HI */
+ OUT_RING(ring, CP_COND_WRITE5_7_WRITE_DATA(1 + fd6_ctx->vsc_draw_strm_pitch));
OUT_PKT7(ring, CP_COND_WRITE5, 8);
OUT_RING(ring, CP_COND_WRITE5_0_FUNCTION(WRITE_GE) |
CP_COND_WRITE5_0_WRITE_MEMORY);
- OUT_RING(ring, CP_COND_WRITE5_1_POLL_ADDR_LO(REG_A6XX_VSC_SIZE2_REG(i)));
+ OUT_RING(ring, CP_COND_WRITE5_1_POLL_ADDR_LO(REG_A6XX_VSC_PRIM_STRM_SIZE_REG(i)));
OUT_RING(ring, CP_COND_WRITE5_2_POLL_ADDR_HI(0));
- OUT_RING(ring, CP_COND_WRITE5_3_REF(fd6_ctx->vsc_data2_pitch));
+ OUT_RING(ring, CP_COND_WRITE5_3_REF(fd6_ctx->vsc_prim_strm_pitch));
OUT_RING(ring, CP_COND_WRITE5_4_MASK(~0));
- OUT_RELOCW(ring, control_ptr(fd6_ctx, vsc_scratch)); /* WRITE_ADDR_LO/HI */
- OUT_RING(ring, CP_COND_WRITE5_7_WRITE_DATA(3 + fd6_ctx->vsc_data2_pitch));
+ OUT_RELOC(ring, control_ptr(fd6_ctx, vsc_scratch)); /* WRITE_ADDR_LO/HI */
+ OUT_RING(ring, CP_COND_WRITE5_7_WRITE_DATA(3 + fd6_ctx->vsc_prim_strm_pitch));
}
OUT_PKT7(ring, CP_WAIT_MEM_WRITES, 0);
OUT_PKT7(ring, CP_MEM_TO_REG, 3);
OUT_RING(ring, CP_MEM_TO_REG_0_REG(OVERFLOW_FLAG_REG) |
- CP_MEM_TO_REG_0_CNT(1 - 1));
+ CP_MEM_TO_REG_0_CNT(0));
OUT_RELOC(ring, control_ptr(fd6_ctx, vsc_scratch)); /* SRC_LO/HI */
/*
BEGIN_RING(ring, 10); /* ensure if/else doesn't get split */
- /* b0 will be set if VSC_DATA or VSC_DATA2 overflow: */
+ /* b0 will be set if VSC_DRAW_STRM or VSC_PRIM_STRM overflow: */
OUT_PKT7(ring, CP_REG_TEST, 1);
OUT_RING(ring, A6XX_CP_REG_TEST_0_REG(OVERFLOW_FLAG_REG) |
A6XX_CP_REG_TEST_0_BIT(0) |
- A6XX_CP_REG_TEST_0_UNK25);
+ A6XX_CP_REG_TEST_0_WAIT_FOR_ME);
OUT_PKT7(ring, CP_COND_REG_EXEC, 2);
- OUT_RING(ring, 0x10000000);
- OUT_RING(ring, 7); /* conditionally execute next 7 dwords */
+ OUT_RING(ring, CP_COND_REG_EXEC_0_MODE(PRED_TEST));
+ OUT_RING(ring, CP_COND_REG_EXEC_1_DWORDS(7));
/* if (b0 set) */ {
/*
OUT_PKT7(ring, CP_REG_TO_MEM, 3);
OUT_RING(ring, CP_REG_TO_MEM_0_REG(OVERFLOW_FLAG_REG) |
CP_REG_TO_MEM_0_CNT(1 - 1));
- OUT_RELOCW(ring, control_ptr(fd6_ctx, vsc_overflow));
+ OUT_RELOC(ring, control_ptr(fd6_ctx, vsc_overflow));
OUT_PKT4(ring, OVERFLOW_FLAG_REG, 1);
OUT_RING(ring, 0x0);
unsigned size = vsc_overflow & ~0x3;
if (buffer == 0x1) {
- /* VSC_PIPE_DATA overflow: */
+ /* VSC_DRAW_STRM overflow: */
- if (size < fd6_ctx->vsc_data_pitch) {
+ if (size < fd6_ctx->vsc_draw_strm_pitch) {
/* we've already increased the size, this overflow is
* from a batch submitted before resize, but executed
* after
return;
}
- fd_bo_del(fd6_ctx->vsc_data);
- fd6_ctx->vsc_data = NULL;
- fd6_ctx->vsc_data_pitch *= 2;
+ fd_bo_del(fd6_ctx->vsc_draw_strm);
+ fd6_ctx->vsc_draw_strm = NULL;
+ fd6_ctx->vsc_draw_strm_pitch *= 2;
- debug_printf("resized VSC_DATA_PITCH to: 0x%x\n", fd6_ctx->vsc_data_pitch);
