@@ -31,427 +31,426 @@
/* Offsets for data table __svml_sasinh_data_internal
*/
-#define SgnMask 0
-#define sOne 32
-#define sPoly 64
-#define iBrkValue 320
-#define iOffExpoMask 352
-#define sBigThreshold 384
-#define sC2 416
-#define sC3 448
-#define sHalf 480
-#define sLargestFinite 512
-#define sLittleThreshold 544
-#define sSign 576
-#define sThirtyOne 608
-#define sTopMask8 640
-#define XScale 672
-#define sLn2 704
+#define SgnMask 0
+#define sOne 32
+#define sPoly 64
+#define iBrkValue 320
+#define iOffExpoMask 352
+#define sBigThreshold 384
+#define sC2 416
+#define sC3 448
+#define sHalf 480
+#define sLargestFinite 512
+#define sLittleThreshold 544
+#define sSign 576
+#define sThirtyOne 608
+#define sTopMask8 640
+#define XScale 672
+#define sLn2 704
#include <sysdep.h>
- .text
- .section .text.avx2,"ax",@progbits
+ .section .text.avx2, "ax", @progbits
ENTRY(_ZGVdN8v_asinhf_avx2)
- pushq %rbp
- cfi_def_cfa_offset(16)
- movq %rsp, %rbp
- cfi_def_cfa(6, 16)
- cfi_offset(6, -16)
- andq $-32, %rsp
- subq $96, %rsp
- vmovaps %ymm0, %ymm9
-
-/* Load the constant 1 and a sign mask */
- vmovups sOne+__svml_sasinh_data_internal(%rip), %ymm8
-
-/* No need to split X when FMA is available in hardware. */
- vmulps %ymm9, %ymm9, %ymm5
- vmovups sTopMask8+__svml_sasinh_data_internal(%rip), %ymm1
-
-/*
- * Finally, express Y + W = X^2 + 1 accurately where Y has <= 8 bits.
- * If |X| <= 1 then |XHi| <= 1 and so |X2Hi| <= 1, so we can treat 1
- * as the dominant component in the compensated summation. Otherwise,
- * if |X| >= 1, then since X2Hi only has 22 significant bits, the basic
- * addition will be exact anyway until we get to |X| >= 2^24. But by
- * that time the log function is well-conditioned enough that the
- * rounding error doesn't matter. Hence we can treat 1 as dominant even
- * if it literally isn't.
- */
- vaddps %ymm5, %ymm8, %ymm13
- vandps %ymm1, %ymm13, %ymm2
- vmovaps %ymm9, %ymm4
- vsubps %ymm13, %ymm8, %ymm11
- vsubps %ymm2, %ymm13, %ymm15
-
-/*
- * Compute R = 1/sqrt(Y + W) * (1 + d)
- * Force R to <= 8 significant bits.
- * This means that R * Y and R^2 * Y are exactly representable.
- */
- vrsqrtps %ymm2, %ymm0
- vfmsub213ps %ymm5, %ymm9, %ymm4
- vaddps %ymm11, %ymm5, %ymm12
-
-/*
- * Get the absolute value of the input, since we will exploit antisymmetry
- * and mostly assume X >= 0 in the core computation
- */
- vandps SgnMask+__svml_sasinh_data_internal(%rip), %ymm9, %ymm6
-
-/*
- * Check whether the input is finite, by checking |X| <= MaxFloat
- * Otherwise set the rangemask so that the callout will get used.
- * Note that this will also use the callout for NaNs since not(NaN <= MaxFloat)
- */
- vcmpnle_uqps sLargestFinite+__svml_sasinh_data_internal(%rip), %ymm6, %ymm10
- vaddps %ymm12, %ymm4, %ymm14
-
-/*
- * Unfortunately, we can still be in trouble if |X| <= 2^-5, since
- * the absolute error 2^-(7+24)-ish in sqrt(1 + X^2) gets scaled up
- * by 1/X and comes close to our threshold. Hence if |X| <= 2^-4,
- * perform an alternative computation
- * sqrt(1 + X^2) - 1 = X^2/2 - X^4/8 + X^6/16
- * X2 = X^2
- */
- vaddps %ymm4, %ymm5, %ymm4
-
-/*
- * The following computation can go wrong for very large X, basically
- * because X^2 overflows. But for large X we have
- * asinh(X) / log(2 X) - 1 =~= 1/(4 * X^2), so for X >= 2^30
- * we can just later stick X back into the log and tweak up the exponent.
