#include "util/u_format.h"
#include "freedreno_resource.h"
+#include "freedreno_query_hw.h"
#include "fd4_emit.h"
#include "fd4_blend.h"
#include "fd4_program.h"
#include "fd4_rasterizer.h"
#include "fd4_texture.h"
-#include "fd4_util.h"
+#include "fd4_format.h"
#include "fd4_zsa.h"
+static const enum adreno_state_block sb[] = {
+ [SHADER_VERTEX] = SB_VERT_SHADER,
+ [SHADER_FRAGMENT] = SB_FRAG_SHADER,
+};
+
/* regid: base const register
* prsc or dwords: buffer containing constant values
* sizedwords: size of const value buffer
*/
void
-fd4_emit_constant(struct fd_ringbuffer *ring,
- enum adreno_state_block sb,
+fd4_emit_const(struct fd_ringbuffer *ring, enum shader_t type,
uint32_t regid, uint32_t offset, uint32_t sizedwords,
const uint32_t *dwords, struct pipe_resource *prsc)
{
uint32_t i, sz;
enum adreno_state_src src;
+ debug_assert((regid % 4) == 0);
+ debug_assert((sizedwords % 4) == 0);
+
if (prsc) {
sz = 0;
src = 0x2; // TODO ??
OUT_PKT3(ring, CP_LOAD_STATE, 2 + sz);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/4) |
CP_LOAD_STATE_0_STATE_SRC(src) |
- CP_LOAD_STATE_0_STATE_BLOCK(sb) |
+ CP_LOAD_STATE_0_STATE_BLOCK(sb[type]) |
CP_LOAD_STATE_0_NUM_UNIT(sizedwords/4));
if (prsc) {
struct fd_bo *bo = fd_resource(prsc)->bo;
}
static void
-emit_constants(struct fd_ringbuffer *ring,
- enum adreno_state_block sb,
- struct fd_constbuf_stateobj *constbuf,
- struct ir3_shader_variant *shader)
+fd4_emit_const_bo(struct fd_ringbuffer *ring, enum shader_t type, boolean write,
+ uint32_t regid, uint32_t num, struct fd_bo **bos, uint32_t *offsets)
{
- uint32_t enabled_mask = constbuf->enabled_mask;
- uint32_t first_immediate;
- uint32_t base = 0;
-
- // XXX TODO only emit dirty consts.. but we need to keep track if
- // they are clobbered by a clear, gmem2mem, or mem2gmem..
- constbuf->dirty_mask = enabled_mask;
-
- /* in particular, with binning shader we may end up with unused
- * consts, ie. we could end up w/ constlen that is smaller
- * than first_immediate. In that case truncate the user consts
- * early to avoid HLSQ lockup caused by writing too many consts
- */
- first_immediate = MIN2(shader->first_immediate, shader->constlen);
+ uint32_t i;
- /* emit user constants: */
- while (enabled_mask) {
- unsigned index = ffs(enabled_mask) - 1;
- struct pipe_constant_buffer *cb = &constbuf->cb[index];
- unsigned size = align(cb->buffer_size, 4) / 4; /* size in dwords */
+ debug_assert((regid % 4) == 0);
+ debug_assert((num % 4) == 0);
- // I expect that size should be a multiple of vec4's:
- assert(size == align(size, 4));
-
- /* gallium could leave const buffers bound above what the
- * current shader uses.. don't let that confuse us.
- */
- if (base >= (4 * first_immediate))
- break;
-
- if (constbuf->dirty_mask & (1 << index)) {
- /* and even if the start of the const buffer is before
- * first_immediate, the end may not be:
- */
- size = MIN2(size, (4 * first_immediate) - base);
- fd4_emit_constant(ring, sb, base,
- cb->buffer_offset, size,
- cb->user_buffer, cb->buffer);
- constbuf->dirty_mask &= ~(1 << index);
- }
-
- base += size;
- enabled_mask &= ~(1 << index);
- }
-
- /* emit shader immediates: */
- if (shader) {
- int size = shader->immediates_count;
- base = shader->first_immediate;
-
- /* truncate size to avoid writing constants that shader
- * does not use:
- */
- size = MIN2(size + base, shader->constlen) - base;
-
- /* convert out of vec4: */
- base *= 4;
- size *= 4;
-
- if (size > 0) {
