r300: convert to new relocations format (see libdrm-radeon)

This commit is contained in:
Jerome Glisse 2008-11-12 17:00:28 +01:00
parent d07d137931
commit c26ec97b13
7 changed files with 106 additions and 50 deletions

View file

@ -80,7 +80,7 @@ COMMON_SYMLINKS = \
radeon_bo_legacy.h \
radeon_cs_legacy.h
DRI_LIB_DEPS += -ldrm_radeon
DRI_LIB_DEPS += -ldrm-radeon
##### TARGETS #####

View file

@ -297,12 +297,14 @@ static void emit_tex_offsets(r300ContextPtr r300, struct r300_state_atom * atom)
OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1);
r300TexObj *t = r300->hw.textures[i];
if (t && !t->image_override) {
OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0, 0);
OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0,
RADEON_GEM_DOMAIN_VRAM, 0, 0);
} else if (!t) {
OUT_BATCH(r300->radeon.radeonScreen->texOffset[0]);
} else {
if (t->bo) {
OUT_BATCH_RELOC(t->tile_bits, t->bo, 0, 0);
OUT_BATCH_RELOC(t->tile_bits, t->bo, 0,
RADEON_GEM_DOMAIN_VRAM, 0, 0);
} else {
OUT_BATCH(t->override_offset);
}
@ -339,7 +341,7 @@ static void emit_cb_offset(r300ContextPtr r300, struct r300_state_atom * atom)
BEGIN_BATCH(4);
OUT_BATCH_REGSEQ(R300_RB3D_COLOROFFSET0, 1);
OUT_BATCH_RELOC(0, rrb->bo, 0, 0);
OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
OUT_BATCH_REGSEQ(R300_RB3D_COLORPITCH0, 1);
OUT_BATCH(cbpitch);
END_BATCH();
@ -365,7 +367,7 @@ static void emit_zb_offset(r300ContextPtr r300, struct r300_state_atom * atom)
BEGIN_BATCH(4);
OUT_BATCH_REGSEQ(R300_ZB_DEPTHOFFSET, 1);
OUT_BATCH_RELOC(0, rrb->bo, 0, 0);
OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
OUT_BATCH_REGVAL(R300_ZB_DEPTHPITCH, zbpitch);
END_BATCH();
}

View file

@ -84,14 +84,20 @@ void r300BeginBatch(r300ContextPtr r300,
/**
* Write a relocated dword to the command buffer.
*/
#define OUT_BATCH_RELOC(data, bo, offset, flags) \
#define OUT_BATCH_RELOC(data, bo, offset, rd, wd, flags) \
do { \
if (offset) {\
fprintf(stderr, "(%s:%s:%d) offset : %d\n",\
__FILE__, __FUNCTION__, __LINE__, offset);\
}\
radeon_cs_write_dword(b_l_r300->cmdbuf.cs, offset);\
radeon_cs_write_reloc(b_l_r300->cmdbuf.cs,bo,0,(bo)->size,flags);\
radeon_cs_write_reloc(b_l_r300->cmdbuf.cs, \
bo, \
offset, \
(bo)->size, \
rd, \
wd, \
flags);\
} while(0)
/**

View file

@ -101,7 +101,7 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags,
assert(rrb != 0);
BEGIN_BATCH_NO_AUTOSTATE(4);
OUT_BATCH_REGSEQ(R300_RB3D_COLOROFFSET0, 1);
OUT_BATCH_RELOC(0, rrb->bo, 0, 0);
OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
OUT_BATCH_REGVAL(R300_RB3D_COLORPITCH0, cbpitch);
END_BATCH();
}
@ -117,7 +117,7 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags,
}
BEGIN_BATCH_NO_AUTOSTATE(4);
OUT_BATCH_REGSEQ(R300_ZB_DEPTHOFFSET, 1);
OUT_BATCH_RELOC(0, rrbd->bo, 0, 0);
OUT_BATCH_RELOC(0, rrbd->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
OUT_BATCH_REGVAL(R300_ZB_DEPTHPITCH, cbpitch);
END_BATCH();
}

