1 /**************************************************************************
3 * Copyright 2009 VMware, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
30 * Helper arithmetic functions.
32 * @author Jose Fonseca <jfonseca@vmware.com>
40 #include "gallivm/lp_bld.h"
44 struct lp_build_context
;
49 * Complement, i.e., 1 - a.
52 lp_build_comp(struct lp_build_context
*bld
,
56 lp_build_add(struct lp_build_context
*bld
,
61 lp_build_horizontal_add(struct lp_build_context
*bld
,
65 lp_build_hadd_partial4(struct lp_build_context
*bld
,
66 LLVMValueRef vectors
[],
70 lp_build_sub(struct lp_build_context
*bld
,
75 lp_build_mul(struct lp_build_context
*bld
,
80 lp_build_mul_32_lohi_cpu(struct lp_build_context
*bld
,
83 LLVMValueRef
*res_hi
);
86 lp_build_mul_32_lohi(struct lp_build_context
*bld
,
89 LLVMValueRef
*res_hi
);
92 lp_build_mul_imm(struct lp_build_context
*bld
,
97 lp_build_div(struct lp_build_context
*bld
,
102 /* llvm.fmuladd.* intrinsic */
104 lp_build_fmuladd(LLVMBuilderRef builder
,
111 lp_build_mad(struct lp_build_context
*bld
,
118 * Set when the weights for normalized are prescaled, that is, in range
119 * 0..2**n, as opposed to range 0..2**(n-1).
121 #define LP_BLD_LERP_PRESCALED_WEIGHTS (1 << 0)
124 * Used internally when using wide intermediates for normalized lerps.
128 #define LP_BLD_LERP_WIDE_NORMALIZED (1 << 1)
131 lp_build_lerp(struct lp_build_context
*bld
,
138 lp_build_lerp_2d(struct lp_build_context
*bld
,
148 lp_build_lerp_3d(struct lp_build_context
*bld
,
163 * Specifies floating point NaN behavior.
165 enum gallivm_nan_behavior
{
166 /* Results are undefined with NaN. Results in fastest code */
167 GALLIVM_NAN_BEHAVIOR_UNDEFINED
,
168 /* If one of the inputs is NaN, NaN is returned */
169 GALLIVM_NAN_RETURN_NAN
,
170 /* If one of the inputs is NaN, the other operand is returned */
171 GALLIVM_NAN_RETURN_OTHER
,
172 /* If one of the inputs is NaN, the other operand is returned,
173 * but we guarantee the second operand is not a NaN.
174 * In min/max it will be as fast as undefined with sse opcodes,
175 * and archs having native return_other can benefit too. */
176 GALLIVM_NAN_RETURN_OTHER_SECOND_NONNAN
,
177 /* If one of the inputs is NaN, NaN is returned,
178 * but we guarantee the first operand is not a NaN.
179 * In min/max it will be as fast as undefined with sse opcodes,
180 * and archs having native return_nan can benefit too. */
181 GALLIVM_NAN_RETURN_NAN_FIRST_NONNAN
,
186 lp_build_min(struct lp_build_context
*bld
,
191 lp_build_min_ext(struct lp_build_context
*bld
,
194 enum gallivm_nan_behavior nan_behavior
);
197 lp_build_max(struct lp_build_context
*bld
,
202 lp_build_max_ext(struct lp_build_context
*bld
,
205 enum gallivm_nan_behavior nan_behavior
);
208 lp_build_clamp(struct lp_build_context
*bld
,
214 lp_build_clamp_zero_one_nanzero(struct lp_build_context
*bld
,
218 lp_build_abs(struct lp_build_context
*bld
,
222 lp_build_negate(struct lp_build_context
*bld
,
226 lp_build_sgn(struct lp_build_context
*bld
,
230 lp_build_set_sign(struct lp_build_context
*bld
,
231 LLVMValueRef a
, LLVMValueRef sign
);
234 lp_build_int_to_float(struct lp_build_context
*bld
,
238 lp_build_round(struct lp_build_context
*bld
,
242 lp_build_floor(struct lp_build_context
*bld
,
246 lp_build_ceil(struct lp_build_context
*bld
,
250 lp_build_trunc(struct lp_build_context
*bld
,
254 lp_build_fract(struct lp_build_context
*bld
,
258 lp_build_fract_safe(struct lp_build_context
*bld
,
262 lp_build_ifloor(struct lp_build_context
*bld
,
265 lp_build_iceil(struct lp_build_context
*bld
,
269 lp_build_iround(struct lp_build_context
*bld
,
273 lp_build_itrunc(struct lp_build_context
*bld
,
277 lp_build_ifloor_fract(struct lp_build_context
*bld
,
279 LLVMValueRef
*out_ipart
,
280 LLVMValueRef
*out_fpart
);
283 lp_build_ifloor_fract_safe(struct lp_build_context
*bld
,
285 LLVMValueRef
*out_ipart
,
286 LLVMValueRef
*out_fpart
);
289 lp_build_sqrt(struct lp_build_context
*bld
,
293 lp_build_rcp(struct lp_build_context
*bld
,
297 lp_build_rsqrt(struct lp_build_context
*bld
,
301 lp_build_fast_rsqrt_available(struct lp_type type
);
304 lp_build_fast_rsqrt(struct lp_build_context
*bld
,
308 lp_build_polynomial(struct lp_build_context
*bld
,
310 const double *coeffs
,
311 unsigned num_coeffs
);
314 lp_build_cos(struct lp_build_context
*bld
,
318 lp_build_sin(struct lp_build_context
*bld
,
322 lp_build_pow(struct lp_build_context
*bld
,
327 lp_build_exp(struct lp_build_context
*bld
,
331 lp_build_log(struct lp_build_context
*bld
,
335 lp_build_log_safe(struct lp_build_context
*bld
,
339 lp_build_exp2(struct lp_build_context
*bld
,
343 lp_build_extract_exponent(struct lp_build_context
*bld
,
348 lp_build_extract_mantissa(struct lp_build_context
*bld
,
352 lp_build_log2(struct lp_build_context
*bld
,
356 lp_build_log2_safe(struct lp_build_context
*bld
,
360 lp_build_fast_log2(struct lp_build_context
*bld
,
364 lp_build_ilog2(struct lp_build_context
*bld
,
368 lp_build_log2_approx(struct lp_build_context
*bld
,
371 LLVMValueRef
*p_floor_log2
,
372 LLVMValueRef
*p_log2
,
373 boolean handle_nans
);
376 lp_build_mod(struct lp_build_context
*bld
,
381 lp_build_isnan(struct lp_build_context
*bld
,
385 lp_build_isfinite(struct lp_build_context
*bld
,
390 lp_build_is_inf_or_nan(struct gallivm_state
*gallivm
,
391 const struct lp_type type
,
396 lp_build_fpstate_get(struct gallivm_state
*gallivm
);
399 lp_build_fpstate_set_denorms_zero(struct gallivm_state
*gallivm
,
402 lp_build_fpstate_set(struct gallivm_state
*gallivm
,
405 #endif /* !LP_BLD_ARIT_H */