summaryrefslogtreecommitdiffstats
path: root/src/compiler/spirv/vtn_glsl450.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/compiler/spirv/vtn_glsl450.c')
-rw-r--r--src/compiler/spirv/vtn_glsl450.c155
1 files changed, 2 insertions, 153 deletions
diff --git a/src/compiler/spirv/vtn_glsl450.c b/src/compiler/spirv/vtn_glsl450.c
index dd72a86e21c..2d66512bc42 100644
--- a/src/compiler/spirv/vtn_glsl450.c
+++ b/src/compiler/spirv/vtn_glsl450.c
@@ -234,157 +234,6 @@ build_asin(nir_builder *b, nir_ssa_def *x, float p0, float p1)
expr_tail)));
}
-/**
- * Compute xs[0] + xs[1] + xs[2] + ... using fadd.
- */
-static nir_ssa_def *
-build_fsum(nir_builder *b, nir_ssa_def **xs, int terms)
-{
- nir_ssa_def *accum = xs[0];
-
- for (int i = 1; i < terms; i++)
- accum = nir_fadd(b, accum, xs[i]);
-
- return accum;
-}
-
-static nir_ssa_def *
-build_atan(nir_builder *b, nir_ssa_def *y_over_x)
-{
- const uint32_t bit_size = y_over_x->bit_size;
-
- nir_ssa_def *abs_y_over_x = nir_fabs(b, y_over_x);
- nir_ssa_def *one = nir_imm_floatN_t(b, 1.0f, bit_size);
-
- /*
- * range-reduction, first step:
- *
- * / y_over_x if |y_over_x| <= 1.0;
- * x = <
- * \ 1.0 / y_over_x otherwise
- */
- nir_ssa_def *x = nir_fdiv(b, nir_fmin(b, abs_y_over_x, one),
- nir_fmax(b, abs_y_over_x, one));
-
- /*
- * approximate atan by evaluating polynomial:
- *
- * x * 0.9999793128310355 - x^3 * 0.3326756418091246 +
- * x^5 * 0.1938924977115610 - x^7 * 0.1173503194786851 +
- * x^9 * 0.0536813784310406 - x^11 * 0.0121323213173444
- */
- nir_ssa_def *x_2 = nir_fmul(b, x, x);
- nir_ssa_def *x_3 = nir_fmul(b, x_2, x);
- nir_ssa_def *x_5 = nir_fmul(b, x_3, x_2);
- nir_ssa_def *x_7 = nir_fmul(b, x_5, x_2);
- nir_ssa_def *x_9 = nir_fmul(b, x_7, x_2);
- nir_ssa_def *x_11 = nir_fmul(b, x_9, x_2);
-
- nir_ssa_def *polynomial_terms[] = {
- nir_fmul_imm(b, x, 0.9999793128310355f),
- nir_fmul_imm(b, x_3, -0.3326756418091246f),
- nir_fmul_imm(b, x_5, 0.1938924977115610f),
- nir_fmul_imm(b, x_7, -0.1173503194786851f),
- nir_fmul_imm(b, x_9, 0.0536813784310406f),
- nir_fmul_imm(b, x_11, -0.0121323213173444f),
- };
-
- nir_ssa_def *tmp =
- build_fsum(b, polynomial_terms, ARRAY_SIZE(polynomial_terms));
-
- /* range-reduction fixup */
- tmp = nir_fadd(b, tmp,
- nir_fmul(b, nir_b2f(b, nir_flt(b, one, abs_y_over_x), bit_size),
- nir_fadd_imm(b, nir_fmul_imm(b, tmp, -2.0f), M_PI_2f)));
-
- /* sign fixup */
- return nir_fmul(b, tmp, nir_fsign(b, y_over_x));
-}
-
-static nir_ssa_def *
-build_atan2(nir_builder *b, nir_ssa_def *y, nir_ssa_def *x)
-{
- assert(y->bit_size == x->bit_size);
- const uint32_t bit_size = x->bit_size;
-
- nir_ssa_def *zero = nir_imm_floatN_t(b, 0, bit_size);
- nir_ssa_def *one = nir_imm_floatN_t(b, 1, bit_size);
-
- /* If we're on the left half-plane rotate the coordinates π/2 clock-wise
- * for the y=0 discontinuity to end up aligned with the vertical
- * discontinuity of atan(s/t) along t=0. This also makes sure that we
- * don't attempt to divide by zero along the vertical line, which may give
- * unspecified results on non-GLSL 4.1-capable hardware.