View file

@ -202,8 +202,10 @@ static void r300FireEB(r300ContextPtr rmesa, int vertex_count, int type)
if (!rmesa->radeon.radeonScreen->driScreen->dri2.enabled) {
OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2);
OUT_BATCH(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
OUT_BATCH_RELOC(0, rmesa->state.elt_dma_bo,
rmesa->state.elt_dma_offset, 0);
OUT_BATCH_RELOC(rmesa->state.elt_dma_offset,
rmesa->state.elt_dma_bo,
rmesa->state.elt_dma_offset,
RADEON_GEM_DOMAIN_GTT, 0, 0);
OUT_BATCH(vertex_count);
} else {
OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2);
@ -214,7 +216,7 @@ static void r300FireEB(r300ContextPtr rmesa, int vertex_count, int type)
rmesa->state.elt_dma_bo,
0,
rmesa->state.elt_dma_bo->size,
0);
RADEON_GEM_DOMAIN_GTT, 0, 0);
}
END_BATCH();
}
@ -245,17 +247,30 @@ static void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset)
voffset = rmesa->state.aos[i + 0].offset +
offset * 4 * rmesa->state.aos[i + 0].stride;
OUT_BATCH_RELOC(0, rmesa->state.aos[i].bo, voffset, 0);
OUT_BATCH_RELOC(voffset,
rmesa->state.aos[i].bo,
voffset,
RADEON_GEM_DOMAIN_GTT,
0, 0);
voffset = rmesa->state.aos[i + 1].offset +
offset * 4 * rmesa->state.aos[i + 1].stride;
OUT_BATCH_RELOC(0, rmesa->state.aos[i+1].bo, voffset, 0);
OUT_BATCH_RELOC(voffset,
rmesa->state.aos[i+1].bo,
voffset,
RADEON_GEM_DOMAIN_GTT,
0, 0);
}
if (nr & 1) {
OUT_BATCH((rmesa->state.aos[nr - 1].components << 0) |
(rmesa->state.aos[nr - 1].stride << 8));
OUT_BATCH_RELOC(0, rmesa->state.aos[nr - 1].bo,
rmesa->state.aos[nr - 1].offset + offset * 4 * rmesa->state.aos[nr - 1].stride, 0);
voffset = rmesa->state.aos[nr - 1].offset +
offset * 4 * rmesa->state.aos[nr - 1].stride;
OUT_BATCH_RELOC(voffset,
rmesa->state.aos[nr - 1].bo,
voffset,
RADEON_GEM_DOMAIN_GTT,
0, 0);
}
} else {
for (i = 0; i + 1 < nr; i += 2) {
@ -275,32 +290,37 @@ static void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset)
if (nr & 1) {
OUT_BATCH((rmesa->state.aos[nr - 1].components << 0) |
(rmesa->state.aos[nr - 1].stride << 8));
OUT_BATCH(rmesa->state.aos[nr - 1].offset + offset * 4 *
rmesa->state.aos[nr - 1].stride);
voffset = rmesa->state.aos[nr - 1].offset +
offset * 4 * rmesa->state.aos[nr - 1].stride;
OUT_BATCH(voffset);
}
for (i = 0; i + 1 < nr; i += 2) {
#if 0
fprintf(stderr, "3D_LOAD_VBPNTR 0x%08X & 0x%08X\n",
rmesa->state.aos[i+0].bo->handle,
rmesa->state.aos[i+1].bo->handle);
#endif
voffset = rmesa->state.aos[i + 0].offset +
offset * 4 * rmesa->state.aos[i + 0].stride;
radeon_cs_write_reloc(rmesa->cmdbuf.cs,
rmesa->state.aos[i+0].bo,
0,
voffset,
rmesa->state.aos[i+0].bo->size,
0);
RADEON_GEM_DOMAIN_GTT,
0, 0);
voffset = rmesa->state.aos[i + 1].offset +
offset * 4 * rmesa->state.aos[i + 1].stride;
radeon_cs_write_reloc(rmesa->cmdbuf.cs,
rmesa->state.aos[i+1].bo,
0,
voffset,
rmesa->state.aos[i+1].bo->size,
0);
RADEON_GEM_DOMAIN_GTT,
0, 0);
}
if (nr & 1) {
voffset = rmesa->state.aos[nr - 1].offset +
offset * 4 * rmesa->state.aos[nr - 1].stride;
radeon_cs_write_reloc(rmesa->cmdbuf.cs,
rmesa->state.aos[nr-1].bo,
0,
voffset,
rmesa->state.aos[nr-1].bo->size,
0);
RADEON_GEM_DOMAIN_GTT,
0, 0);
}
}
END_BATCH();

