wine/vkd3d
0
0
Fork 0
mirror of https://gitlab.winehq.org/wine/vkd3d.git synced 2025-04-13 05:43:18 -07:00
Code Issues Packages Projects Releases Wiki Activity

vkd3d-shader/ir: Handle index ranges that do not touch a signature element for each register.

Browse Source 
A good part of the I/O normaliser job is to merge together signature
elements that are spanned by DCL_INDEX_RANGE instructions. The
current algorithm assumes that each index range touches exactly
one signature element for each index spanned by the range. The
assumption is used in shader_signature_merge() in the form of
expecting that, if the index range is N registers long, then,
once you find the first signature element of an index range, the
other elements that will have to be merged with it are exactly the
following N-1 according to the order given by
signature_element_register_compare() or
signature_element_mask_compare(), depending on the signature type.

This doesn't necessarily happen. For example, The Falconeer has a
few hull shaders in which this happens:

    hs_fork_phase
    dcl_hs_fork_phase_instance_count 13
    dcl_input vForkInstanceId
    dcl_output o4.z
    dcl_output o5.z
    dcl_output o6.z
    dcl_output o7.z
    dcl_output o12.z
    dcl_output o13.z
    dcl_output o14.z
    dcl_output o15.z
    dcl_output o16.z
    dcl_output o17.z
    dcl_output o18.z
    dcl_output o19.z
    dcl_output o20.z
    dcl_temps 1
    dcl_index_range o4.z 17
    iadd r0.x, vForkInstanceId.x, l(4)
    ult r0.y, vForkInstanceId.x, l(4)
    movc r0.x, r0.y, vForkInstanceId.x, r0.x
    mov o[r0.x + 4].z, l(0)
    ret

Here the index range "skips" o8.z through o11.z, because those
registers only use mask .xy. The current algorithm fails on such
a shader.

Even depending on the signature element order doesn't look ideal.
I don't have a full counterexample for that, but it looks fragile,
especially given that the register allocation algorithm in FXC is
notoriously full of unexpected corner cases.

We solve both problems by slightly changing the architecture of
the normaliser: first we move computing the masks for the merge
signature element from signature_element_range_expand_mask(),
which is executed while merging signature, to
io_normaliser_add_index_range(), which is executed before merging
signatures. Then, while we are merging signatures, we can decide
for each single signature element whether it has to be retained
or not, and how it should be patched. The algorithm becomes
independent of the order, because each signature element can be
processed individually.
This commit is contained in:
Giovanni Mascellani
2024-12-19 10:53:33 +01:00
committed by Henri Verbeet
parent b5350f9387
commit 4b84fb486b
Notes: Henri Verbeet 2025-02-19 18:01:43 +01:00 
Approved-by: Giovanni Mascellani (@giomasce)
Approved-by: Henri Verbeet (@hverbeet)
Approved-by: Francisco Casas (@fcasas)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/1332
1 changed files with 53 additions and 74 deletions
Download Patch File Download Diff File Expand all files Collapse all files

