fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-25 00:00:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

nir: remove alu ddx/ddy

Browse source 
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31014>
This commit is contained in:
Georg Lehmann 2024-09-04 13:12:47 +02:00 • committed by Marge Bot
parent bf0d1a42b4
commit f9d2aad7a3
3 changed files with 3 additions and 56 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

34
src/compiler/nir/nir_lower_wpos_ytransform.c
View file

@ -24,11 +24,11 @@
#include "nir.h"
#include "nir_builder.h"
/* Lower gl_FragCoord (and fddy) to account for driver's requested coordinate-
/* Lower gl_FragCoord (and ddy) to account for driver's requested coordinate-
* origin and pixel-center vs. shader. If transformation is required, a
* gl_FbWposYTransform uniform is inserted (with the specified state-slots)
* and additional instructions are inserted to transform gl_FragCoord (and
* fddy src arg).
* ddy src arg).
*
* This is based on the logic in emit_wpos()/emit_wpos_adjustment() in TGSI
* compiler.
@ -232,29 +232,7 @@ lower_fragcoord(lower_wpos_ytransform_state *state, nir_intrinsic_instr *intr)
emit_wpos_adjustment(state, intr, invert, adjX, adjY);
}
/* turns 'fddy(p)' into 'fddy(fmul(p, transform.x))' */
static void
lower_fddy(lower_wpos_ytransform_state *state, nir_alu_instr *fddy)
{
nir_builder *b = &state->b;
nir_def *p, *pt, *trans;
nir_def *wpostrans = get_transform(state);
b->cursor = nir_before_instr(&fddy->instr);
p = nir_ssa_for_alu_src(b, fddy, 0);
trans = nir_channel(b, wpostrans, 0);
if (p->bit_size == 16)
trans = nir_f2f16(b, trans);
pt = nir_fmul(b, p, trans);
nir_src_rewrite(&fddy->src[0].src, pt);
for (unsigned i = 0; i < 4; i++)
fddy->src[0].swizzle[i] = MIN2(i, pt->num_components - 1);
}
/* turns 'ddy(p)' into 'ddy(fmul(p, transform.x))' */
static void
lower_ddy(lower_wpos_ytransform_state *state, nir_intrinsic_instr *ddy)
{
@ -352,12 +330,6 @@ lower_wpos_ytransform_instr(nir_builder *b, nir_instr *instr,
intr->intrinsic == nir_intrinsic_ddy_coarse) {
lower_ddy(state, intr);
}
} else if (instr->type == nir_instr_type_alu) {
nir_alu_instr *alu = nir_instr_as_alu(instr);
if (alu->op == nir_op_fddy ||
alu->op == nir_op_fddy_fine ||
alu->op == nir_op_fddy_coarse)
lower_fddy(state, alu);
}
return state->transform != NULL;

13
src/compiler/nir/nir_opcodes.py
View file

@ -348,19 +348,6 @@ unop("fcos", tfloat, "bit_size == 64 ? cos(src0) : cosf(src0)")
unop_convert("frexp_exp", tint32, tfloat, "frexp(src0, &dst);")
unop_convert("frexp_sig", tfloat, tfloat, "int n; dst = frexp(src0, &n);")
# Partial derivatives.
deriv_template = """
Calculate the screen-space partial derivative using {} derivatives of the input
with respect to the {}-axis. The constant folding is trivial as the derivative
of a constant is 0 if the constant is not Inf or NaN.
"""
for mode, suffix in [("either fine or coarse", ""), ("fine", "_fine"), ("coarse", "_coarse")]:
for axis in ["x", "y"]:
unop(f"fdd{axis}{suffix}", tfloat, "isfinite(src0) ? 0.0 : NAN",
algebraic_properties = derivative,
description = deriv_template.format(mode, axis.upper()))
# Floating point pack and unpack operations.
def pack_2x16(fmt, in_type):

12
src/compiler/nir/nir_opt_algebraic.py
View file

@ -2892,18 +2892,6 @@ for op in ['fadd', 'fmul', 'fmulz', 'iadd', 'imul']:
((op, ('bcsel(is_used_once)', a, '#b', c), '#d'), ('bcsel', a, (op, b, d), (op, c, d)))
]
# For derivatives in compute shaders, GLSL_NV_compute_shader_derivatives
# states:
#
# If neither layout qualifier is specified, derivatives in compute shaders
# return zero, which is consistent with the handling of built-in texture
# functions like texture() in GLSL 4.50 compute shaders.
for op in ['fddx', 'fddx_fine', 'fddx_coarse',
'fddy', 'fddy_fine', 'fddy_coarse']:
optimizations += [
((op, 'a'), 0.0, 'info->stage == MESA_SHADER_COMPUTE && info->derivative_group == DERIVATIVE_GROUP_NONE')
]
# Some optimizations for ir3-specific instructions.
optimizations += [
# 'al * bl': If either 'al' or 'bl' is zero, return zero.