+ debug_printf("resized VSC_DRAW_STRM_PITCH to: 0x%x\n",
+ fd6_ctx->vsc_draw_strm_pitch);
} else if (buffer == 0x3) {
- /* VSC_PIPE_DATA2 overflow: */
+ /* VSC_PRIM_STRM overflow: */
- if (size < fd6_ctx->vsc_data2_pitch) {
+ if (size < fd6_ctx->vsc_prim_strm_pitch) {
/* we've already increased the size */
return;
}
- fd_bo_del(fd6_ctx->vsc_data2);
- fd6_ctx->vsc_data2 = NULL;
- fd6_ctx->vsc_data2_pitch *= 2;
+ fd_bo_del(fd6_ctx->vsc_prim_strm);
+ fd6_ctx->vsc_prim_strm = NULL;
+ fd6_ctx->vsc_prim_strm_pitch *= 2;
- debug_printf("resized VSC_DATA2_PITCH to: 0x%x\n", fd6_ctx->vsc_data2_pitch);
+ debug_printf("resized VSC_PRIM_STRM_PITCH to: 0x%x\n",
+ fd6_ctx->vsc_prim_strm_pitch);
} else {
/* NOTE: it's possible, for example, for overflow to corrupt the
* is skipped for tiles that have no visible geometry.
*/
static void
-emit_conditional_ib(struct fd_batch *batch, struct fd_tile *tile,
+emit_conditional_ib(struct fd_batch *batch, const struct fd_tile *tile,
struct fd_ringbuffer *target)
{
struct fd_ringbuffer *ring = batch->gmem;
OUT_PKT7(ring, CP_REG_TEST, 1);
OUT_RING(ring, A6XX_CP_REG_TEST_0_REG(REG_A6XX_VSC_STATE_REG(tile->p)) |
A6XX_CP_REG_TEST_0_BIT(tile->n) |
- A6XX_CP_REG_TEST_0_UNK25);
+ A6XX_CP_REG_TEST_0_WAIT_FOR_ME);
OUT_PKT7(ring, CP_COND_REG_EXEC, 2);
- OUT_RING(ring, 0x10000000);
- OUT_RING(ring, 4 * count); /* conditionally execute next 4*count dwords */
+ OUT_RING(ring, CP_COND_REG_EXEC_0_MODE(PRED_TEST));
+ OUT_RING(ring, CP_COND_REG_EXEC_1_DWORDS(4 * count));
for (unsigned i = 0; i < count; i++) {
uint32_t dwords;
static void
set_scissor(struct fd_ringbuffer *ring, uint32_t x1, uint32_t y1, uint32_t x2, uint32_t y2)
{
- OUT_PKT4(ring, REG_A6XX_GRAS_SC_WINDOW_SCISSOR_TL, 2);
- OUT_RING(ring, A6XX_GRAS_SC_WINDOW_SCISSOR_TL_X(x1) |
- A6XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(y1));
- OUT_RING(ring, A6XX_GRAS_SC_WINDOW_SCISSOR_BR_X(x2) |
- A6XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(y2));
-
- OUT_PKT4(ring, REG_A6XX_GRAS_RESOLVE_CNTL_1, 2);
- OUT_RING(ring, A6XX_GRAS_RESOLVE_CNTL_1_X(x1) |
- A6XX_GRAS_RESOLVE_CNTL_1_Y(y1));
- OUT_RING(ring, A6XX_GRAS_RESOLVE_CNTL_2_X(x2) |
- A6XX_GRAS_RESOLVE_CNTL_2_Y(y2));
+ OUT_REG(ring,
+ A6XX_GRAS_SC_WINDOW_SCISSOR_TL(.x = x1, .y = y1),
+ A6XX_GRAS_SC_WINDOW_SCISSOR_BR(.x = x2, .y = y2));
+
+ OUT_REG(ring,
+ A6XX_GRAS_RESOLVE_CNTL_1(.x = x1, .y = y1),
+ A6XX_GRAS_RESOLVE_CNTL_2(.x = x2, .y = y2));
}
static void
set_bin_size(struct fd_ringbuffer *ring, uint32_t w, uint32_t h, uint32_t flag)
{
- OUT_PKT4(ring, REG_A6XX_GRAS_BIN_CONTROL, 1);
- OUT_RING(ring, A6XX_GRAS_BIN_CONTROL_BINW(w) |