- fd4_emit_constant(ring, sb, base,
- 0, size, shader->immediates[0].val, NULL);
+ OUT_PKT3(ring, CP_LOAD_STATE, 2 + num);
+ OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/4) |
+ CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
+ CP_LOAD_STATE_0_STATE_BLOCK(sb[type]) |
+ CP_LOAD_STATE_0_NUM_UNIT(num/4));
+ OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
+ CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));
+
+ for (i = 0; i < num; i++) {
+ if (bos[i]) {
+ if (write) {
+ OUT_RELOCW(ring, bos[i], offsets[i], 0, 0);
+ } else {
+ OUT_RELOC(ring, bos[i], offsets[i], 0, 0);
+ }
+ } else {
+ OUT_RING(ring, 0xbad00000 | (i << 16));
}
}
}
emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
enum adreno_state_block sb, struct fd_texture_stateobj *tex)
{
- unsigned i;
+ static const uint32_t bcolor_reg[] = {
+ [SB_VERT_TEX] = REG_A4XX_TPL1_TP_VS_BORDER_COLOR_BASE_ADDR,
+ [SB_FRAG_TEX] = REG_A4XX_TPL1_TP_FS_BORDER_COLOR_BASE_ADDR,
+ };
+ struct fd4_context *fd4_ctx = fd4_context(ctx);
+ unsigned i, off;
+ void *ptr;
+
+ u_upload_alloc(fd4_ctx->border_color_uploader,
+ 0, BORDER_COLOR_UPLOAD_SIZE,
+ BORDER_COLOR_UPLOAD_SIZE, &off,
+ &fd4_ctx->border_color_buf,
+ &ptr);
+
+ fd_setup_border_colors(tex, ptr, 0);
if (tex->num_samplers > 0) {
+ int num_samplers;
+
+ /* not sure if this is an a420.0 workaround, but we seem
+ * to need to emit these in pairs.. emit a final dummy
+ * entry if odd # of samplers:
+ */
+ num_samplers = align(tex->num_samplers, 2);
+
/* output sampler state: */
- OUT_PKT3(ring, CP_LOAD_STATE, 2 + 2 + (2 * tex->num_samplers));
+ OUT_PKT3(ring, CP_LOAD_STATE, 2 + (2 * num_samplers));
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(sb) |
- CP_LOAD_STATE_0_NUM_UNIT(tex->num_samplers));
+ CP_LOAD_STATE_0_NUM_UNIT(num_samplers));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
for (i = 0; i < tex->num_samplers; i++) {
OUT_RING(ring, sampler->texsamp0);
OUT_RING(ring, sampler->texsamp1);
}
- /* maybe an a420.0 (or a4xx.0) workaround?? or just driver bug? */
- OUT_RING(ring, 0x00000000);
- OUT_RING(ring, 0x00000000);
+
+ for (; i < num_samplers; i++) {
+ OUT_RING(ring, 0x00000000);
+ OUT_RING(ring, 0x00000000);
+ }
}
if (tex->num_textures > 0) {
const struct fd4_pipe_sampler_view *view = tex->textures[i] ?
fd4_pipe_sampler_view(tex->textures[i]) :
&dummy_view;
- struct fd_resource *rsc = view->tex_resource;
- struct fd_resource_slice *slice = fd_resource_slice(rsc, 0);
+
OUT_RING(ring, view->texconst0);
OUT_RING(ring, view->texconst1);
OUT_RING(ring, view->texconst2);
OUT_RING(ring, view->texconst3);
- OUT_RELOC(ring, rsc->bo, slice->offset, 0, 0);
+ if (view->base.texture) {
+ struct fd_resource *rsc = fd_resource(view->base.texture);
+ OUT_RELOC(ring, rsc->bo, view->offset, view->texconst4, 0);
+ } else {
+ OUT_RING(ring, 0x00000000);
+ }
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
}
}
+
+ OUT_PKT0(ring, bcolor_reg[sb], 1);
+ OUT_RELOC(ring, fd_resource(fd4_ctx->border_color_buf)->bo, off, 0, 0);
+
+ u_upload_unmap(fd4_ctx->border_color_uploader);
}
/* emit texture state for mem->gmem restore operation.. eventually it would
* special cases..
*/
void
-fd4_emit_gmem_restore_tex(struct fd_ringbuffer *ring, struct pipe_surface *psurf)
+fd4_emit_gmem_restore_tex(struct fd_ringbuffer *ring, unsigned nr_bufs,
+ struct pipe_surface **bufs)
{
- struct fd_resource *rsc = fd_resource(psurf->texture);
- unsigned lvl = psurf->u.tex.level;