- * Actually we scale X by 2^-30 and tweak the exponent up by 31,
- * to stay in the safe range for the later log computation.
- * Compute a flag now telling us when do do this.
- */
- vcmplt_oqps sBigThreshold+__svml_sasinh_data_internal(%rip), %ymm6, %ymm7
- vaddps %ymm15, %ymm14, %ymm3
-
-/*
- * Now 1 / (1 + d)
- * = 1 / (1 + (sqrt(1 - e) - 1))
- * = 1 / sqrt(1 - e)
- * = 1 + 1/2 * e + 3/8 * e^2 + 5/16 * e^3 + 35/128 * e^4 + ...
- * So compute the first three nonconstant terms of that, so that
- * we have a relative correction (1 + Corr) to apply to S etc.
- * C1 = 1/2
- * C2 = 3/8
- * C3 = 5/16
- */
- vmovups sC3+__svml_sasinh_data_internal(%rip), %ymm12
- vmovmskps %ymm10, %edx
- vandps %ymm1, %ymm0, %ymm10
-
-/*
- * Compute S = (Y/sqrt(Y + W)) * (1 + d)
- * and T = (W/sqrt(Y + W)) * (1 + d)
- * so that S + T = sqrt(Y + W) * (1 + d)
- * S is exact, and the rounding error in T is OK.
- */
- vmulps %ymm10, %ymm2, %ymm15
- vmulps %ymm3, %ymm10, %ymm14
- vmovups sHalf+__svml_sasinh_data_internal(%rip), %ymm3
- vsubps %ymm8, %ymm15, %ymm0
-
-/*
- * Obtain sqrt(1 + X^2) - 1 in two pieces
- * sqrt(1 + X^2) - 1
- * = sqrt(Y + W) - 1
- * = (S + T) * (1 + Corr) - 1
- * = [S - 1] + [T + (S + T) * Corr]
- * We need a compensated summation for the last part. We treat S - 1
- * as the larger part; it certainly is until about X < 2^-4, and in that
- * case, the error is affordable since X dominates over sqrt(1 + X^2) - 1
- * Final sum is dTmp5 (hi) + dTmp7 (lo)
- */
- vaddps %ymm14, %ymm15, %ymm13
-
-/*
- * Compute e = -(2 * d + d^2)
- * The first FMR is exact, and the rounding error in the other is acceptable
- * since d and e are ~ 2^-8
- */
- vmovaps %ymm8, %ymm11
- vfnmadd231ps %ymm15, %ymm10, %ymm11
- vfnmadd231ps %ymm14, %ymm10, %ymm11
- vfmadd213ps sC2+__svml_sasinh_data_internal(%rip), %ymm11, %ymm12
- vfmadd213ps %ymm3, %ymm11, %ymm12
- vmulps %ymm12, %ymm11, %ymm1
-
-/* Now multiplex the two possible computations */
- vcmple_oqps sLittleThreshold+__svml_sasinh_data_internal(%rip), %ymm6, %ymm11
- vfmadd213ps %ymm14, %ymm13, %ymm1
- vaddps %ymm0, %ymm1, %ymm2
- vsubps %ymm2, %ymm0, %ymm10
-
-/* sX2over2 = X^2/2 */
- vmulps %ymm4, %ymm3, %ymm0
- vaddps %ymm10, %ymm1, %ymm1
-
-/* sX4over4 = X^4/4 */
- vmulps %ymm0, %ymm0, %ymm5
-
-/* sX46 = -X^4/4 + X^6/8 */
- vfmsub231ps %ymm0, %ymm5, %ymm5
-
-/* sX46over2 = -X^4/8 + x^6/16 */
- vmulps %ymm5, %ymm3, %ymm3
- vaddps %ymm3, %ymm0, %ymm5
- vblendvps %ymm11, %ymm5, %ymm2, %ymm2
- vsubps %ymm5, %ymm0, %ymm4
-
-/*
- * Now do another compensated sum to add |X| + [sqrt(1 + X^2) - 1].