- A6XX_GRAS_BIN_CONTROL_BINH(h) | flag);
-
- OUT_PKT4(ring, REG_A6XX_RB_BIN_CONTROL, 1);
- OUT_RING(ring, A6XX_RB_BIN_CONTROL_BINW(w) |
- A6XX_RB_BIN_CONTROL_BINH(h) | flag);
-
+ OUT_REG(ring, A6XX_GRAS_BIN_CONTROL(.binw = w, .binh = h, .dword = flag));
+ OUT_REG(ring, A6XX_RB_BIN_CONTROL(.binw = w, .binh = h, .dword = flag));
/* no flag for RB_BIN_CONTROL2... */
- OUT_PKT4(ring, REG_A6XX_RB_BIN_CONTROL2, 1);
- OUT_RING(ring, A6XX_RB_BIN_CONTROL2_BINW(w) |
- A6XX_RB_BIN_CONTROL2_BINH(h));
+ OUT_REG(ring, A6XX_RB_BIN_CONTROL2(.binw = w, .binh = h));
}
static void
emit_binning_pass(struct fd_batch *batch)
{
struct fd_ringbuffer *ring = batch->gmem;
- struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
struct fd6_context *fd6_ctx = fd6_context(batch->ctx);
- uint32_t x1 = gmem->minx;
- uint32_t y1 = gmem->miny;
- uint32_t x2 = gmem->minx + gmem->width - 1;
- uint32_t y2 = gmem->miny + gmem->height - 1;
-
debug_assert(!batch->tessellation);
- set_scissor(ring, x1, y1, x2, y2);
+ set_scissor(ring, 0, 0, gmem->width - 1, gmem->height - 1);
emit_marker6(ring, 7);
OUT_PKT7(ring, CP_SET_MARKER, 1);
OUT_WFI5(ring);
- OUT_PKT4(ring, REG_A6XX_VFD_MODE_CNTL, 1);
- OUT_RING(ring, A6XX_VFD_MODE_CNTL_BINNING_PASS);
+ OUT_REG(ring, A6XX_VFD_MODE_CNTL(.binning_pass = true));
update_vsc_pipe(batch);
A6XX_SP_TP_WINDOW_OFFSET_Y(0));
/* emit IB to binning drawcmds: */
+ fd_log(batch, "GMEM: START BINNING IB");
fd6_emit_ib(ring, batch->draw);
+ fd_log(batch, "GMEM: END BINNING IB");
fd_reset_wfi(batch);
OUT_PKT7(ring, CP_WAIT_FOR_ME, 0);
+ fd_log(batch, "START VSC OVERFLOW TEST");
emit_vsc_overflow_test(batch);
+ fd_log(batch, "END VSC OVERFLOW TEST");
OUT_PKT7(ring, CP_SET_VISIBILITY_OVERRIDE, 1);
OUT_RING(ring, 0x0);
struct fd_context *ctx = batch->ctx;
struct fd_ringbuffer *ring = batch->gmem;
struct pipe_framebuffer_state *pfb = &batch->framebuffer;
- struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
fd6_emit_restore(batch, ring);
fd6_emit_lrz_flush(ring);
- if (batch->lrz_clear)
+ if (batch->lrz_clear) {
+ fd_log(batch, "START LRZ CLEAR");
fd6_emit_ib(ring, batch->lrz_clear);
+ fd_log(batch, "END LRZ CLEAR");
+ }
fd6_cache_inv(batch, ring);
OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
OUT_RING(ring, 0x0);
+ /* blob controls "local" in IB2, but I think that is not required */
+ OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_LOCAL, 1);
+ OUT_RING(ring, 0x1);
+
fd_wfi(batch, ring);
OUT_PKT4(ring, REG_A6XX_RB_CCU_CNTL, 1);
OUT_RING(ring, fd6_context(ctx)->magic.RB_CCU_CNTL_gmem);
- emit_zs(ring, pfb->zsbuf, &ctx->gmem);
- emit_mrt(ring, pfb, &ctx->gmem);
+ emit_zs(ring, pfb->zsbuf, batch->gmem_state);