- struct fd_resource_slice *slice = &rsc->slices[lvl];
- uint32_t layer_offset = slice->size0 * psurf->u.tex.first_layer;
- enum pipe_format format = fd4_gmem_restore_format(psurf->format);
+ unsigned char mrt_comp[A4XX_MAX_RENDER_TARGETS];
+ int i;
- debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
+ for (i = 0; i < A4XX_MAX_RENDER_TARGETS; i++) {
+ mrt_comp[i] = (i < nr_bufs) ? 0xf : 0;
+ }
/* output sampler state: */
- OUT_PKT3(ring, CP_LOAD_STATE, 4);
+ OUT_PKT3(ring, CP_LOAD_STATE, 2 + (2 * nr_bufs));
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_TEX) |
- CP_LOAD_STATE_0_NUM_UNIT(1));
+ CP_LOAD_STATE_0_NUM_UNIT(nr_bufs));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
- OUT_RING(ring, A4XX_TEX_SAMP_0_XY_MAG(A4XX_TEX_NEAREST) |
- A4XX_TEX_SAMP_0_XY_MIN(A4XX_TEX_NEAREST) |
- A4XX_TEX_SAMP_0_WRAP_S(A4XX_TEX_CLAMP_TO_EDGE) |
- A4XX_TEX_SAMP_0_WRAP_T(A4XX_TEX_CLAMP_TO_EDGE) |
- A4XX_TEX_SAMP_0_WRAP_R(A4XX_TEX_REPEAT));
- OUT_RING(ring, 0x00000000);
+ for (i = 0; i < nr_bufs; i++) {
+ OUT_RING(ring, A4XX_TEX_SAMP_0_XY_MAG(A4XX_TEX_NEAREST) |
+ A4XX_TEX_SAMP_0_XY_MIN(A4XX_TEX_NEAREST) |
+ A4XX_TEX_SAMP_0_WRAP_S(A4XX_TEX_CLAMP_TO_EDGE) |
+ A4XX_TEX_SAMP_0_WRAP_T(A4XX_TEX_CLAMP_TO_EDGE) |
+ A4XX_TEX_SAMP_0_WRAP_R(A4XX_TEX_REPEAT));
+ OUT_RING(ring, 0x00000000);
+ }
/* emit texture state: */
- OUT_PKT3(ring, CP_LOAD_STATE, 10);
+ OUT_PKT3(ring, CP_LOAD_STATE, 2 + (8 * nr_bufs));
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_TEX) |
- CP_LOAD_STATE_0_NUM_UNIT(1));
+ CP_LOAD_STATE_0_NUM_UNIT(nr_bufs));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
- OUT_RING(ring, A4XX_TEX_CONST_0_FMT(fd4_pipe2tex(format)) |
- A4XX_TEX_CONST_0_TYPE(A4XX_TEX_2D) |
- fd4_tex_swiz(format, PIPE_SWIZZLE_RED, PIPE_SWIZZLE_GREEN,
- PIPE_SWIZZLE_BLUE, PIPE_SWIZZLE_ALPHA));
- OUT_RING(ring, A4XX_TEX_CONST_1_WIDTH(psurf->width) |
- A4XX_TEX_CONST_1_HEIGHT(psurf->height));
- OUT_RING(ring, A4XX_TEX_CONST_2_PITCH(slice->pitch * rsc->cpp));
- OUT_RING(ring, 0x00000000);
- OUT_RELOC(ring, rsc->bo, layer_offset, 0, 0);
- OUT_RING(ring, 0x00000000);
- OUT_RING(ring, 0x00000000);
- OUT_RING(ring, 0x00000000);
+ for (i = 0; i < nr_bufs; i++) {
+ if (bufs[i]) {
+ struct fd_resource *rsc = fd_resource(bufs[i]->texture);
+ /* note: PIPE_BUFFER disallowed for surfaces */
+ unsigned lvl = bufs[i]->u.tex.level;
+ struct fd_resource_slice *slice = fd_resource_slice(rsc, lvl);
+ uint32_t offset = fd_resource_offset(rsc, lvl, bufs[i]->u.tex.first_layer);
+ enum pipe_format format = fd4_gmem_restore_format(bufs[i]->format);
+
+ /* The restore blit_zs shader expects stencil in sampler 0,
+ * and depth in sampler 1
+ */
+ if (rsc->stencil && (i == 0)) {
+ rsc = rsc->stencil;
+ format = fd4_gmem_restore_format(rsc->base.b.format);
+ }
+
+ /* z32 restore is accomplished using depth write. If there is
+ * no stencil component (ie. PIPE_FORMAT_Z32_FLOAT_S8X24_UINT)
+ * then no render target:
+ *
+ * (The same applies for z32_s8x24, since for stencil sampler
+ * state the above 'if' will replace 'format' with s8)
+ */
+ if ((format == PIPE_FORMAT_Z32_FLOAT) ||
+ (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT))
+ mrt_comp[i] = 0;
+
+ debug_assert(bufs[i]->u.tex.first_layer == bufs[i]->u.tex.last_layer);
+
+ OUT_RING(ring, A4XX_TEX_CONST_0_FMT(fd4_pipe2tex(format)) |
+ A4XX_TEX_CONST_0_TYPE(A4XX_TEX_2D) |
+ fd4_tex_swiz(format, PIPE_SWIZZLE_RED, PIPE_SWIZZLE_GREEN,
+ PIPE_SWIZZLE_BLUE, PIPE_SWIZZLE_ALPHA));
+ OUT_RING(ring, A4XX_TEX_CONST_1_WIDTH(bufs[i]->width) |