127
libs/vkd3d-shader/ir.c
View File

@ -1737,6 +1737,8 @@ struct io_normaliser_register_data
struct
{
uint8_t register_count;
uint32_t mask;
uint32_t used_mask;
} component[VKD3D_VEC4_SIZE];
};
@ -1812,13 +1814,13 @@ static unsigned int range_map_get_register_count(struct io_normaliser_register_d
static enum vkd3d_result range_map_set_register_range(struct io_normaliser *normaliser,
struct io_normaliser_register_data range_map[], unsigned int register_idx,
unsigned int register_count, uint32_t write_mask, bool is_dcl_indexrange)
unsigned int register_count, uint32_t mask, uint32_t used_mask, bool is_dcl_indexrange)
{
unsigned int i, j, r, c, component_idx, component_count;
VKD3D_ASSERT(write_mask <= VKD3DSP_WRITEMASK_ALL);
component_idx = vsir_write_mask_get_component_idx(write_mask);
component_count = vsir_write_mask_component_count(write_mask);
VKD3D_ASSERT(mask <= VKD3DSP_WRITEMASK_ALL);
component_idx = vsir_write_mask_get_component_idx(mask);
component_count = vsir_write_mask_component_count(mask);
VKD3D_ASSERT(register_idx < MAX_REG_OUTPUT && MAX_REG_OUTPUT - register_idx >= register_count);
@ -1840,6 +1842,8 @@ static enum vkd3d_result range_map_set_register_range(struct io_normaliser *norm
return VKD3D_OK;
}
range_map[register_idx].component[component_idx].register_count = register_count;
range_map[register_idx].component[component_idx].mask = mask;
range_map[register_idx].component[component_idx].used_mask = used_mask;
for (i = 0; i < register_count; ++i)
{
@ -1858,6 +1862,8 @@ static enum vkd3d_result range_map_set_register_range(struct io_normaliser *norm
return VKD3D_ERROR_INVALID_SHADER;
}
range_map[r].component[c].register_count = UINT8_MAX;
range_map[r].component[c].mask = mask;
range_map[r].component[c].used_mask = used_mask;
}
}
@ -1869,10 +1875,10 @@ static enum vkd3d_result io_normaliser_add_index_range(struct io_normaliser *nor
{
const struct vkd3d_shader_index_range *range = &ins->declaration.index_range;
const struct vkd3d_shader_register *reg = &range->dst.reg;
struct io_normaliser_register_data *range_map;
const struct shader_signature *signature;
struct io_normaliser_register_data (*range_map);
struct signature_element *element;
unsigned int reg_idx, write_mask;
uint32_t mask, used_mask;
unsigned int reg_idx, i;
switch (reg->type)
{
@ -1903,9 +1909,21 @@ static enum vkd3d_result io_normaliser_add_index_range(struct io_normaliser *nor
}
reg_idx = reg->idx[reg->idx_count - 1].offset;
write_mask = range->dst.write_mask;
element = vsir_signature_find_element_for_reg(signature, reg_idx, write_mask);
return range_map_set_register_range(normaliser, range_map, reg_idx, range->register_count, element->mask, true);
mask = range->dst.write_mask;
used_mask = 0;
for (i = 0; i < range->register_count; ++i)
{
struct signature_element *element;
if ((element = vsir_signature_find_element_for_reg(signature, reg_idx + i, mask)))
{
mask |= element->mask;
used_mask |= element->used_mask;
}
}
return range_map_set_register_range(normaliser, range_map, reg_idx, range->register_count, mask, used_mask, true);
}
static int signature_element_mask_compare(const void *a, const void *b)
@ -1959,7 +1977,7 @@ static enum vkd3d_result shader_signature_map_patch_constant_index_ranges(struct
continue;
if ((ret = range_map_set_register_range(normaliser, range_map,
e->register_index, register_count, e->mask, false) < 0))
e->register_index, register_count, e->mask, e->used_mask, false) < 0))
return ret;
}
@ -2006,51 +2024,6 @@ static int signature_element_index_compare(const void *a, const void *b)
return vkd3d_u32_compare(e->sort_index, f->sort_index);
}
static unsigned int signature_element_range_expand_mask(struct signature_element *e, unsigned int register_count,
struct io_normaliser_register_data range_map[])
{
unsigned int i, j, component_idx, component_count, merged_write_mask = e->mask;
/* dcl_indexrange instructions can declare a subset of the full mask, and the masks of
* the elements within the range may differ. TPF's handling of arrayed inputs with
* dcl_indexrange is really just a hack. Here we create a mask which covers all element
* masks, and check for collisions with other ranges. */
for (i = 1; i < register_count; ++i)
merged_write_mask |= e[i].mask;
if (merged_write_mask == e->mask)
return merged_write_mask;
/* Reaching this point is very rare to begin with, and collisions are even rarer or
* impossible. If the latter shows up, the fallback in shader_signature_find_element_for_reg()
* may be sufficient. */
component_idx = vsir_write_mask_get_component_idx(e->mask);
component_count = vsir_write_mask_component_count(e->mask);
for (i = e->register_index; i < e->register_index + register_count; ++i)
{
for (j = 0; j < component_idx; ++j)
if (range_map[i].component[j].register_count)
break;
for (j = component_idx + component_count; j < VKD3D_VEC4_SIZE; ++j)
if (range_map[i].component[j].register_count)
break;
}
if (i == register_count)
{
WARN("Expanding mask %#x to %#x for %s, base reg %u, count %u.\n", e->mask, merged_write_mask,
e->semantic_name, e->register_index, register_count);
return merged_write_mask;
}
WARN("Cannot expand mask %#x to %#x for %s, base reg %u, count %u.\n", e->mask, merged_write_mask,
e->semantic_name, e->register_index, register_count);
return e->mask;
}
static enum vkd3d_result shader_signature_merge(struct io_normaliser *normaliser,
struct shader_signature *s, struct io_normaliser_register_data range_map[],
bool is_patch_constant)
@ -2124,39 +2097,45 @@ static enum vkd3d_result shader_signature_merge(struct io_normaliser *normaliser
if (is_patch_constant
&& (ret = shader_signature_map_patch_constant_index_ranges(normaliser, s, range_map)) < 0)
return ret;
goto out;
for (i = 0, new_count = 0; i < element_count; i += register_count, elements[new_count++] = *e)
for (i = 0, new_count = 0; i < element_count; ++i)
{
e = &elements[i];
register_count = 1;
if (e->register_index >= MAX_REG_OUTPUT)
{
elements[new_count++] = *e;
continue;
}
register_count = range_map_get_register_count(range_map, e->register_index, e->mask);
VKD3D_ASSERT(register_count != UINT8_MAX);
register_count += !register_count;
if (register_count > 1)
if (register_count == UINT8_MAX)
{
TRACE("Merging %s, base reg %u, count %u.\n", e->semantic_name, e->register_index, register_count);
e->register_count = register_count;
e->mask = signature_element_range_expand_mask(e, register_count, range_map);
for (j = 1; j < register_count; ++j)
{
f = &elements[i + j];
vkd3d_free((void *)f->semantic_name);
}
TRACE("Register %u mask %#x semantic %s%u has already been merged, dropping it.\n",
e->register_index, e->mask, e->semantic_name, e->semantic_index);
vkd3d_free((void *)e->semantic_name);
continue;
}
if (register_count > 0)
{
TRACE("Register %u mask %#x semantic %s%u is used as merge destination.\n",
e->register_index, e->mask, e->semantic_name, e->semantic_index);
e->register_count = register_count;
e->mask = range_map[e->register_index].component[vsir_write_mask_get_component_idx(e->mask)].mask;
e->used_mask = range_map[e->register_index].component[vsir_write_mask_get_component_idx(e->mask)].used_mask;
}
elements[new_count++] = *e;
}
element_count = new_count;
s->element_count = new_count;
out:
/* Restoring the original order is required for sensible trace output. */
qsort(elements, element_count, sizeof(elements[0]), signature_element_index_compare);
s->element_count = element_count;
qsort(s->elements, s->element_count, sizeof(elements[0]), signature_element_index_compare);
return ret;
}