+ emit_mrt(ring, pfb, batch->gmem_state);
emit_msaa(ring, pfb->samples);
patch_fb_read(batch);
/* before mem2gmem */
static void
-fd6_emit_tile_prep(struct fd_batch *batch, struct fd_tile *tile)
+fd6_emit_tile_prep(struct fd_batch *batch, const struct fd_tile *tile)
{
struct fd_context *ctx = batch->ctx;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
struct fd6_context *fd6_ctx = fd6_context(ctx);
struct fd_ringbuffer *ring = batch->gmem;
emit_marker6(ring, 7);
OUT_PKT7(ring, CP_SET_MARKER, 1);
- OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_GMEM) | 0x10);
+ OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_GMEM));
emit_marker6(ring, 7);
uint32_t x1 = tile->xoff;
set_scissor(ring, x1, y1, x2, y2);
if (use_hw_binning(batch)) {
- struct fd_vsc_pipe *pipe = &ctx->vsc_pipe[tile->p];
+ const struct fd_vsc_pipe *pipe = &gmem->vsc_pipe[tile->p];
OUT_PKT7(ring, CP_WAIT_FOR_ME, 0);
OUT_PKT7(ring, CP_REG_TEST, 1);
OUT_RING(ring, A6XX_CP_REG_TEST_0_REG(OVERFLOW_FLAG_REG) |
A6XX_CP_REG_TEST_0_BIT(0) |
- A6XX_CP_REG_TEST_0_UNK25);
+ A6XX_CP_REG_TEST_0_WAIT_FOR_ME);
OUT_PKT7(ring, CP_COND_REG_EXEC, 2);
- OUT_RING(ring, 0x10000000);
- OUT_RING(ring, 11); /* conditionally execute next 11 dwords */
+ OUT_RING(ring, CP_COND_REG_EXEC_0_MODE(PRED_TEST));
+ OUT_RING(ring, CP_COND_REG_EXEC_1_DWORDS(11));
/* if (no overflow) */ {
OUT_PKT7(ring, CP_SET_BIN_DATA5, 7);
OUT_RING(ring, CP_SET_BIN_DATA5_0_VSC_SIZE(pipe->w * pipe->h) |
CP_SET_BIN_DATA5_0_VSC_N(tile->n));
- OUT_RELOC(ring, fd6_ctx->vsc_data, /* VSC_PIPE[p].DATA_ADDRESS */
- (tile->p * fd6_ctx->vsc_data_pitch), 0, 0);
- OUT_RELOC(ring, fd6_ctx->vsc_data, /* VSC_SIZE_ADDRESS + (p * 4) */
- (tile->p * 4) + (32 * fd6_ctx->vsc_data_pitch), 0, 0);
- OUT_RELOC(ring, fd6_ctx->vsc_data2,
- (tile->p * fd6_ctx->vsc_data2_pitch), 0, 0);
+ OUT_RELOC(ring, fd6_ctx->vsc_draw_strm, /* per-pipe draw-stream address */
+ (tile->p * fd6_ctx->vsc_draw_strm_pitch), 0, 0);
+ OUT_RELOC(ring, fd6_ctx->vsc_draw_strm, /* VSC_DRAW_STRM_ADDRESS + (p * 4) */
+ (tile->p * 4) + (32 * fd6_ctx->vsc_draw_strm_pitch), 0, 0);
+ OUT_RELOC(ring, fd6_ctx->vsc_prim_strm,
+ (tile->p * fd6_ctx->vsc_prim_strm_pitch), 0, 0);
OUT_PKT7(ring, CP_SET_VISIBILITY_OVERRIDE, 1);
OUT_RING(ring, 0x0);
set_window_offset(ring, x1, y1);
- struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
set_bin_size(ring, gmem->bin_w, gmem->bin_h, 0x6000000);
OUT_PKT7(ring, CP_SET_MODE, 1);
OUT_RING(ring, 0x0);
-
- OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8804, 1);
- OUT_RING(ring, 0x0);
-
- OUT_PKT4(ring, REG_A6XX_SP_TP_UNKNOWN_B304, 1);
- OUT_RING(ring, 0x0);
-
- OUT_PKT4(ring, REG_A6XX_GRAS_UNKNOWN_80A4, 1);
- OUT_RING(ring, 0x0);