- */
- nir_ssa_def *flip = nir_fge(b, zero, x);
- nir_ssa_def *s = nir_bcsel(b, flip, nir_fabs(b, x), y);
- nir_ssa_def *t = nir_bcsel(b, flip, y, nir_fabs(b, x));
-
- /* If the magnitude of the denominator exceeds some huge value, scale down
- * the arguments in order to prevent the reciprocal operation from flushing
- * its result to zero, which would cause precision problems, and for s
- * infinite would cause us to return a NaN instead of the correct finite
- * value.
- *
- * If fmin and fmax are respectively the smallest and largest positive
- * normalized floating point values representable by the implementation,
- * the constants below should be in agreement with:
- *
- * huge <= 1 / fmin
- * scale <= 1 / fmin / fmax (for |t| >= huge)
- *
- * In addition scale should be a negative power of two in order to avoid
- * loss of precision. The values chosen below should work for most usual
- * floating point representations with at least the dynamic range of ATI's
- * 24-bit representation.
- */
- const double huge_val = bit_size >= 32 ? 1e18 : 16384;
- nir_ssa_def *huge = nir_imm_floatN_t(b, huge_val, bit_size);
- nir_ssa_def *scale = nir_bcsel(b, nir_fge(b, nir_fabs(b, t), huge),
- nir_imm_floatN_t(b, 0.25, bit_size), one);
- nir_ssa_def *rcp_scaled_t = nir_frcp(b, nir_fmul(b, t, scale));
- nir_ssa_def *s_over_t = nir_fmul(b, nir_fmul(b, s, scale), rcp_scaled_t);
-
- /* For |x| = |y| assume tan = 1 even if infinite (i.e. pretend momentarily
- * that ∞/∞ = 1) in order to comply with the rather artificial rules
- * inherited from IEEE 754-2008, namely:
- *
- * "atan2(±∞, −∞) is ±3π/4
- * atan2(±∞, +∞) is ±π/4"
- *
- * Note that this is inconsistent with the rules for the neighborhood of
- * zero that are based on iterated limits:
- *
- * "atan2(±0, −0) is ±π
- * atan2(±0, +0) is ±0"
- *
- * but GLSL specifically allows implementations to deviate from IEEE rules
- * at (0,0), so we take that license (i.e. pretend that 0/0 = 1 here as
- * well).
- */
- nir_ssa_def *tan = nir_bcsel(b, nir_feq(b, nir_fabs(b, x), nir_fabs(b, y)),
- one, nir_fabs(b, s_over_t));
-
- /* Calculate the arctangent and fix up the result if we had flipped the
- * coordinate system.
- */
- nir_ssa_def *arc =
- nir_fadd(b, nir_fmul_imm(b, nir_b2f(b, flip, bit_size), M_PI_2f),
- build_atan(b, tan));
-
- /* Rather convoluted calculation of the sign of the result. When x < 0 we
- * cannot use fsign because we need to be able to distinguish between
- * negative and positive zero. We don't use bitwise arithmetic tricks for
- * consistency with the GLSL front-end. When x >= 0 rcp_scaled_t will
- * always be non-negative so this won't be able to distinguish between
- * negative and positive zero, but we don't care because atan2 is
- * continuous along the whole positive y = 0 half-line, so it won't affect
- * the result significantly.
- */
- return nir_bcsel(b, nir_flt(b, nir_fmin(b, y, rcp_scaled_t), zero),
- nir_fneg(b, arc), arc);
-}
-
static nir_op
vtn_nir_alu_op_for_spirv_glsl_opcode(struct vtn_builder *b,
enum GLSLstd450 opcode,
@@ -662,11 +511,11 @@ handle_glsl450_alu(struct vtn_builder *b, enum GLSLstd450 entrypoint,
return;
case GLSLstd450Atan:
- val->ssa->def = build_atan(nb, src[0]);
+ val->ssa->def = nir_atan(nb, src[0]);
return;
case GLSLstd450Atan2:
- val->ssa->def = build_atan2(nb, src[0], src[1]);
+ val->ssa->def = nir_atan2(nb, src[0], src[1]);
return;
case GLSLstd450Frexp: {