+ A4XX_TEX_CONST_1_HEIGHT(bufs[i]->height));
+ OUT_RING(ring, A4XX_TEX_CONST_2_PITCH(slice->pitch * rsc->cpp) |
+ A4XX_TEX_CONST_2_FETCHSIZE(fd4_pipe2fetchsize(format)));
+ OUT_RING(ring, 0x00000000);
+ OUT_RELOC(ring, rsc->bo, offset, 0, 0);
+ OUT_RING(ring, 0x00000000);
+ OUT_RING(ring, 0x00000000);
+ OUT_RING(ring, 0x00000000);
+ } else {
+ OUT_RING(ring, A4XX_TEX_CONST_0_FMT(0) |
+ A4XX_TEX_CONST_0_TYPE(A4XX_TEX_2D) |
+ A4XX_TEX_CONST_0_SWIZ_X(A4XX_TEX_ONE) |
+ A4XX_TEX_CONST_0_SWIZ_Y(A4XX_TEX_ONE) |
+ A4XX_TEX_CONST_0_SWIZ_Z(A4XX_TEX_ONE) |
+ A4XX_TEX_CONST_0_SWIZ_W(A4XX_TEX_ONE));
+ OUT_RING(ring, A4XX_TEX_CONST_1_WIDTH(0) |
+ A4XX_TEX_CONST_1_HEIGHT(0));
+ OUT_RING(ring, A4XX_TEX_CONST_2_PITCH(0));
+ OUT_RING(ring, 0x00000000);
+ OUT_RING(ring, 0x00000000);
+ OUT_RING(ring, 0x00000000);
+ OUT_RING(ring, 0x00000000);
+ OUT_RING(ring, 0x00000000);
+ }
+ }
+
+ OUT_PKT0(ring, REG_A4XX_RB_RENDER_COMPONENTS, 1);
+ OUT_RING(ring, A4XX_RB_RENDER_COMPONENTS_RT0(mrt_comp[0]) |
+ A4XX_RB_RENDER_COMPONENTS_RT1(mrt_comp[1]) |
+ A4XX_RB_RENDER_COMPONENTS_RT2(mrt_comp[2]) |
+ A4XX_RB_RENDER_COMPONENTS_RT3(mrt_comp[3]) |
+ A4XX_RB_RENDER_COMPONENTS_RT4(mrt_comp[4]) |
+ A4XX_RB_RENDER_COMPONENTS_RT5(mrt_comp[5]) |
+ A4XX_RB_RENDER_COMPONENTS_RT6(mrt_comp[6]) |
+ A4XX_RB_RENDER_COMPONENTS_RT7(mrt_comp[7]));
}
void
fd4_emit_vertex_bufs(struct fd_ringbuffer *ring, struct fd4_emit *emit)
{
- uint32_t i, j, last = 0;
+ int32_t i, j, last = -1;
uint32_t total_in = 0;
const struct fd_vertex_state *vtx = emit->vtx;
struct ir3_shader_variant *vp = fd4_emit_get_vp(emit);
- unsigned n = MIN2(vtx->vtx->num_elements, vp->inputs_count);
-
- /* hw doesn't like to be configured for zero vbo's, it seems: */
- if (vtx->vtx->num_elements == 0)
- return;
-
- for (i = 0; i < n; i++)
- if (vp->inputs[i].compmask)
+ unsigned vertex_regid = regid(63, 0);
+ unsigned instance_regid = regid(63, 0);
+ unsigned vtxcnt_regid = regid(63, 0);
+
+ /* Note that sysvals come *after* normal inputs: */
+ for (i = 0; i < vp->inputs_count; i++) {
+ if (!vp->inputs[i].compmask)
+ continue;
+ if (vp->inputs[i].sysval) {
+ switch(vp->inputs[i].slot) {
+ case SYSTEM_VALUE_BASE_VERTEX:
+ /* handled elsewhere */
+ break;
+ case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE:
+ vertex_regid = vp->inputs[i].regid;
+ break;
+ case SYSTEM_VALUE_INSTANCE_ID:
+ instance_regid = vp->inputs[i].regid;
+ break;
+ case SYSTEM_VALUE_VERTEX_CNT:
+ vtxcnt_regid = vp->inputs[i].regid;
+ break;
+ default:
+ unreachable("invalid system value");
+ break;
+ }
+ } else if (i < vtx->vtx->num_elements) {
last = i;
+ }
+ }
for (i = 0, j = 0; i <= last; i++) {
+ assert(!vp->inputs[i].sysval);
if (vp->inputs[i].compmask) {
struct pipe_vertex_element *elem = &vtx->vtx->pipe[i];
const struct pipe_vertex_buffer *vb =
struct fd_resource *rsc = fd_resource(vb->buffer);
enum pipe_format pfmt = elem->src_format;
enum a4xx_vtx_fmt fmt = fd4_pipe2vtx(pfmt);
- bool switchnext = (i != last);
+ bool switchnext = (i != last) ||
+ (vertex_regid != regid(63, 0)) ||
+ (instance_regid != regid(63, 0)) ||
+ (vtxcnt_regid != regid(63, 0));
+ bool isint = util_format_is_pure_integer(pfmt);
uint32_t fs = util_format_get_blocksize(pfmt);
uint32_t off = vb->buffer_offset + elem->src_offset;
uint32_t size = fd_bo_size(rsc->bo) - off;
OUT_PKT0(ring, REG_A4XX_VFD_FETCH(j), 4);
OUT_RING(ring, A4XX_VFD_FETCH_INSTR_0_FETCHSIZE(fs - 1) |
A4XX_VFD_FETCH_INSTR_0_BUFSTRIDE(vb->stride) |