} else {
set_window_offset(ring, x1, y1);
static void
set_blit_scissor(struct fd_batch *batch, struct fd_ringbuffer *ring)
{
- struct pipe_scissor_state blit_scissor;
- struct pipe_framebuffer_state *pfb = &batch->framebuffer;
+ struct pipe_scissor_state blit_scissor = batch->max_scissor;
- blit_scissor.minx = 0;
- blit_scissor.miny = 0;
- blit_scissor.maxx = align(pfb->width, batch->ctx->screen->gmem_alignw);
- blit_scissor.maxy = align(pfb->height, batch->ctx->screen->gmem_alignh);
+ blit_scissor.minx = ROUND_DOWN_TO(blit_scissor.minx, 16);
+ blit_scissor.miny = ROUND_DOWN_TO(blit_scissor.miny, 4);
+ blit_scissor.maxx = ALIGN(blit_scissor.maxx, 16);
+ blit_scissor.maxy = ALIGN(blit_scissor.maxy, 4);
OUT_PKT4(ring, REG_A6XX_RB_BLIT_SCISSOR_TL, 2);
OUT_RING(ring,
struct pipe_surface *psurf,
bool stencil)
{
- struct fd_resource_slice *slice;
+ struct fdl_slice *slice;
struct fd_resource *rsc = fd_resource(psurf->texture);
enum pipe_format pfmt = psurf->format;
- uint32_t offset, ubwc_offset;
+ uint32_t offset;
bool ubwc_enabled;
debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
offset = fd_resource_offset(rsc, psurf->u.tex.level,
psurf->u.tex.first_layer);
ubwc_enabled = fd_resource_ubwc_enabled(rsc, psurf->u.tex.level);
- ubwc_offset = fd_resource_ubwc_offset(rsc, psurf->u.tex.level,
- psurf->u.tex.first_layer);
debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
- enum a6xx_color_fmt format = fd6_pipe2color(pfmt);
- uint32_t stride = slice->pitch * rsc->cpp;
+ enum a6xx_format format = fd6_pipe2color(pfmt);
+ uint32_t stride = slice->pitch;
uint32_t size = slice->size0;
- enum a3xx_color_swap swap = rsc->tile_mode ? WZYX : fd6_pipe2swap(pfmt);
+ enum a3xx_color_swap swap = fd6_resource_swap(rsc, pfmt);
enum a3xx_msaa_samples samples =
fd_msaa_samples(rsc->base.nr_samples);
- uint32_t tile_mode;
+ uint32_t tile_mode = fd_resource_tile_mode(&rsc->base, psurf->u.tex.level);
- if (rsc->tile_mode &&
- fd_resource_level_linear(&rsc->base, psurf->u.tex.level))
- tile_mode = TILE6_LINEAR;
- else
- tile_mode = rsc->tile_mode;
+ OUT_REG(ring,
+ A6XX_RB_BLIT_DST_INFO(.tile_mode = tile_mode, .samples = samples,
+ .color_format = format, .color_swap = swap, .flags = ubwc_enabled),
+ A6XX_RB_BLIT_DST(.bo = rsc->bo, .bo_offset = offset),
+ A6XX_RB_BLIT_DST_PITCH(.a6xx_rb_blit_dst_pitch = stride),
+ A6XX_RB_BLIT_DST_ARRAY_PITCH(.a6xx_rb_blit_dst_array_pitch = size));
- OUT_PKT4(ring, REG_A6XX_RB_BLIT_DST_INFO, 5);
- OUT_RING(ring,
- A6XX_RB_BLIT_DST_INFO_TILE_MODE(tile_mode) |
- A6XX_RB_BLIT_DST_INFO_SAMPLES(samples) |
- A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(format) |
- A6XX_RB_BLIT_DST_INFO_COLOR_SWAP(swap) |