- * It's always safe to assume |X| is larger.
- * This is the final 2-part argument to the log1p function
- */
- vaddps %ymm2, %ymm6, %ymm14
-
-/*
- * Now resume the main code.
- * reduction: compute r,n
- */
- vmovups iBrkValue+__svml_sasinh_data_internal(%rip), %ymm5
- vaddps %ymm4, %ymm3, %ymm10
-
-/*
- * Now we feed into the log1p code, using H in place of _VARG1 and
- * also adding L into Xl.
- * compute 1+x as high, low parts
- */
- vmaxps %ymm14, %ymm8, %ymm15
- vminps %ymm14, %ymm8, %ymm0
- vblendvps %ymm11, %ymm10, %ymm1, %ymm12
- vsubps %ymm14, %ymm6, %ymm1
- vaddps %ymm0, %ymm15, %ymm3
-
-/* Now multiplex to the case X = 2^-30 * input, Xl = sL = 0 in the "big" case. */
- vmulps XScale+__svml_sasinh_data_internal(%rip), %ymm6, %ymm6
- vaddps %ymm1, %ymm2, %ymm13
- vsubps %ymm3, %ymm15, %ymm15
- vaddps %ymm13, %ymm12, %ymm1
- vaddps %ymm15, %ymm0, %ymm2
- vblendvps %ymm7, %ymm3, %ymm6, %ymm0
- vaddps %ymm2, %ymm1, %ymm4
- vpsubd %ymm5, %ymm0, %ymm1
- vpsrad $23, %ymm1, %ymm6
- vpand iOffExpoMask+__svml_sasinh_data_internal(%rip), %ymm1, %ymm2
- vmovups sPoly+224+__svml_sasinh_data_internal(%rip), %ymm1
- vpslld $23, %ymm6, %ymm10
- vpaddd %ymm5, %ymm2, %ymm13
- vcvtdq2ps %ymm6, %ymm0
- vpsubd %ymm10, %ymm8, %ymm12
-
-/* polynomial evaluation */
- vsubps %ymm8, %ymm13, %ymm8
-
-/* Add 31 to the exponent in the "large" case to get log(2 * input) */
- vaddps sThirtyOne+__svml_sasinh_data_internal(%rip), %ymm0, %ymm3
- vandps %ymm7, %ymm4, %ymm11
- vmulps %ymm12, %ymm11, %ymm14
- vblendvps %ymm7, %ymm0, %ymm3, %ymm0
- vaddps %ymm8, %ymm14, %ymm2
- vfmadd213ps sPoly+192+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
- vfmadd213ps sPoly+160+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
- vfmadd213ps sPoly+128+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
- vfmadd213ps sPoly+96+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
- vfmadd213ps sPoly+64+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
- vfmadd213ps sPoly+32+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
- vfmadd213ps sPoly+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
- vmulps %ymm1, %ymm2, %ymm4
- vfmadd213ps %ymm2, %ymm2, %ymm4
-
-/* final reconstruction */
- vfmadd132ps sLn2+__svml_sasinh_data_internal(%rip), %ymm4, %ymm0
-
-/* Finally, reincorporate the original sign. */
- vandps sSign+__svml_sasinh_data_internal(%rip), %ymm9, %ymm7
- vxorps %ymm0, %ymm7, %ymm0
- testl %edx, %edx
-
-/* Go to special inputs processing branch */
- jne L(SPECIAL_VALUES_BRANCH)
- # LOE rbx r12 r13 r14 r15 edx ymm0 ymm9
-
-/* Restore registers
- * and exit the function
- */
+ pushq %rbp
+ cfi_def_cfa_offset(16)
+ movq %rsp, %rbp
+ cfi_def_cfa(6, 16)
+ cfi_offset(6, -16)
+ andq $-32, %rsp
+ subq $96, %rsp
+ vmovaps %ymm0, %ymm9
+
+ /* Load the constant 1 and a sign mask */
+ vmovups sOne+__svml_sasinh_data_internal(%rip), %ymm8
+
+ /* No need to split X when FMA is available in hardware. */
+ vmulps %ymm9, %ymm9, %ymm5
+ vmovups sTopMask8+__svml_sasinh_data_internal(%rip), %ymm1
+
+ /*
+ * Finally, express Y + W = X^2 + 1 accurately where Y has <= 8 bits.
