fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-20 11:40:10 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

radeonsi: use common build buffer descriptor helpers

Browse source 
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29385>
This commit is contained in:
Samuel Pitoiset 2024-05-24 14:29:03 +02:00
parent 3c5173aabf
commit 139bc6b813
4 changed files with 131 additions and 222 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

75
src/gallium/drivers/radeonsi/si_descriptors.c
View file

@ -1033,26 +1033,23 @@ static void si_init_buffer_resources(struct si_context *sctx,
si_init_descriptors(descs, shader_userdata_rel_index, 4, num_buffers); si_init_descriptors(descs, shader_userdata_rel_index, 4, num_buffers);
const struct ac_buffer_state buffer_state = {
.format = PIPE_FORMAT_R32_FLOAT,
.swizzle =
{
PIPE_SWIZZLE_X,
PIPE_SWIZZLE_Y,
PIPE_SWIZZLE_Z,
PIPE_SWIZZLE_W,
},
.gfx10_oob_select = V_008F0C_OOB_SELECT_RAW,
};
/* Initialize buffer descriptors, so that we don't have to do it at bind time. */ /* Initialize buffer descriptors, so that we don't have to do it at bind time. */
for (unsigned i = 0; i < num_buffers; i++) { for (unsigned i = 0; i < num_buffers; i++) {
uint32_t *desc = descs->list + i * 4; uint32_t *desc = descs->list + i * 4;
desc[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) | S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) | ac_set_buf_desc_word3(sctx->gfx_level, &buffer_state, &desc[3]);
S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) | S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W);
if (sctx->gfx_level >= GFX12) {
desc[3] |= S_008F0C_FORMAT_GFX12(V_008F0C_GFX11_FORMAT_32_FLOAT) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_RAW);
} else if (sctx->gfx_level >= GFX11) {
desc[3] |= S_008F0C_FORMAT_GFX10(V_008F0C_GFX11_FORMAT_32_FLOAT) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_RAW);
} else if (sctx->gfx_level >= GFX10) {
desc[3] |= S_008F0C_FORMAT_GFX10(V_008F0C_GFX10_FORMAT_32_FLOAT) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_RAW) | S_008F0C_RESOURCE_LEVEL(1);
} else {
desc[3] |= S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32);
}
} }
} }
@ -1482,37 +1479,31 @@ void si_set_ring_buffer(struct si_context *sctx, uint slot, struct pipe_resource
/* Set the descriptor. */ /* Set the descriptor. */
uint32_t *desc = descs->list + slot * 4; uint32_t *desc = descs->list + slot * 4;
desc[0] = va;
desc[1] = S_008F04_BASE_ADDRESS_HI(va >> 32) | S_008F04_STRIDE(stride); uint32_t swizzle_enable;
desc[2] = num_records;
desc[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) | S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) | S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) |
S_008F0C_INDEX_STRIDE(index_stride) | S_008F0C_ADD_TID_ENABLE(add_tid);
if (sctx->gfx_level >= GFX11) { if (sctx->gfx_level >= GFX11) {