- COND(ubwc_enabled, A6XX_RB_BLIT_DST_INFO_FLAGS));
- OUT_RELOCW(ring, rsc->bo, offset, 0, 0); /* RB_BLIT_DST_LO/HI */
- OUT_RING(ring, A6XX_RB_BLIT_DST_PITCH(stride));
- OUT_RING(ring, A6XX_RB_BLIT_DST_ARRAY_PITCH(size));
-
- OUT_PKT4(ring, REG_A6XX_RB_BLIT_BASE_GMEM, 1);
- OUT_RING(ring, base);
+ OUT_REG(ring, A6XX_RB_BLIT_BASE_GMEM(.dword = base));
if (ubwc_enabled) {
OUT_PKT4(ring, REG_A6XX_RB_BLIT_FLAG_DST_LO, 3);
- OUT_RELOCW(ring, rsc->bo, ubwc_offset, 0, 0);
- OUT_RING(ring, A6XX_RB_BLIT_FLAG_DST_PITCH_PITCH(rsc->ubwc_pitch) |
- A6XX_RB_BLIT_FLAG_DST_PITCH_ARRAY_PITCH(rsc->ubwc_size));
+ fd6_emit_flag_reference(ring, rsc,
+ psurf->u.tex.level, psurf->u.tex.first_layer);
}
fd6_emit_blit(batch, ring);
struct pipe_surface *psurf,
unsigned buffer)
{
- uint32_t info = 0;
- bool stencil = false;
+ bool stencil = (buffer == FD_BUFFER_STENCIL);
- switch (buffer) {
- case FD_BUFFER_COLOR:
- info |= A6XX_RB_BLIT_INFO_UNK0;
- break;
- case FD_BUFFER_STENCIL:
- info |= A6XX_RB_BLIT_INFO_UNK0;
- stencil = true;
- break;
- case FD_BUFFER_DEPTH:
- info |= A6XX_RB_BLIT_INFO_DEPTH | A6XX_RB_BLIT_INFO_UNK0;
- break;
- }
-
- if (util_format_is_pure_integer(psurf->format))
- info |= A6XX_RB_BLIT_INFO_INTEGER;
-
- OUT_PKT4(ring, REG_A6XX_RB_BLIT_INFO, 1);
- OUT_RING(ring, info | A6XX_RB_BLIT_INFO_GMEM);
+ OUT_REG(ring, A6XX_RB_BLIT_INFO(
+ .gmem = true, .unk0 = true,
+ .depth = (buffer == FD_BUFFER_DEPTH),
+ .integer = util_format_is_pure_integer(psurf->format)));
emit_blit(batch, ring, base, psurf, stencil);
}
emit_clears(struct fd_batch *batch, struct fd_ringbuffer *ring)
{
struct pipe_framebuffer_state *pfb = &batch->framebuffer;
- struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
enum a3xx_msaa_samples samples = fd_msaa_samples(pfb->samples);
uint32_t buffers = batch->fast_cleared;
break;
}
- if (util_format_is_pure_uint(pfmt)) {
- util_format_write_4ui(pfmt, swapped.ui, 0, &uc, 0, 0, 0, 1, 1);
- } else if (util_format_is_pure_sint(pfmt)) {
- util_format_write_4i(pfmt, swapped.i, 0, &uc, 0, 0, 0, 1, 1);
- } else {
- util_pack_color(swapped.f, pfmt, &uc);
- }
+ util_pack_color_union(pfmt, &uc, &swapped);
OUT_PKT4(ring, REG_A6XX_RB_BLIT_DST_INFO, 1);
OUT_RING(ring, A6XX_RB_BLIT_DST_INFO_TILE_MODE(TILE6_LINEAR) |
OUT_PKT4(ring, REG_A6XX_RB_BLIT_DST_INFO, 1);
OUT_RING(ring, A6XX_RB_BLIT_DST_INFO_TILE_MODE(TILE6_LINEAR) |
A6XX_RB_BLIT_DST_INFO_SAMPLES(samples) |
- A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(RB6_R8_UINT));
+ A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(FMT6_8_UINT));
OUT_PKT4(ring, REG_A6XX_RB_BLIT_INFO, 1);
OUT_RING(ring, A6XX_RB_BLIT_INFO_GMEM |
static void