+ COND(elem->instance_divisor, A4XX_VFD_FETCH_INSTR_0_INSTANCED) |
COND(switchnext, A4XX_VFD_FETCH_INSTR_0_SWITCHNEXT));
OUT_RELOC(ring, rsc->bo, off, 0, 0);
OUT_RING(ring, A4XX_VFD_FETCH_INSTR_2_SIZE(size));
- OUT_RING(ring, 0x00000001);
+ OUT_RING(ring, A4XX_VFD_FETCH_INSTR_3_STEPRATE(MAX2(1, elem->instance_divisor)));
OUT_PKT0(ring, REG_A4XX_VFD_DECODE_INSTR(j), 1);
OUT_RING(ring, A4XX_VFD_DECODE_INSTR_CONSTFILL |
A4XX_VFD_DECODE_INSTR_REGID(vp->inputs[i].regid) |
A4XX_VFD_DECODE_INSTR_SHIFTCNT(fs) |
A4XX_VFD_DECODE_INSTR_LASTCOMPVALID |
+ COND(isint, A4XX_VFD_DECODE_INSTR_INT) |
COND(switchnext, A4XX_VFD_DECODE_INSTR_SWITCHNEXT));
total_in += vp->inputs[i].ncomp;
}
}
+ /* hw doesn't like to be configured for zero vbo's, it seems: */
+ if (last < 0) {
+ /* just recycle the shader bo, we just need to point to *something*
+ * valid:
+ */
+ struct fd_bo *dummy_vbo = vp->bo;
+ bool switchnext = (vertex_regid != regid(63, 0)) ||
+ (instance_regid != regid(63, 0)) ||
+ (vtxcnt_regid != regid(63, 0));
+
+ OUT_PKT0(ring, REG_A4XX_VFD_FETCH(0), 4);
+ OUT_RING(ring, A4XX_VFD_FETCH_INSTR_0_FETCHSIZE(0) |
+ A4XX_VFD_FETCH_INSTR_0_BUFSTRIDE(0) |
+ COND(switchnext, A4XX_VFD_FETCH_INSTR_0_SWITCHNEXT));
+ OUT_RELOC(ring, dummy_vbo, 0, 0, 0);
+ OUT_RING(ring, A4XX_VFD_FETCH_INSTR_2_SIZE(1));
+ OUT_RING(ring, A4XX_VFD_FETCH_INSTR_3_STEPRATE(1));
+
+ OUT_PKT0(ring, REG_A4XX_VFD_DECODE_INSTR(0), 1);
+ OUT_RING(ring, A4XX_VFD_DECODE_INSTR_CONSTFILL |
+ A4XX_VFD_DECODE_INSTR_WRITEMASK(0x1) |
+ A4XX_VFD_DECODE_INSTR_FORMAT(VFMT4_8_UNORM) |
+ A4XX_VFD_DECODE_INSTR_SWAP(XYZW) |
+ A4XX_VFD_DECODE_INSTR_REGID(regid(0,0)) |
+ A4XX_VFD_DECODE_INSTR_SHIFTCNT(1) |
+ A4XX_VFD_DECODE_INSTR_LASTCOMPVALID |
+ COND(switchnext, A4XX_VFD_DECODE_INSTR_SWITCHNEXT));
+
+ total_in = 1;
+ j = 1;
+ }
+
OUT_PKT0(ring, REG_A4XX_VFD_CONTROL_0, 5);
OUT_RING(ring, A4XX_VFD_CONTROL_0_TOTALATTRTOVS(total_in) |
0xa0000 | /* XXX */
A4XX_VFD_CONTROL_0_STRMDECINSTRCNT(j) |
A4XX_VFD_CONTROL_0_STRMFETCHINSTRCNT(j));
OUT_RING(ring, A4XX_VFD_CONTROL_1_MAXSTORAGE(129) | // XXX
- A4XX_VFD_CONTROL_1_REGID4VTX(regid(63,0)) |
- A4XX_VFD_CONTROL_1_REGID4INST(regid(63,0)));
+ A4XX_VFD_CONTROL_1_REGID4VTX(vertex_regid) |
+ A4XX_VFD_CONTROL_1_REGID4INST(instance_regid));
OUT_RING(ring, 0x00000000); /* XXX VFD_CONTROL_2 */
- OUT_RING(ring, 0x0000fc00); /* XXX VFD_CONTROL_3 */
+ OUT_RING(ring, A4XX_VFD_CONTROL_3_REGID_VTXCNT(vtxcnt_regid));
OUT_RING(ring, 0x00000000); /* XXX VFD_CONTROL_4 */
+
+ /* cache invalidate, otherwise vertex fetch could see
+ * stale vbo contents:
+ */
+ OUT_PKT0(ring, REG_A4XX_UCHE_INVALIDATE0, 2);
+ OUT_RING(ring, 0x00000000);
+ OUT_RING(ring, 0x00000012);
}
void
emit_marker(ring, 5);
+ if ((dirty & FD_DIRTY_FRAMEBUFFER) && !emit->key.binning_pass) {
+ struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+ unsigned char mrt_comp[A4XX_MAX_RENDER_TARGETS] = {0};
+
+ for (unsigned i = 0; i < A4XX_MAX_RENDER_TARGETS; i++) {
+ mrt_comp[i] = ((i < pfb->nr_cbufs) && pfb->cbufs[i]) ? 0xf : 0;
+ }
+
+ OUT_PKT0(ring, REG_A4XX_RB_RENDER_COMPONENTS, 1);
+ OUT_RING(ring, A4XX_RB_RENDER_COMPONENTS_RT0(mrt_comp[0]) |
+ A4XX_RB_RENDER_COMPONENTS_RT1(mrt_comp[1]) |
+ A4XX_RB_RENDER_COMPONENTS_RT2(mrt_comp[2]) |
+ A4XX_RB_RENDER_COMPONENTS_RT3(mrt_comp[3]) |
+ A4XX_RB_RENDER_COMPONENTS_RT4(mrt_comp[4]) |
+ A4XX_RB_RENDER_COMPONENTS_RT5(mrt_comp[5]) |
+ A4XX_RB_RENDER_COMPONENTS_RT6(mrt_comp[6]) |
+ A4XX_RB_RENDER_COMPONENTS_RT7(mrt_comp[7]));
+ }
+