assert(!swizzle || element_size == 1 || element_size == 3); /* 4 or 16 bytes */ swizzle_enable = swizzle ? element_size : 0;
desc[1] |= S_008F04_SWIZZLE_ENABLE_GFX11(swizzle ? element_size : 0);
} else if (sctx->gfx_level >= GFX9) {
assert(!swizzle || element_size == 1); /* only 4 bytes on GFX9 */
desc[1] |= S_008F04_SWIZZLE_ENABLE_GFX6(swizzle);
} else { } else {
desc[1] |= S_008F04_SWIZZLE_ENABLE_GFX6(swizzle); swizzle_enable = swizzle;
desc[3] |= S_008F0C_ELEMENT_SIZE(element_size);
} }
if (sctx->gfx_level >= GFX12) { const struct ac_buffer_state buffer_state = {
desc[3] |= S_008F0C_FORMAT_GFX12(V_008F0C_GFX11_FORMAT_32_FLOAT) | .va = va,
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_DISABLED); .size = num_records,
} else if (sctx->gfx_level >= GFX11) { .format = PIPE_FORMAT_R32_FLOAT,
desc[3] |= S_008F0C_FORMAT_GFX10(V_008F0C_GFX11_FORMAT_32_FLOAT) | .swizzle = {
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_DISABLED); PIPE_SWIZZLE_X, PIPE_SWIZZLE_Y, PIPE_SWIZZLE_Z, PIPE_SWIZZLE_W,
} else if (sctx->gfx_level >= GFX10) { },
desc[3] |= S_008F0C_FORMAT_GFX10(V_008F0C_GFX10_FORMAT_32_FLOAT) | .stride = stride,
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_DISABLED) | S_008F0C_RESOURCE_LEVEL(1); .swizzle_enable = swizzle_enable,
} else { .gfx10_oob_select = V_008F0C_OOB_SELECT_DISABLED,
desc[3] |= S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) | .index_stride = index_stride,
S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32); .element_size = element_size,
} .add_tid = add_tid,
};
ac_build_buffer_descriptor(sctx->gfx_level, &buffer_state, desc);
pipe_resource_reference(&buffers->buffers[slot], buffer); pipe_resource_reference(&buffers->buffers[slot], buffer);
radeon_add_to_buffer_list(sctx, &sctx->gfx_cs, si_resource(buffer), radeon_add_to_buffer_list(sctx, &sctx->gfx_cs, si_resource(buffer),

103
src/gallium/drivers/radeonsi/si_nir_lower_abi.c
View file

@ -44,20 +44,17 @@ static nir_def *build_attr_ring_desc(nir_builder *b, struct si_shader *shader,
ac_nir_load_arg(b, &args->ac, args->gs_attr_address); ac_nir_load_arg(b, &args->ac, args->gs_attr_address);
unsigned stride = 16 * shader->info.nr_param_exports; unsigned stride = 16 * shader->info.nr_param_exports;
uint32_t desc[4];
ac_build_attr_ring_descriptor(sel->screen->info.gfx_level,
(uint64_t)sel->screen->info.address32_hi << 32,
0xffffffff, stride, desc);
nir_def *comp[] = { nir_def *comp[] = {
attr_address, attr_address,
nir_imm_int(b, S_008F04_BASE_ADDRESS_HI(sel->screen->info.address32_hi) | nir_imm_int(b, desc[1]),
S_008F04_STRIDE(stride) | nir_imm_int(b, desc[2]),
S_008F04_SWIZZLE_ENABLE_GFX11(3) /* 16B */), nir_imm_int(b, desc[3]),
nir_imm_int(b, 0xffffffff),
nir_imm_int(b, S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) |
S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) |
(sel->screen->info.gfx_level >= GFX12 ?