emit_restore_blits(struct fd_batch *batch, struct fd_ringbuffer *ring)
{
- struct fd_context *ctx = batch->ctx;
- struct fd_gmem_stateobj *gmem = &ctx->gmem;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
struct pipe_framebuffer_state *pfb = &batch->framebuffer;
if (batch->restore & FD_BUFFER_COLOR) {
* transfer from system memory to gmem
*/
static void
-fd6_emit_tile_mem2gmem(struct fd_batch *batch, struct fd_tile *tile)
+fd6_emit_tile_mem2gmem(struct fd_batch *batch, const struct fd_tile *tile)
{
}
/* before IB to rendering cmds: */
static void
-fd6_emit_tile_renderprep(struct fd_batch *batch, struct fd_tile *tile)
+fd6_emit_tile_renderprep(struct fd_batch *batch, const struct fd_tile *tile)
{
+ fd_log(batch, "TILE: START CLEAR/RESTORE");
if (batch->fast_cleared || !use_hw_binning(batch)) {
fd6_emit_ib(batch->gmem, batch->tile_setup);
} else {
emit_conditional_ib(batch, tile, batch->tile_setup);
}
+ fd_log(batch, "TILE: END CLEAR/RESTORE");
}
static void
static void
prepare_tile_fini_ib(struct fd_batch *batch)
{
- struct fd_context *ctx = batch->ctx;
- struct fd_gmem_stateobj *gmem = &ctx->gmem;
+ const struct fd_gmem_stateobj *gmem = batch->gmem_state;
struct pipe_framebuffer_state *pfb = &batch->framebuffer;
struct fd_ringbuffer *ring;
}
static void
-fd6_emit_tile(struct fd_batch *batch, struct fd_tile *tile)
+fd6_emit_tile(struct fd_batch *batch, const struct fd_tile *tile)
{
if (!use_hw_binning(batch)) {
fd6_emit_ib(batch->gmem, batch->draw);
}
static void
-fd6_emit_tile_gmem2mem(struct fd_batch *batch, struct fd_tile *tile)
+fd6_emit_tile_gmem2mem(struct fd_batch *batch, const struct fd_tile *tile)
{
struct fd_ringbuffer *ring = batch->gmem;
OUT_PKT7(ring, CP_REG_TEST, 1);
OUT_RING(ring, A6XX_CP_REG_TEST_0_REG(OVERFLOW_FLAG_REG) |
A6XX_CP_REG_TEST_0_BIT(0) |
- A6XX_CP_REG_TEST_0_UNK25);
+ A6XX_CP_REG_TEST_0_WAIT_FOR_ME);
OUT_PKT7(ring, CP_COND_REG_EXEC, 2);
- OUT_RING(ring, 0x10000000);
- OUT_RING(ring, 2); /* conditionally execute next 2 dwords */
+ OUT_RING(ring, CP_COND_REG_EXEC_0_MODE(PRED_TEST));
+ OUT_RING(ring, CP_COND_REG_EXEC_1_DWORDS(2));
/* if (no overflow) */ {
OUT_PKT7(ring, CP_SET_MARKER, 1);
- OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(0x5) | 0x10);
+ OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_ENDVIS));
}
}
emit_marker6(ring, 7);
OUT_PKT7(ring, CP_SET_MARKER, 1);
- OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_RESOLVE) | 0x10);
+ OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_RESOLVE));
emit_marker6(ring, 7);
+ fd_log(batch, "TILE: START RESOLVE");
if (batch->fast_cleared || !use_hw_binning(batch)) {
fd6_emit_ib(batch->gmem, batch->tile_fini);