if ((dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG)) && !emit->key.binning_pass) {
uint32_t val = fd4_zsa_stateobj(ctx->zsa)->rb_render_control;
OUT_RINGP(ring, val, &fd4_context(ctx)->rbrc_patches);
}
- if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_STENCIL_REF)) {
+ if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_FRAMEBUFFER)) {
struct fd4_zsa_stateobj *zsa = fd4_zsa_stateobj(ctx->zsa);
- struct pipe_stencil_ref *sr = &ctx->stencil_ref;
+ struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+ uint32_t rb_alpha_control = zsa->rb_alpha_control;
- OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
- OUT_RING(ring, zsa->gras_alpha_control);
+ if (util_format_is_pure_integer(pipe_surface_format(pfb->cbufs[0])))
+ rb_alpha_control &= ~A4XX_RB_ALPHA_CONTROL_ALPHA_TEST;
+
+ OUT_PKT0(ring, REG_A4XX_RB_ALPHA_CONTROL, 1);
+ OUT_RING(ring, rb_alpha_control);
- OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 1);
+ OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 2);
OUT_RING(ring, zsa->rb_stencil_control);
+ OUT_RING(ring, zsa->rb_stencil_control2);
+ }
+
+ if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_STENCIL_REF)) {
+ struct fd4_zsa_stateobj *zsa = fd4_zsa_stateobj(ctx->zsa);
+ struct pipe_stencil_ref *sr = &ctx->stencil_ref;
OUT_PKT0(ring, REG_A4XX_RB_STENCILREFMASK, 2);
OUT_RING(ring, zsa->rb_stencilrefmask |
}
if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG)) {
- uint32_t val = fd4_zsa_stateobj(ctx->zsa)->rb_depth_control;
- if (fp->writes_pos) {
- val |= A4XX_RB_DEPTH_CONTROL_FRAG_WRITES_Z;
- val |= A4XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE;
- }
- if (fp->has_kill) {
- val |= A4XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE;
- }
+ struct fd4_zsa_stateobj *zsa = fd4_zsa_stateobj(ctx->zsa);
+ bool fragz = fp->has_kill | fp->writes_pos;
+
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
- OUT_RING(ring, val);
+ OUT_RING(ring, zsa->rb_depth_control |
+ COND(fragz, A4XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE) |
+ COND(fragz && fp->frag_coord, A4XX_RB_DEPTH_CONTROL_FORCE_FRAGZ_TO_FS));
+
+ /* maybe this register/bitfield needs a better name.. this
+ * appears to be just disabling early-z
+ */
+ OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
+ OUT_RING(ring, zsa->gras_alpha_control |
+ COND(fragz, A4XX_GRAS_ALPHA_CONTROL_ALPHA_TEST_ENABLE) |
+ COND(fragz && fp->frag_coord, A4XX_GRAS_ALPHA_CONTROL_FORCE_FRAGZ_TO_FS));
}
if (dirty & FD_DIRTY_RASTERIZER) {
OUT_PKT0(ring, REG_A4XX_GRAS_SU_POLY_OFFSET_SCALE, 2);
OUT_RING(ring, rasterizer->gras_su_poly_offset_scale);
OUT_RING(ring, rasterizer->gras_su_poly_offset_offset);
- }
- if (dirty & (FD_DIRTY_RASTERIZER | FD_DIRTY_PROG)) {
- uint32_t val = fd4_rasterizer_stateobj(ctx->rasterizer)
- ->gras_cl_clip_cntl;
OUT_PKT0(ring, REG_A4XX_GRAS_CL_CLIP_CNTL, 1);
- OUT_RING(ring, val);
+ OUT_RING(ring, rasterizer->gras_cl_clip_cntl);
}
/* NOTE: since primitive_restart is not actually part of any
* when it changes.
*/
if (emit->info) {
- uint32_t val = fd4_rasterizer_stateobj(ctx->rasterizer)
- ->pc_prim_vtx_cntl;
+ const struct pipe_draw_info *info = emit->info;
+ struct fd4_rasterizer_stateobj *rast =
+ fd4_rasterizer_stateobj(ctx->rasterizer);
+ uint32_t val = rast->pc_prim_vtx_cntl;
+
+ if (info->indexed && info->primitive_restart)
+ val |= A4XX_PC_PRIM_VTX_CNTL_PRIMITIVE_RESTART;
val |= COND(vp->writes_psize, A4XX_PC_PRIM_VTX_CNTL_PSIZE);
