From 272bbbffb0138024564cbf8410a30ea9e25eb179 Mon Sep 17 00:00:00 2001 From: Christoph Bumiller Date: Sun, 21 Mar 2010 13:17:02 +0100 Subject: [PATCH] nv50: get rid of the static_init stateobj Relocations of per-screen buffers are now emitted directly, and include the necessary method to get changes in constbuf addresses committed to the hw. It should also be a bit cheaper than the way stateobjs emit relocation markers, use a little less pushbuf space. --- src/gallium/drivers/nv50/nv50_screen.c | 276 +++++++++--------- src/gallium/drivers/nv50/nv50_screen.h | 4 +- .../drivers/nv50/nv50_state_validate.c | 2 +- 3 files changed, 142 insertions(+), 140 deletions(-) diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c index d7f5863fb71..8fab06c55da 100644 --- a/src/gallium/drivers/nv50/nv50_screen.c +++ b/src/gallium/drivers/nv50/nv50_screen.c @@ -190,8 +190,6 @@ nv50_screen_destroy(struct pipe_screen *pscreen) nouveau_bo_ref(NULL, &screen->tic); if (screen->tsc) nouveau_bo_ref(NULL, &screen->tsc); - if (screen->static_init) - so_ref(NULL, &screen->static_init); nouveau_notifier_free(&screen->sync); nouveau_grobj_free(&screen->tesla); @@ -204,16 +202,65 @@ nv50_screen_destroy(struct pipe_screen *pscreen) FREE(screen); } +#define BGN_RELOC(ch, bo, gr, m, n, fl) \ + OUT_RELOC(ch, bo, (n << 18) | (gr->subc << 13) | m, fl, 0, 0) + +void +nv50_screen_relocs(struct nv50_screen *screen) +{ + struct nouveau_channel *chan = screen->base.channel; + struct nouveau_grobj *tesla = screen->tesla; + unsigned i; + const unsigned rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_DUMMY; + + MARK_RING (chan, 28, 26); + + /* cause grobj autobind */ + BEGIN_RING(chan, tesla, 0x0100, 1); + OUT_RING (chan, 0); + + BGN_RELOC (chan, screen->tic, tesla, NV50TCL_TIC_ADDRESS_HIGH, 2, rl); + OUT_RELOCh(chan, screen->tic, 0, rl); + OUT_RELOCl(chan, screen->tic, 0, rl); + + BGN_RELOC (chan, screen->tsc, tesla, NV50TCL_TSC_ADDRESS_HIGH, 2, rl); + OUT_RELOCh(chan, screen->tsc, 0, rl); + OUT_RELOCl(chan, screen->tsc, 0, rl); + + BGN_RELOC (chan, screen->constbuf_misc[0], + tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3, rl); + OUT_RELOCh(chan, screen->constbuf_misc[0], 0, rl); + OUT_RELOCl(chan, screen->constbuf_misc[0], 0, rl); + OUT_RELOC (chan, screen->constbuf_misc[0], + (NV50_CB_PMISC << 16) | 0x0200, rl, 0, 0); + + BGN_RELOC (chan, screen->constbuf_misc[0], + tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3, rl); + OUT_RELOCh(chan, screen->constbuf_misc[0], 0x200, rl); + OUT_RELOCl(chan, screen->constbuf_misc[0], 0x200, rl); + OUT_RELOC (chan, screen->constbuf_misc[0], + (NV50_CB_AUX << 16) | 0x0200, rl, 0, 0); + + for (i = 0; i < 3; ++i) { + BGN_RELOC (chan, screen->constbuf_parm[i], + tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3, rl); + OUT_RELOCh(chan, screen->constbuf_parm[i], 0, rl); + OUT_RELOCl(chan, screen->constbuf_parm[i], 0, rl); + OUT_RELOC (chan, screen->constbuf_parm[i], + ((NV50_CB_PVP + i) << 16) | 0x0800, rl, 0, 0); + } +} + struct pipe_screen * nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) { struct nv50_screen *screen = CALLOC_STRUCT(nv50_screen); struct nouveau_channel *chan; struct pipe_screen *pscreen; - struct nouveau_stateobj *so; unsigned chipset = dev->chipset; unsigned tesla_class = 0; int ret, i; + const unsigned rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD; if (!screen) return NULL; @@ -296,64 +343,58 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) } /* Static M2MF init */ - so = so_new(1, 3, 0); - so_method(so, screen->m2mf, NV04_MEMORY_TO_MEMORY_FORMAT_DMA_NOTIFY, 3); - so_data (so, screen->sync->handle); - so_data (so, chan->vram->handle); - so_data (so, chan->vram->handle); - so_emit(chan, so); - so_ref (NULL, &so); + BEGIN_RING(chan, screen->m2mf, + NV04_MEMORY_TO_MEMORY_FORMAT_DMA_NOTIFY, 3); + OUT_RING (chan, screen->sync->handle); + OUT_RING (chan, chan->vram->handle); + OUT_RING (chan, chan->vram->handle); /* Static 2D init */ - so = so_new(4, 7, 0); - so_method(so, screen->eng2d, NV50_2D_DMA_NOTIFY, 4); - so_data (so, screen->sync->handle); - so_data (so, chan->vram->handle); - so_data (so, chan->vram->handle); - so_data (so, chan->vram->handle); - so_method(so, screen->eng2d, NV50_2D_OPERATION, 1); - so_data (so, NV50_2D_OPERATION_SRCCOPY); - so_method(so, screen->eng2d, NV50_2D_CLIP_ENABLE, 1); - so_data (so, 0); - so_method(so, screen->eng2d, 0x0888, 1); - so_data (so, 1); - so_emit(chan, so); - so_ref(NULL, &so); + BEGIN_RING(chan, screen->eng2d, NV50_2D_DMA_NOTIFY, 4); + OUT_RING (chan, screen->sync->handle); + OUT_RING (chan, chan->vram->handle); + OUT_RING (chan, chan->vram->handle); + OUT_RING (chan, chan->vram->handle); + BEGIN_RING(chan, screen->eng2d, NV50_2D_OPERATION, 1); + OUT_RING (chan, NV50_2D_OPERATION_SRCCOPY); + BEGIN_RING(chan, screen->eng2d, NV50_2D_CLIP_ENABLE, 1); + OUT_RING (chan, 0); + BEGIN_RING(chan, screen->eng2d, 0x0888, 1); + OUT_RING (chan, 1); /* Static tesla init */ - so = so_new(47, 95, 24); - - so_method(so, screen->tesla, NV50TCL_COND_MODE, 1); - so_data (so, NV50TCL_COND_MODE_ALWAYS); - so_method(so, screen->tesla, NV50TCL_DMA_NOTIFY, 1); - so_data (so, screen->sync->handle); - so_method(so, screen->tesla, NV50TCL_DMA_ZETA, 11); + BEGIN_RING(chan, screen->tesla, NV50TCL_COND_MODE, 1); + OUT_RING (chan, NV50TCL_COND_MODE_ALWAYS); + BEGIN_RING(chan, screen->tesla, NV50TCL_DMA_NOTIFY, 1); + OUT_RING (chan, screen->sync->handle); + BEGIN_RING(chan, screen->tesla, NV50TCL_DMA_ZETA, 11); for (i = 0; i < 11; i++) - so_data(so, chan->vram->handle); - so_method(so, screen->tesla, NV50TCL_DMA_COLOR(0), - NV50TCL_DMA_COLOR__SIZE); + OUT_RING (chan, chan->vram->handle); + BEGIN_RING(chan, screen->tesla, + NV50TCL_DMA_COLOR(0), NV50TCL_DMA_COLOR__SIZE); for (i = 0; i < NV50TCL_DMA_COLOR__SIZE; i++) - so_data(so, chan->vram->handle); - so_method(so, screen->tesla, NV50TCL_RT_CONTROL, 1); - so_data (so, 1); + OUT_RING (chan, chan->vram->handle); + + BEGIN_RING(chan, screen->tesla, NV50TCL_RT_CONTROL, 1); + OUT_RING (chan, 1); /* activate all 32 lanes (threads) in a warp */ - so_method(so, screen->tesla, NV50TCL_WARP_HALVES, 1); - so_data (so, 0x2); - so_method(so, screen->tesla, 0x1400, 1); - so_data (so, 0xf); + BEGIN_RING(chan, screen->tesla, NV50TCL_WARP_HALVES, 1); + OUT_RING (chan, 2); + BEGIN_RING(chan, screen->tesla, 0x1400, 1); + OUT_RING (chan, 0xf); /* max TIC (bits 4:8) & TSC (ignored) bindings, per program type */ for (i = 0; i < 3; ++i) { - so_method(so, screen->tesla, NV50TCL_TEX_LIMITS(i), 1); - so_data (so, 0x54); + BEGIN_RING(chan, screen->tesla, NV50TCL_TEX_LIMITS(i), 1); + OUT_RING (chan, 0x54); } /* origin is top left (set to 1 for bottom left) */ - so_method(so, screen->tesla, NV50TCL_Y_ORIGIN_BOTTOM, 1); - so_data (so, 0); - so_method(so, screen->tesla, NV50TCL_VP_REG_ALLOC_RESULT, 1); - so_data (so, 8); + BEGIN_RING(chan, screen->tesla, NV50TCL_Y_ORIGIN_BOTTOM, 1); + OUT_RING (chan, 0); + BEGIN_RING(chan, screen->tesla, NV50TCL_VP_REG_ALLOC_RESULT, 1); + OUT_RING (chan, 8); /* constant buffers for immediates and VP/FP parameters */ ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, (32 * 4) * 4, @@ -362,6 +403,14 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) nv50_screen_destroy(pscreen); return NULL; } + BEGIN_RING(chan, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); + OUT_RELOCh(chan, screen->constbuf_misc[0], 0, rl); + OUT_RELOCl(chan, screen->constbuf_misc[0], 0, rl); + OUT_RING (chan, (NV50_CB_PMISC << 16) | 0x0200); + BEGIN_RING(chan, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); + OUT_RELOCh(chan, screen->constbuf_misc[0], 0x200, rl); + OUT_RELOCl(chan, screen->constbuf_misc[0], 0x200, rl); + OUT_RING (chan, (NV50_CB_AUX << 16) | 0x0200); for (i = 0; i < 3; i++) { ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, (256 * 4) * 4, @@ -370,6 +419,10 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) nv50_screen_destroy(pscreen); return NULL; } + BEGIN_RING(chan, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); + OUT_RELOCh(chan, screen->constbuf_parm[i], 0, rl); + OUT_RELOCl(chan, screen->constbuf_parm[i], 0, rl); + OUT_RING (chan, ((NV50_CB_PVP + i) << 16) | 0x0800); } if (nouveau_resource_init(&screen->immd_heap[0], 0, 128) || @@ -381,80 +434,16 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) return NULL; } - /* - // map constant buffers: - // B = buffer ID (maybe more than 1 byte) - // N = CB index used in shader instruction - // P = program type (0 = VP, 2 = GP, 3 = FP) - so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); - so_data (so, 0x000BBNP1); - */ - - so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); - so_reloc (so, screen->constbuf_misc[0], 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->constbuf_misc[0], 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); - so_data (so, (NV50_CB_PMISC << 16) | 0x00000200); - so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); - so_data (so, 0x00000001 | (NV50_CB_PMISC << 12)); - so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); - so_data (so, 0x00000021 | (NV50_CB_PMISC << 12)); - so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); - so_data (so, 0x00000031 | (NV50_CB_PMISC << 12)); - - /* bind auxiliary constbuf to immediate data bo */ - so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); - so_reloc (so, screen->constbuf_misc[0], (128 * 4) * 4, - NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->constbuf_misc[0], (128 * 4) * 4, - NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); - so_data (so, (NV50_CB_AUX << 16) | 0x00000200); - so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); - so_data (so, 0x00000201 | (NV50_CB_AUX << 12)); - so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); - so_data (so, 0x00000221 | (NV50_CB_AUX << 12)); - - so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); - so_reloc (so, screen->constbuf_parm[PIPE_SHADER_VERTEX], 0, - NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->constbuf_parm[PIPE_SHADER_VERTEX], 0, - NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); - so_data (so, (NV50_CB_PVP << 16) | 0x00000800); - so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); - so_data (so, 0x00000101 | (NV50_CB_PVP << 12)); - - so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); - so_reloc (so, screen->constbuf_parm[PIPE_SHADER_GEOMETRY], 0, - NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->constbuf_parm[PIPE_SHADER_GEOMETRY], 0, - NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); - so_data (so, (NV50_CB_PGP << 16) | 0x00000800); - so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); - so_data (so, 0x00000121 | (NV50_CB_PGP << 12)); - - so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); - so_reloc (so, screen->constbuf_parm[PIPE_SHADER_FRAGMENT], 0, - NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->constbuf_parm[PIPE_SHADER_FRAGMENT], 0, - NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); - so_data (so, (NV50_CB_PFP << 16) | 0x00000800); - so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); - so_data (so, 0x00000131 | (NV50_CB_PFP << 12)); - ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, 3 * 32 * (8 * 4), &screen->tic); if (ret) { nv50_screen_destroy(pscreen); return NULL; } - - so_method(so, screen->tesla, NV50TCL_TIC_ADDRESS_HIGH, 3); - so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); - so_data (so, 3 * 32 - 1); + BEGIN_RING(chan, screen->tesla, NV50TCL_TIC_ADDRESS_HIGH, 3); + OUT_RELOCh(chan, screen->tic, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD); + OUT_RELOCl(chan, screen->tic, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD); + OUT_RING (chan, 3 * 32 - 1); ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, 3 * 32 * (8 * 4), &screen->tsc); @@ -462,37 +451,50 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) nv50_screen_destroy(pscreen); return NULL; } - - so_method(so, screen->tesla, NV50TCL_TSC_ADDRESS_HIGH, 3); - so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); - so_data (so, 0x00000000); /* ignored if TSC_LINKED (0x1234) = 1 */ - + BEGIN_RING(chan, screen->tesla, NV50TCL_TSC_ADDRESS_HIGH, 3); + OUT_RELOCh(chan, screen->tsc, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD); + OUT_RELOCl(chan, screen->tsc, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD); + OUT_RING (chan, 0); /* ignored if TSC_LINKED (0x1234) == 1 */ + + /* map constant buffers: + * B = buffer ID (maybe more than 1 byte) + * N = CB index used in shader instruction + * P = program type (0 = VP, 2 = GP, 3 = FP) + * SET_PROGRAM_CB = 0x000BBNP1 + */ + BEGIN_RING_NI(chan, screen->tesla, NV50TCL_SET_PROGRAM_CB, 8); + /* bind immediate buffer */ + OUT_RING (chan, 0x001 | (NV50_CB_PMISC << 12)); + OUT_RING (chan, 0x021 | (NV50_CB_PMISC << 12)); + OUT_RING (chan, 0x031 | (NV50_CB_PMISC << 12)); + /* bind auxiliary constbuf to immediate data bo */ + OUT_RING (chan, 0x201 | (NV50_CB_AUX << 12)); + OUT_RING (chan, 0x221 | (NV50_CB_AUX << 12)); + /* bind parameter buffers */ + OUT_RING (chan, 0x101 | (NV50_CB_PVP << 12)); + OUT_RING (chan, 0x121 | (NV50_CB_PGP << 12)); + OUT_RING (chan, 0x131 | (NV50_CB_PFP << 12)); /* Vertex array limits - max them out */ for (i = 0; i < 16; i++) { - so_method(so, screen->tesla, NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i), 2); - so_data (so, 0x000000ff); - so_data (so, 0xffffffff); + BEGIN_RING(chan, screen->tesla, + NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i), 2); + OUT_RING (chan, 0x000000ff); + OUT_RING (chan, 0xffffffff); } - so_method(so, screen->tesla, NV50TCL_DEPTH_RANGE_NEAR(0), 2); - so_data (so, fui(0.0)); - so_data (so, fui(1.0)); + BEGIN_RING(chan, screen->tesla, NV50TCL_DEPTH_RANGE_NEAR(0), 2); + OUT_RINGf (chan, 0.0f); + OUT_RINGf (chan, 1.0f); /* no dynamic combination of TIC & TSC entries => only BIND_TIC used */ - so_method(so, screen->tesla, NV50TCL_LINKED_TSC, 1); - so_data (so, 1); + BEGIN_RING(chan, screen->tesla, NV50TCL_LINKED_TSC, 1); + OUT_RING (chan, 1); - so_method(so, screen->tesla, NV50TCL_EDGEFLAG_ENABLE, 1); - so_data (so, 1); /* default edgeflag to TRUE */ + BEGIN_RING(chan, screen->tesla, NV50TCL_EDGEFLAG_ENABLE, 1); + OUT_RING (chan, 1); /* default edgeflag to TRUE */ - so_emit(chan, so); - so_ref (so, &screen->static_init); - so_ref (NULL, &so); - nouveau_pushbuf_flush(chan, 0); + FIRE_RING (chan); screen->force_push = debug_get_bool_option("NV50_ALWAYS_PUSH", FALSE); return pscreen; diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h index ec19ea655b1..15bd4eed399 100644 --- a/src/gallium/drivers/nv50/nv50_screen.h +++ b/src/gallium/drivers/nv50/nv50_screen.h @@ -27,8 +27,6 @@ struct nv50_screen { struct nouveau_bo *tic; struct nouveau_bo *tsc; - struct nouveau_stateobj *static_init; - boolean force_push; }; @@ -38,4 +36,6 @@ nv50_screen(struct pipe_screen *screen) return (struct nv50_screen *)screen; } +extern void nv50_screen_relocs(struct nv50_screen *); + #endif diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c index 63d73b5ce83..b7e355283c8 100644 --- a/src/gallium/drivers/nv50/nv50_state_validate.c +++ b/src/gallium/drivers/nv50/nv50_state_validate.c @@ -435,7 +435,7 @@ nv50_state_validate(struct nv50_context *nv50, unsigned wait_dwords) so_emit_reloc_markers(chan, nv50->state.hw[3]); /* vp */ so_emit_reloc_markers(chan, nv50->state.hw[4]); /* fp */ so_emit_reloc_markers(chan, nv50->state.hw[17]); /* vb */ - so_emit_reloc_markers(chan, nv50->screen->static_init); + nv50_screen_relocs(nv50->screen); /* No idea.. */ BEGIN_RING(chan, tesla, 0x142c, 1); -- 2.30.2