View file

@ -646,7 +646,7 @@ void r300EmitVertexAOS(r300ContextPtr rmesa, GLuint vertex_size, struct radeon_b
OUT_BATCH_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, 2);
OUT_BATCH(1);
OUT_BATCH(vertex_size | (vertex_size << 8));
OUT_BATCH_RELOC(0, bo, offset, 0);
OUT_BATCH_RELOC(offset, bo, offset, RADEON_GEM_DOMAIN_GTT, 0, 0);
END_BATCH();
}

View file

@ -94,19 +94,34 @@ static int cs_write_dword(struct radeon_cs *cs, uint32_t dword)
static int cs_write_reloc(struct radeon_cs *cs,
struct radeon_bo *bo,
uint32_t soffset,
uint32_t eoffset,
uint32_t domains)
uint32_t start_offset,
uint32_t end_offset,
uint32_t read_domain,
uint32_t write_domain,
uint32_t flags)
{
struct cs_reloc_legacy *relocs;
int i;
relocs = (struct cs_reloc_legacy *)cs->relocs;
/* check reloc window */
if (eoffset > bo->size) {
/* check domains */
if ((read_domain && write_domain) || (!read_domain && !write_domain)) {
/* in one CS a bo can only be in read or write domain but not
* in read & write domain at the same sime
*/
return -EINVAL;
}
if (soffset > eoffset) {
if (read_domain == RADEON_GEM_DOMAIN_CPU) {
return -EINVAL;
}
if (write_domain == RADEON_GEM_DOMAIN_CPU) {
return -EINVAL;
}
/* check reloc window */
if (end_offset > bo->size) {
return -EINVAL;
}
if (start_offset > end_offset) {
return -EINVAL;
}
/* check if bo is already referenced */
@ -114,16 +129,28 @@ static int cs_write_reloc(struct radeon_cs *cs,
uint32_t *indices;
if (relocs[i].base.bo->handle == bo->handle) {
/* update start offset and size */
if (eoffset > relocs[i].base.eoffset) {
relocs[i].base.eoffset = eoffset;
/* update start and end offset */
if (start_offset < relocs[i].base.start_offset) {
relocs[i].base.start_offset = start_offset;
}
if (soffset < relocs[i].base.soffset) {
relocs[i].base.soffset = soffset;
if (end_offset > relocs[i].base.end_offset) {
relocs[i].base.end_offset = end_offset;
}
relocs[i].base.size = relocs[i].base.eoffset -
relocs[i].base.soffset;
relocs[i].base.domains |= domains;
/* Check domains must be in read or write. As we check already
* checked that in argument one of the read or write domain was
* set we only need to check that if previous reloc as the read
* domain set then the read_domain should also be set for this
* new relocation.
*/
if (relocs[i].base.read_domain && !read_domain) {
return -EINVAL;
}
if (relocs[i].base.write_domain && !write_domain) {
return -EINVAL;
}
relocs[i].base.read_domain |= read_domain;
relocs[i].base.write_domain |= write_domain;
/* save indice */
relocs[i].cindices += 1;
indices = (uint32_t*)realloc(relocs[i].indices,
relocs[i].cindices * 4);
@ -145,10 +172,11 @@ static int cs_write_reloc(struct radeon_cs *cs,
}
cs->relocs = relocs;
relocs[cs->crelocs].base.bo = bo;
relocs[cs->crelocs].base.soffset = soffset;
relocs[cs->crelocs].base.eoffset = eoffset;
relocs[cs->crelocs].base.size = eoffset - soffset;
relocs[cs->crelocs].base.domains = domains;
relocs[cs->crelocs].base.start_offset = start_offset;
relocs[cs->crelocs].base.end_offset = end_offset;
relocs[cs->crelocs].base.read_domain = read_domain;
relocs[cs->crelocs].base.write_domain = write_domain;
relocs[cs->crelocs].base.flags = flags;
relocs[cs->crelocs].indices = (uint32_t*)malloc(4);
if (relocs[cs->crelocs].indices == NULL) {
return -ENOMEM;
@ -221,8 +249,8 @@ static int cs_process_relocs(struct radeon_cs *cs)
for (j = 0; j < relocs[i].cindices; j++) {
uint32_t soffset, eoffset;
soffset = relocs[i].base.soffset;
eoffset = relocs[i].base.eoffset;
soffset = relocs[i].base.start_offset;
eoffset = relocs[i].base.end_offset;
r = radeon_bo_legacy_validate(relocs[i].base.bo,
&soffset, &eoffset);
if (r) {