S_008F0C_FORMAT_GFX12(V_008F0C_GFX11_FORMAT_32_32_32_32_FLOAT) :
S_008F0C_FORMAT_GFX10(V_008F0C_GFX11_FORMAT_32_32_32_32_FLOAT)) |
S_008F0C_INDEX_STRIDE(2) /* 32 elements */),
}; };
return nir_vec(b, comp, 4); return nir_vec(b, comp, 4);
@ -133,33 +130,17 @@ static nir_def *build_tess_ring_desc(nir_builder *b, struct si_screen *screen,
struct si_shader_args *args) struct si_shader_args *args)
{ {
nir_def *addr = ac_nir_load_arg(b, &args->ac, args->tes_offchip_addr); nir_def *addr = ac_nir_load_arg(b, &args->ac, args->tes_offchip_addr);
uint32_t desc[4];
uint32_t rsrc3 = ac_build_raw_buffer_descriptor(screen->info.gfx_level,
S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) | (uint64_t)screen->info.address32_hi << 32,
S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) | 0xffffffff, desc);
S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W);
if (screen->info.gfx_level >= GFX12) {
rsrc3 |= S_008F0C_FORMAT_GFX12(V_008F0C_GFX11_FORMAT_32_FLOAT) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_RAW);
} else if (screen->info.gfx_level >= GFX11) {
rsrc3 |= S_008F0C_FORMAT_GFX10(V_008F0C_GFX11_FORMAT_32_FLOAT) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_RAW);
} else if (screen->info.gfx_level >= GFX10) {
rsrc3 |= S_008F0C_FORMAT_GFX10(V_008F0C_GFX10_FORMAT_32_FLOAT) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_RAW) |
S_008F0C_RESOURCE_LEVEL(1);
} else {
rsrc3 |= S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32);
}
nir_def *comp[4] = { nir_def *comp[4] = {
addr, addr,
nir_imm_int(b, S_008F04_BASE_ADDRESS_HI(screen->info.address32_hi)), nir_imm_int(b, desc[1]),
nir_imm_int(b, 0xffffffff), nir_imm_int(b, desc[2]),
nir_imm_int(b, rsrc3), nir_imm_int(b, desc[3]),
}; };
return nir_vec(b, comp, 4); return nir_vec(b, comp, 4);
@ -215,46 +196,34 @@ static void build_gsvs_ring_desc(nir_builder *b, struct lower_abi_state *s)
if (!num_components) if (!num_components)
continue; continue;
nir_def *desc[4];
desc[0] = nir_unpack_64_2x32_split_x(b, base_addr);
desc[1] = nir_unpack_64_2x32_split_y(b, base_addr);
unsigned stride = 4 * num_components * sel->info.base.gs.vertices_out; unsigned stride = 4 * num_components * sel->info.base.gs.vertices_out;
/* Limit on the stride field for <= GFX7. */ /* Limit on the stride field for <= GFX7. */
assert(stride < (1 << 14)); assert(stride < (1 << 14));
desc[1] = nir_ior_imm(
b, desc[1], S_008F04_STRIDE(stride) | S_008F04_SWIZZLE_ENABLE_GFX6(1));
unsigned num_records = s->shader->wave_size; unsigned num_records = s->shader->wave_size;
desc[2] = nir_imm_int(b, num_records);
uint32_t rsrc3 = const struct ac_buffer_state buffer_state = {
S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) | .size = num_records,
S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) | .format = PIPE_FORMAT_R32_FLOAT,
S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) | .swizzle = {
S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) | PIPE_SWIZZLE_X, PIPE_SWIZZLE_Y, PIPE_SWIZZLE_Z, PIPE_SWIZZLE_W,
S_008F0C_INDEX_STRIDE(1) | /* index_stride = 16 (elements) */ },
S_008F0C_ADD_TID_ENABLE(1); .stride = stride,
.swizzle_enable = true,
.element_size = 1,
.index_stride = 1,
.add_tid = true,
.gfx10_oob_select = V_008F0C_OOB_SELECT_DISABLED,
};
uint32_t tmp_desc[4];
if (sel->screen->info.gfx_level >= GFX10) { ac_build_buffer_descriptor(sel->screen->info.gfx_level, &buffer_state, tmp_desc);
rsrc3 |=
S_008F0C_FORMAT_GFX10(V_008F0C_GFX10_FORMAT_32_FLOAT) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_DISABLED) |
S_008F0C_RESOURCE_LEVEL(1);
} else {
/* If MUBUF && ADD_TID_ENABLE, DATA_FORMAT means STRIDE[14:17] on gfx8-9, so set 0. */
unsigned data_format =
sel->screen->info.gfx_level == GFX8 || sel->screen->info.gfx_level == GFX9 ?