} else {
emit_conditional_ib(batch, tile, batch->tile_fini);
}
-
- OUT_PKT7(ring, CP_SET_MARKER, 1);
- OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(0x7));
+ fd_log(batch, "TILE: END RESOLVE");
}
static void
{
struct fd_ringbuffer *ring = batch->gmem;
+ if (batch->epilogue)
+ fd6_emit_ib(batch->gmem, batch->epilogue);
+
OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_CNTL, 1);
- OUT_RING(ring, A6XX_GRAS_LRZ_CNTL_ENABLE | A6XX_GRAS_LRZ_CNTL_UNK3);
+ OUT_RING(ring, A6XX_GRAS_LRZ_CNTL_ENABLE);
fd6_emit_lrz_flush(ring);
- fd6_event_write(batch, ring, CACHE_FLUSH_TS, true);
+ fd6_event_write(batch, ring, PC_CCU_RESOLVE_TS, true);
if (use_hw_binning(batch)) {
check_vsc_overflow(batch->ctx);
}
}
- fd6_event_write(batch, ring, 0x1d, true);
+ fd6_event_write(batch, ring, PC_CCU_FLUSH_COLOR_TS, true);
}
static void
DRM_FREEDRENO_GEM_TYPE_KMEM, "tessparam");
OUT_PKT4(ring, REG_A6XX_PC_TESSFACTOR_ADDR_LO, 2);
- OUT_RELOCW(ring, batch->tessfactor_bo, 0, 0, 0);
+ OUT_RELOC(ring, batch->tessfactor_bo, 0, 0, 0);
batch->tess_addrs_constobj->cur = batch->tess_addrs_constobj->start;
- OUT_RELOCW(batch->tess_addrs_constobj, batch->tessparam_bo, 0, 0, 0);
- OUT_RELOCW(batch->tess_addrs_constobj, batch->tessfactor_bo, 0, 0, 0);
+ OUT_RELOC(batch->tess_addrs_constobj, batch->tessparam_bo, 0, 0, 0);
+ OUT_RELOC(batch->tess_addrs_constobj, batch->tessfactor_bo, 0, 0, 0);
}
static void
fd6_emit_restore(batch, ring);
- set_scissor(ring, 0, 0, pfb->width - 1, pfb->height - 1);
+ if (pfb->width > 0 && pfb->height > 0)
+ set_scissor(ring, 0, 0, pfb->width - 1, pfb->height - 1);
+ else
+ set_scissor(ring, 0, 0, 0, 0);
set_window_offset(ring, 0, 0);
fd6_emit_lrz_flush(ring);
+ if (batch->lrz_clear)
+ fd6_emit_ib(ring, batch->lrz_clear);
+
emit_marker6(ring, 7);
OUT_PKT7(ring, CP_SET_MARKER, 1);
- OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_BYPASS) | 0x10); /* | 0x10 ? */
+ OUT_RING(ring, A6XX_CP_SET_MARKER_0_MODE(RM6_BYPASS));
emit_marker6(ring, 7);
if (batch->tessellation)
OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
OUT_RING(ring, 0x0);
+ /* blob controls "local" in IB2, but I think that is not required */
+ OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_LOCAL, 1);
+ OUT_RING(ring, 0x1);
+
fd6_event_write(batch, ring, PC_CCU_INVALIDATE_COLOR, false);
fd6_cache_inv(batch, ring);
{
struct fd_ringbuffer *ring = batch->gmem;
+ if (batch->epilogue)
+ fd6_emit_ib(batch->gmem, batch->epilogue);
+
OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
OUT_RING(ring, 0x0);
fd6_emit_lrz_flush(ring);
- fd6_event_write(batch, ring, UNK_1D, true);
+ fd6_event_write(batch, ring, PC_CCU_FLUSH_COLOR_TS, true);
}
void