- val |= COND(fp->total_in > 0, A4XX_PC_PRIM_VTX_CNTL_VAROUT);
+
+ if (fp->total_in > 0) {
+ uint32_t varout = align(fp->total_in, 16) / 16;
+ if (varout > 1)
+ varout = align(varout, 2);
+ val |= A4XX_PC_PRIM_VTX_CNTL_VAROUT(varout);
+ }
OUT_PKT0(ring, REG_A4XX_PC_PRIM_VTX_CNTL, 2);
OUT_RING(ring, val);
- OUT_RING(ring, 0x12); /* XXX UNKNOWN_21C5 */
+ OUT_RING(ring, rast->pc_prim_vtx_cntl2);
}
if (dirty & FD_DIRTY_SCISSOR) {
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZSCALE_0(ctx->viewport.scale[2]));
}
- if (dirty & FD_DIRTY_PROG)
- fd4_program_emit(ring, emit);
+ if (dirty & (FD_DIRTY_PROG | FD_DIRTY_FRAMEBUFFER)) {
+ struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+ fd4_program_emit(ring, emit, pfb->nr_cbufs, pfb->cbufs);
+ }
- if ((dirty & (FD_DIRTY_PROG | FD_DIRTY_CONSTBUF)) &&
- /* evil hack to deal sanely with clear path: */
- (emit->prog == &ctx->prog)) {
- fd_wfi(ctx, ring);
- emit_constants(ring, SB_VERT_SHADER,
- &ctx->constbuf[PIPE_SHADER_VERTEX],
- (emit->prog->dirty & FD_SHADER_DIRTY_VP) ? vp : NULL);
- if (!emit->key.binning_pass) {
- emit_constants(ring, SB_FRAG_SHADER,
- &ctx->constbuf[PIPE_SHADER_FRAGMENT],
- (emit->prog->dirty & FD_SHADER_DIRTY_FP) ? fp : NULL);
- }
+ if (emit->prog == &ctx->prog) { /* evil hack to deal sanely with clear path */
+ ir3_emit_consts(vp, ring, ctx, emit->info, dirty);
+ if (!emit->key.binning_pass)
+ ir3_emit_consts(fp, ring, ctx, emit->info, dirty);
+ /* mark clean after emitting consts: */
+ ctx->prog.dirty = 0;
}
- if ((dirty & FD_DIRTY_BLEND) && ctx->blend) {
+ if ((dirty & FD_DIRTY_BLEND)) {
struct fd4_blend_stateobj *blend = fd4_blend_stateobj(ctx->blend);
uint32_t i;
- for (i = 0; i < 8; i++) {
+ for (i = 0; i < A4XX_MAX_RENDER_TARGETS; i++) {
+ enum pipe_format format = pipe_surface_format(
+ ctx->framebuffer.cbufs[i]);
+ bool is_int = util_format_is_pure_integer(format);
+ bool has_alpha = util_format_has_alpha(format);
+ uint32_t control = blend->rb_mrt[i].control;
+ uint32_t blend_control = blend->rb_mrt[i].blend_control_alpha;
+
+ if (is_int) {
+ control &= A4XX_RB_MRT_CONTROL_COMPONENT_ENABLE__MASK;
+ control |= A4XX_RB_MRT_CONTROL_ROP_CODE(ROP_COPY);
+ }
+
+ if (has_alpha) {
+ blend_control |= blend->rb_mrt[i].blend_control_rgb;
+ } else {
+ blend_control |= blend->rb_mrt[i].blend_control_no_alpha_rgb;
+ control &= ~A4XX_RB_MRT_CONTROL_BLEND2;
+ }
+
OUT_PKT0(ring, REG_A4XX_RB_MRT_CONTROL(i), 1);
- OUT_RING(ring, blend->rb_mrt[i].control);
+ OUT_RING(ring, control);
OUT_PKT0(ring, REG_A4XX_RB_MRT_BLEND_CONTROL(i), 1);
- OUT_RING(ring, blend->rb_mrt[i].blend_control);
+ OUT_RING(ring, blend_control);
}
+
+ OUT_PKT0(ring, REG_A4XX_RB_FS_OUTPUT, 1);
+ OUT_RING(ring, blend->rb_fs_output |
+ A4XX_RB_FS_OUTPUT_SAMPLE_MASK(0xffff));
+ }
+
+ if (dirty & (FD_DIRTY_BLEND_COLOR | FD_DIRTY_FRAMEBUFFER)) {
+ struct pipe_blend_color *bcolor = &ctx->blend_color;
+ struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+ float factor = 65535.0;
+ int i;
+
+ for (i = 0; i < pfb->nr_cbufs; i++) {
+ enum pipe_format format = pipe_surface_format(pfb->cbufs[i]);
+ const struct util_format_description *desc =
+ util_format_description(format);
+ int j;
+
+ if (desc->is_mixed)
+ continue;
+
+ j = util_format_get_first_non_void_channel(format);
+ if (j == -1)
+ continue;
+
+ if (desc->channel[j].size > 8 || !desc->channel[j].normalized ||
+ desc->channel[j].pure_integer)
+ continue;
+
+ /* Just use the first unorm8/snorm8 render buffer. Can't keep
+ * everyone happy.