+ * If |X| <= 1 then |XHi| <= 1 and so |X2Hi| <= 1, so we can treat 1
+ * as the dominant component in the compensated summation. Otherwise,
+ * if |X| >= 1, then since X2Hi only has 22 significant bits, the basic
+ * addition will be exact anyway until we get to |X| >= 2^24. But by
+ * that time the log function is well-conditioned enough that the
+ * rounding error doesn't matter. Hence we can treat 1 as dominant even
+ * if it literally isn't.
+ */
+ vaddps %ymm5, %ymm8, %ymm13
+ vandps %ymm1, %ymm13, %ymm2
+ vmovaps %ymm9, %ymm4
+ vsubps %ymm13, %ymm8, %ymm11
+ vsubps %ymm2, %ymm13, %ymm15
+
+ /*
+ * Compute R = 1/sqrt(Y + W) * (1 + d)
+ * Force R to <= 8 significant bits.
+ * This means that R * Y and R^2 * Y are exactly representable.
+ */
+ vrsqrtps %ymm2, %ymm0
+ vfmsub213ps %ymm5, %ymm9, %ymm4
+ vaddps %ymm11, %ymm5, %ymm12
+
+ /*
+ * Get the absolute value of the input, since we will exploit antisymmetry
+ * and mostly assume X >= 0 in the core computation
+ */
+ vandps SgnMask+__svml_sasinh_data_internal(%rip), %ymm9, %ymm6
+
+ /*
+ * Check whether the input is finite, by checking |X| <= MaxFloat
+ * Otherwise set the rangemask so that the callout will get used.
+ * Note that this will also use the callout for NaNs since not(NaN <= MaxFloat)
+ */
+ vcmpnle_uqps sLargestFinite+__svml_sasinh_data_internal(%rip), %ymm6, %ymm10
+ vaddps %ymm12, %ymm4, %ymm14
+
+ /*
+ * Unfortunately, we can still be in trouble if |X| <= 2^-5, since
+ * the absolute error 2^-(7+24)-ish in sqrt(1 + X^2) gets scaled up
+ * by 1/X and comes close to our threshold. Hence if |X| <= 2^-4,
+ * perform an alternative computation
+ * sqrt(1 + X^2) - 1 = X^2/2 - X^4/8 + X^6/16
+ * X2 = X^2
+ */
+ vaddps %ymm4, %ymm5, %ymm4
+
+ /*
+ * The following computation can go wrong for very large X, basically
+ * because X^2 overflows. But for large X we have
+ * asinh(X) / log(2 X) - 1 =~= 1/(4 * X^2), so for X >= 2^30
+ * we can just later stick X back into the log and tweak up the exponent.
+ * Actually we scale X by 2^-30 and tweak the exponent up by 31,
+ * to stay in the safe range for the later log computation.
+ * Compute a flag now telling us when do do this.
+ */
+ vcmplt_oqps sBigThreshold+__svml_sasinh_data_internal(%rip), %ymm6, %ymm7
+ vaddps %ymm15, %ymm14, %ymm3
+
+ /*
+ * Now 1 / (1 + d)
+ * = 1 / (1 + (sqrt(1 - e) - 1))
+ * = 1 / sqrt(1 - e)
+ * = 1 + 1/2 * e + 3/8 * e^2 + 5/16 * e^3 + 35/128 * e^4 + ...
+ * So compute the first three nonconstant terms of that, so that
+ * we have a relative correction (1 + Corr) to apply to S etc.