0 : V_008F0C_BUF_DATA_FORMAT_32;
rsrc3 |= nir_def *desc[4];
S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) | desc[0] = nir_unpack_64_2x32_split_x(b, base_addr);
S_008F0C_DATA_FORMAT(data_format) | desc[1] = nir_ior_imm(b, nir_unpack_64_2x32_split_y(b, base_addr), tmp_desc[1]);
S_008F0C_ELEMENT_SIZE(1); /* element_size = 4 (bytes) */ desc[2] = nir_imm_int(b, tmp_desc[2]);
} desc[3] = nir_imm_int(b, tmp_desc[3]);
desc[3] = nir_imm_int(b, rsrc3);
s->gsvs_ring[stream] = nir_vec(b, desc, 4); s->gsvs_ring[stream] = nir_vec(b, desc, 4);

37
src/gallium/drivers/radeonsi/si_nir_lower_resource.c
View file

@ -28,28 +28,25 @@ struct lower_resource_state {
static nir_def *load_ubo_desc_fast_path(nir_builder *b, nir_def *addr_lo, static nir_def *load_ubo_desc_fast_path(nir_builder *b, nir_def *addr_lo,
struct si_shader_selector *sel) struct si_shader_selector *sel)
{ {
nir_def *addr_hi = const struct ac_buffer_state buffer_state = {
nir_imm_int(b, S_008F04_BASE_ADDRESS_HI(sel->screen->info.address32_hi)); .va = (uint64_t)sel->screen->info.address32_hi << 32,
.size = sel->info.constbuf0_num_slots * 16,
.format = PIPE_FORMAT_R32_FLOAT,
.swizzle =
{
PIPE_SWIZZLE_X,
PIPE_SWIZZLE_Y,
PIPE_SWIZZLE_Z,
PIPE_SWIZZLE_W,
},
.gfx10_oob_select = V_008F0C_OOB_SELECT_RAW,
};
uint32_t desc[4];
uint32_t rsrc3 = ac_build_buffer_descriptor(sel->screen->info.gfx_level, &buffer_state, desc);
S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) | S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) | S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W);
if (sel->screen->info.gfx_level >= GFX12) return nir_vec4(b, addr_lo, nir_imm_int(b, desc[1]), nir_imm_int(b, desc[2]),
rsrc3 |= S_008F0C_FORMAT_GFX12(V_008F0C_GFX11_FORMAT_32_FLOAT) | nir_imm_int(b, desc[3]));
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_RAW);
else if (sel->screen->info.gfx_level >= GFX11)
rsrc3 |= S_008F0C_FORMAT_GFX10(V_008F0C_GFX11_FORMAT_32_FLOAT) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_RAW);
else if (sel->screen->info.gfx_level >= GFX10)
rsrc3 |= S_008F0C_FORMAT_GFX10(V_008F0C_GFX10_FORMAT_32_FLOAT) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_RAW) | S_008F0C_RESOURCE_LEVEL(1);
else
rsrc3 |= S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32);
return nir_vec4(b, addr_lo, addr_hi, nir_imm_int(b, sel->info.constbuf0_num_slots * 16),
nir_imm_int(b, rsrc3));
} }
static nir_def *clamp_index(nir_builder *b, nir_def *index, unsigned max) static nir_def *clamp_index(nir_builder *b, nir_def *index, unsigned max)

138
src/gallium/drivers/radeonsi/si_state.c
View file

@ -4015,52 +4015,21 @@ void si_make_buffer_descriptor(struct si_screen *screen, struct si_resource *buf
if (screen->info.gfx_level == GFX8) if (screen->info.gfx_level == GFX8)
num_records *= stride; num_records *= stride;
state[4] = 0; const struct ac_buffer_state buffer_state = {
state[5] = S_008F04_STRIDE(stride); .size = num_records,
state[6] = num_records; .format = format,
state[7] = S_008F0C_DST_SEL_X(ac_map_swizzle(desc->swizzle[0])) | .swizzle =
S_008F0C_DST_SEL_Y(ac_map_swizzle(desc->swizzle[1])) | {
S_008F0C_DST_SEL_Z(ac_map_swizzle(desc->swizzle[2])) | desc->swizzle[0],
S_008F0C_DST_SEL_W(ac_map_swizzle(desc->swizzle[3])); desc->swizzle[1],
desc->swizzle[2],
desc->swizzle[3],
},
.stride = stride,
.gfx10_oob_select = V_008F0C_OOB_SELECT_STRUCTURED_WITH_OFFSET,
};
if (screen->info.gfx_level >= GFX10) { ac_build_buffer_descriptor(screen->info.gfx_level, &buffer_state, &state[4]);
const struct gfx10_format *fmt = &ac_get_gfx10_format_table(screen->info.gfx_level)[format];
/* OOB_SELECT chooses the out-of-bounds check.