+ */
+ if (desc->channel[j].type == UTIL_FORMAT_TYPE_SIGNED)
+ factor = 32767.0;
+ break;
+ }
+
+ OUT_PKT0(ring, REG_A4XX_RB_BLEND_RED, 8);
+ OUT_RING(ring, A4XX_RB_BLEND_RED_UINT(bcolor->color[0] * factor) |
+ A4XX_RB_BLEND_RED_FLOAT(bcolor->color[0]));
+ OUT_RING(ring, A4XX_RB_BLEND_RED_F32(bcolor->color[0]));
+ OUT_RING(ring, A4XX_RB_BLEND_GREEN_UINT(bcolor->color[1] * factor) |
+ A4XX_RB_BLEND_GREEN_FLOAT(bcolor->color[1]));
+ OUT_RING(ring, A4XX_RB_BLEND_GREEN_F32(bcolor->color[1]));
+ OUT_RING(ring, A4XX_RB_BLEND_BLUE_UINT(bcolor->color[2] * factor) |
+ A4XX_RB_BLEND_BLUE_FLOAT(bcolor->color[2]));
+ OUT_RING(ring, A4XX_RB_BLEND_BLUE_F32(bcolor->color[2]));
+ OUT_RING(ring, A4XX_RB_BLEND_ALPHA_UINT(bcolor->color[3] * factor) |
+ A4XX_RB_BLEND_ALPHA_FLOAT(bcolor->color[3]));
+ OUT_RING(ring, A4XX_RB_BLEND_ALPHA_F32(bcolor->color[3]));
}
if (dirty & FD_DIRTY_VERTTEX) {
OUT_PKT0(ring, REG_A4XX_GRAS_DEBUG_ECO_CONTROL, 1);
OUT_RING(ring, 0x00000000);
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_0EC3, 1);
+ OUT_PKT0(ring, REG_A4XX_SP_MODE_CONTROL, 1);
OUT_RING(ring, 0x00000006);
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_0F03, 1);
+ OUT_PKT0(ring, REG_A4XX_TPL1_TP_MODE_CONTROL, 1);
OUT_RING(ring, 0x0000003a);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_0D01, 1);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000012);
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_0E05, 1);
+ OUT_PKT0(ring, REG_A4XX_HLSQ_MODE_CONTROL, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_0CC5, 1);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_20EF, 1);
OUT_RING(ring, 0x00000000);
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F0, 1);
- OUT_RING(ring, 0x00000000);
-
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F1, 1);
- OUT_RING(ring, 0x00000000);
-
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F2, 1);
- OUT_RING(ring, 0x00000000);
-
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F3, 1);
- OUT_RING(ring, 0x00000000);
-
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F4, 1);
- OUT_RING(ring, 0x00000000);
-
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F5, 1);
- OUT_RING(ring, 0x00000000);
-
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F6, 1);
- OUT_RING(ring, 0x3c007fff);
-
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F7, 1);
- OUT_RING(ring, 0x3f800000);
+ OUT_PKT0(ring, REG_A4XX_RB_BLEND_RED, 4);
+ OUT_RING(ring, A4XX_RB_BLEND_RED_UINT(0) |
+ A4XX_RB_BLEND_RED_FLOAT(0.0));
+ OUT_RING(ring, A4XX_RB_BLEND_GREEN_UINT(0) |
+ A4XX_RB_BLEND_GREEN_FLOAT(0.0));
+ OUT_RING(ring, A4XX_RB_BLEND_BLUE_UINT(0) |
+ A4XX_RB_BLEND_BLUE_FLOAT(0.0));
+ OUT_RING(ring, A4XX_RB_BLEND_ALPHA_UINT(0x7fff) |
+ A4XX_RB_BLEND_ALPHA_FLOAT(1.0));
OUT_PKT0(ring, REG_A4XX_UNKNOWN_2152, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_TPL1_TP_TEX_OFFSET, 1);
OUT_RING(ring, 0x00000000);
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_2381, 1);
- OUT_RING(ring, 0x00000010);
+ OUT_PKT0(ring, REG_A4XX_TPL1_TP_TEX_COUNT, 1);
+ OUT_RING(ring, A4XX_TPL1_TP_TEX_COUNT_VS(16) |
+ A4XX_TPL1_TP_TEX_COUNT_HS(0) |
+ A4XX_TPL1_TP_TEX_COUNT_DS(0) |
+ A4XX_TPL1_TP_TEX_COUNT_GS(0));
- OUT_PKT0(ring, REG_A4XX_UNKNOWN_23A0, 1);
- OUT_RING(ring, 0x00000010);
+ OUT_PKT0(ring, REG_A4XX_TPL1_TP_FS_TEX_COUNT, 1);
+ OUT_RING(ring, 16);
/* we don't use this yet.. probably best to disable.. */
OUT_PKT3(ring, CP_SET_DRAW_STATE, 2);
OUT_PKT0(ring, REG_A4XX_RB_FS_OUTPUT, 1);
OUT_RING(ring, A4XX_RB_FS_OUTPUT_SAMPLE_MASK(0xffff));
- OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL3, 1);
- OUT_RING(ring, A4XX_RB_RENDER_CONTROL3_COMPONENT_ENABLE(0xf));
+ OUT_PKT0(ring, REG_A4XX_GRAS_CLEAR_CNTL, 1);
+ OUT_RING(ring, A4XX_GRAS_CLEAR_CNTL_NOT_FASTCLEAR);
+
+ OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
+ OUT_RING(ring, 0x0);
- OUT_PKT0(ring, REG_A4XX_RB_FS_OUTPUT_REG, 1);
- OUT_RING(ring, A4XX_RB_FS_OUTPUT_REG_COLOR_PIPE_ENABLE);
+ fd_hw_query_enable(ctx, ring);
ctx->needs_rb_fbd = true;
}
+
+static void
+fd4_emit_ib(struct fd_ringbuffer *ring, struct fd_ringmarker *start,
+ struct fd_ringmarker *end)
+{
+ __OUT_IB(ring, true, start, end);
+}
+
+void
+fd4_emit_init(struct pipe_context *pctx)
+{
+ struct fd_context *ctx = fd_context(pctx);
+ ctx->emit_const = fd4_emit_const;
+ ctx->emit_const_bo = fd4_emit_const_bo;
+ ctx->emit_ib = fd4_emit_ib;
+}
projects / mesa.git / blobdiff