+ * C1 = 1/2
+ * C2 = 3/8
+ * C3 = 5/16
+ */
+ vmovups sC3+__svml_sasinh_data_internal(%rip), %ymm12
+ vmovmskps %ymm10, %edx
+ vandps %ymm1, %ymm0, %ymm10
+
+ /*
+ * Compute S = (Y/sqrt(Y + W)) * (1 + d)
+ * and T = (W/sqrt(Y + W)) * (1 + d)
+ * so that S + T = sqrt(Y + W) * (1 + d)
+ * S is exact, and the rounding error in T is OK.
+ */
+ vmulps %ymm10, %ymm2, %ymm15
+ vmulps %ymm3, %ymm10, %ymm14
+ vmovups sHalf+__svml_sasinh_data_internal(%rip), %ymm3
+ vsubps %ymm8, %ymm15, %ymm0
+
+ /*
+ * Obtain sqrt(1 + X^2) - 1 in two pieces
+ * sqrt(1 + X^2) - 1
+ * = sqrt(Y + W) - 1
+ * = (S + T) * (1 + Corr) - 1
+ * = [S - 1] + [T + (S + T) * Corr]
+ * We need a compensated summation for the last part. We treat S - 1
+ * as the larger part; it certainly is until about X < 2^-4, and in that
+ * case, the error is affordable since X dominates over sqrt(1 + X^2) - 1
+ * Final sum is dTmp5 (hi) + dTmp7 (lo)
+ */
+ vaddps %ymm14, %ymm15, %ymm13
+
+ /*
+ * Compute e = -(2 * d + d^2)
+ * The first FMR is exact, and the rounding error in the other is acceptable
+ * since d and e are ~ 2^-8
+ */
+ vmovaps %ymm8, %ymm11
+ vfnmadd231ps %ymm15, %ymm10, %ymm11
+ vfnmadd231ps %ymm14, %ymm10, %ymm11
+ vfmadd213ps sC2+__svml_sasinh_data_internal(%rip), %ymm11, %ymm12
+ vfmadd213ps %ymm3, %ymm11, %ymm12
+ vmulps %ymm12, %ymm11, %ymm1
+
+ /* Now multiplex the two possible computations */
+ vcmple_oqps sLittleThreshold+__svml_sasinh_data_internal(%rip), %ymm6, %ymm11
+ vfmadd213ps %ymm14, %ymm13, %ymm1
+ vaddps %ymm0, %ymm1, %ymm2
+ vsubps %ymm2, %ymm0, %ymm10
+
+ /* sX2over2 = X^2/2 */
+ vmulps %ymm4, %ymm3, %ymm0
+ vaddps %ymm10, %ymm1, %ymm1
+
+ /* sX4over4 = X^4/4 */
+ vmulps %ymm0, %ymm0, %ymm5
+
+ /* sX46 = -X^4/4 + X^6/8 */
+ vfmsub231ps %ymm0, %ymm5, %ymm5
+
+ /* sX46over2 = -X^4/8 + x^6/16 */
+ vmulps %ymm5, %ymm3, %ymm3
+ vaddps %ymm3, %ymm0, %ymm5
+ vblendvps %ymm11, %ymm5, %ymm2, %ymm2
+ vsubps %ymm5, %ymm0, %ymm4
+
+ /*
+ * Now do another compensated sum to add |X| + [sqrt(1 + X^2) - 1].
+ * It's always safe to assume |X| is larger.
+ * This is the final 2-part argument to the log1p function
+ */
+ vaddps %ymm2, %ymm6, %ymm14
+
+ /*
+ * Now resume the main code.
+ * reduction: compute r, n
+ */
+ vmovups iBrkValue+__svml_sasinh_data_internal(%rip), %ymm5
+ vaddps %ymm4, %ymm3, %ymm10
+
+ /*
+ * Now we feed into the log1p code, using H in place of _VARG1 and
+ * also adding L into Xl.