*
* GFX10:
* - 0: (index >= NUM_RECORDS) || (offset >= STRIDE)
* - 1: index >= NUM_RECORDS
* - 2: NUM_RECORDS == 0
* - 3: if SWIZZLE_ENABLE:
* swizzle_address >= NUM_RECORDS
* else:
* offset >= NUM_RECORDS
*
* GFX11:
* - 0: (index >= NUM_RECORDS) || (offset+payload > STRIDE)
* - 1: index >= NUM_RECORDS
* - 2: NUM_RECORDS == 0
* - 3: if SWIZZLE_ENABLE && STRIDE:
* (index >= NUM_RECORDS) || ( offset+payload > STRIDE)
* else:
* offset+payload > NUM_RECORDS
*/
state[7] |= (screen->info.gfx_level >= GFX12 ?
S_008F0C_FORMAT_GFX12(fmt->img_format) :
S_008F0C_FORMAT_GFX10(fmt->img_format)) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_STRUCTURED_WITH_OFFSET) |
S_008F0C_RESOURCE_LEVEL(screen->info.gfx_level < GFX11);
} else {
int first_non_void;
unsigned num_format, data_format;
first_non_void = util_format_get_first_non_void_channel(format);
num_format = si_translate_buffer_numformat(&screen->b, desc, first_non_void);
data_format = si_translate_buffer_dataformat(&screen->b, desc, first_non_void);
state[7] |= S_008F0C_NUM_FORMAT(num_format) | S_008F0C_DATA_FORMAT(data_format);
}
} }
/** /**
@ -4128,29 +4097,21 @@ static void cdna_emu_make_image_descriptor(struct si_screen *screen, struct si_t
util_format_compose_swizzles(desc->swizzle, state_swizzle, swizzle); util_format_compose_swizzles(desc->swizzle, state_swizzle, swizzle);
/* Buffer descriptor */ /* Buffer descriptor */
state[0] = 0; const struct ac_buffer_state buffer_state = {
state[1] = S_008F04_STRIDE(stride); .size = num_records,
state[2] = num_records; .format = pipe_format,
state[3] = S_008F0C_DST_SEL_X(ac_map_swizzle(swizzle[0])) | .swizzle =
S_008F0C_DST_SEL_Y(ac_map_swizzle(swizzle[1])) | {
S_008F0C_DST_SEL_Z(ac_map_swizzle(swizzle[2])) | desc->swizzle[0],
S_008F0C_DST_SEL_W(ac_map_swizzle(swizzle[3])); desc->swizzle[1],
desc->swizzle[2],
desc->swizzle[3],
},
.stride = stride,
.gfx10_oob_select = V_008F0C_OOB_SELECT_STRUCTURED_WITH_OFFSET,
};
if (screen->info.gfx_level >= GFX10) { ac_build_buffer_descriptor(screen->info.gfx_level, &buffer_state, &state[0]);
const struct gfx10_format *fmt = &ac_get_gfx10_format_table(screen->info.gfx_level)[pipe_format];
state[3] |= (screen->info.gfx_level >= GFX12 ? S_008F0C_FORMAT_GFX12(fmt->img_format) :
S_008F0C_FORMAT_GFX10(fmt->img_format)) |
S_008F0C_OOB_SELECT(V_008F0C_OOB_SELECT_STRUCTURED_WITH_OFFSET) |
S_008F0C_RESOURCE_LEVEL(screen->info.gfx_level < GFX11);
} else {
int first_non_void = util_format_get_first_non_void_channel(pipe_format);
unsigned num_format = si_translate_buffer_numformat(&screen->b, desc, first_non_void);
unsigned data_format = si_translate_buffer_dataformat(&screen->b, desc, first_non_void);
state[3] |= S_008F0C_NUM_FORMAT(num_format) |
S_008F0C_DATA_FORMAT(data_format);
}
/* Additional fields used by image opcode emulation. */ /* Additional fields used by image opcode emulation. */
state[4] = width | (height << 16); state[4] = width | (height << 16);
@ -5050,33 +5011,24 @@ static void *si_create_vertex_elements(struct pipe_context *ctx, unsigned count,
v->vb_alignment_check_mask |= 1 << vbo_index; v->vb_alignment_check_mask |= 1 << vbo_index;
} }
v->elem[i].rsrc_word3 = S_008F0C_DST_SEL_X(ac_map_swizzle(desc->swizzle[0])) | const struct ac_buffer_state buffer_state = {
S_008F0C_DST_SEL_Y(ac_map_swizzle(desc->swizzle[1])) | .format = elements[i].src_format,
S_008F0C_DST_SEL_Z(ac_map_swizzle(desc->swizzle[2])) | .swizzle =
S_008F0C_DST_SEL_W(ac_map_swizzle(desc->swizzle[3])); {
desc->swizzle[0],
desc->swizzle[1],
desc->swizzle[2],
desc->swizzle[3],
},
/* OOB_SELECT chooses the out-of-bounds check:
* - 1: index >= NUM_RECORDS (Structured)
* - 3: offset >= NUM_RECORDS (Raw)
*/
.gfx10_oob_select = v->elem[i].stride ? V_008F0C_OOB_SELECT_STRUCTURED
: V_008F0C_OOB_SELECT_RAW,
};
if (sscreen->info.gfx_level >= GFX10) { ac_set_buf_desc_word3(sscreen->info.gfx_level, &buffer_state, &v->elem[i].rsrc_word3);
const struct gfx10_format *fmt = &ac_get_gfx10_format_table(sscreen->info.gfx_level)[elements[i].src_format];
ASSERTED unsigned last_vertex_format = sscreen->info.gfx_level >= GFX11 ? 64 : 128;
assert(fmt->img_format != 0 && fmt->img_format < last_vertex_format);
v->elem[i].rsrc_word3 |=
(sscreen->info.gfx_level >= GFX12 ?
S_008F0C_FORMAT_GFX12(fmt->img_format) :
S_008F0C_FORMAT_GFX10(fmt->img_format)) |
S_008F0C_RESOURCE_LEVEL(sscreen->info.gfx_level < GFX11) |
/* OOB_SELECT chooses the out-of-bounds check:
* - 1: index >= NUM_RECORDS (Structured)
* - 3: offset >= NUM_RECORDS (Raw)
*/
S_008F0C_OOB_SELECT(v->elem[i].stride ? V_008F0C_OOB_SELECT_STRUCTURED
: V_008F0C_OOB_SELECT_RAW);
} else {
unsigned data_format, num_format;
data_format = si_translate_buffer_dataformat(ctx->screen, desc, first_non_void);
num_format = si_translate_buffer_numformat(ctx->screen, desc, first_non_void);
v->elem[i].rsrc_word3 |= S_008F0C_NUM_FORMAT(num_format) |
S_008F0C_DATA_FORMAT(data_format);
}
} }
if (v->instance_divisor_is_fetched) { if (v->instance_divisor_is_fetched) {