+ * compute 1+x as high, low parts
+ */
+ vmaxps %ymm14, %ymm8, %ymm15
+ vminps %ymm14, %ymm8, %ymm0
+ vblendvps %ymm11, %ymm10, %ymm1, %ymm12
+ vsubps %ymm14, %ymm6, %ymm1
+ vaddps %ymm0, %ymm15, %ymm3
+
+ /* Now multiplex to the case X = 2^-30 * input, Xl = sL = 0 in the "big" case. */
+ vmulps XScale+__svml_sasinh_data_internal(%rip), %ymm6, %ymm6
+ vaddps %ymm1, %ymm2, %ymm13
+ vsubps %ymm3, %ymm15, %ymm15
+ vaddps %ymm13, %ymm12, %ymm1
+ vaddps %ymm15, %ymm0, %ymm2
+ vblendvps %ymm7, %ymm3, %ymm6, %ymm0
+ vaddps %ymm2, %ymm1, %ymm4
+ vpsubd %ymm5, %ymm0, %ymm1
+ vpsrad $23, %ymm1, %ymm6
+ vpand iOffExpoMask+__svml_sasinh_data_internal(%rip), %ymm1, %ymm2
+ vmovups sPoly+224+__svml_sasinh_data_internal(%rip), %ymm1
+ vpslld $23, %ymm6, %ymm10
+ vpaddd %ymm5, %ymm2, %ymm13
+ vcvtdq2ps %ymm6, %ymm0
+ vpsubd %ymm10, %ymm8, %ymm12
+
+ /* polynomial evaluation */
+ vsubps %ymm8, %ymm13, %ymm8
+
+ /* Add 31 to the exponent in the "large" case to get log(2 * input) */
+ vaddps sThirtyOne+__svml_sasinh_data_internal(%rip), %ymm0, %ymm3
+ vandps %ymm7, %ymm4, %ymm11
+ vmulps %ymm12, %ymm11, %ymm14
+ vblendvps %ymm7, %ymm0, %ymm3, %ymm0
+ vaddps %ymm8, %ymm14, %ymm2
+ vfmadd213ps sPoly+192+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213ps sPoly+160+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213ps sPoly+128+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213ps sPoly+96+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213ps sPoly+64+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213ps sPoly+32+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213ps sPoly+__svml_sasinh_data_internal(%rip), %ymm2, %ymm1
+ vmulps %ymm1, %ymm2, %ymm4
+ vfmadd213ps %ymm2, %ymm2, %ymm4
+
+ /* final reconstruction */
+ vfmadd132ps sLn2+__svml_sasinh_data_internal(%rip), %ymm4, %ymm0
+
+ /* Finally, reincorporate the original sign. */
+ vandps sSign+__svml_sasinh_data_internal(%rip), %ymm9, %ymm7
+ vxorps %ymm0, %ymm7, %ymm0
+ testl %edx, %edx
+
+ /* Go to special inputs processing branch */
+ jne L(SPECIAL_VALUES_BRANCH)
+ # LOE rbx r12 r13 r14 r15 edx ymm0 ymm9
+
+ /* Restore registers
+ * and exit the function
+ */
L(EXIT):
- movq %rbp, %rsp
- popq %rbp
- cfi_def_cfa(7, 8)
- cfi_restore(6)
- ret
- cfi_def_cfa(6, 16)
- cfi_offset(6, -16)
-
-/* Branch to process
- * special inputs
- */
+ movq %rbp, %rsp
+ popq %rbp
+ cfi_def_cfa(7, 8)
+ cfi_restore(6)
+ ret
+ cfi_def_cfa(6, 16)
+ cfi_offset(6, -16)
+
+ /* Branch to process
+ * special inputs
+ */
L(SPECIAL_VALUES_BRANCH):
- vmovups %ymm9, 32(%rsp)
- vmovups %ymm0, 64(%rsp)
- # LOE rbx r12 r13 r14 r15 edx ymm0
-
- xorl %eax, %eax
- # LOE rbx r12 r13 r14 r15 eax edx
-
- vzeroupper
- movq %r12, 16(%rsp)
- /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus) */
- .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
- movl %eax, %r12d
- movq %r13, 8(%rsp)
- /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus) */
- .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
- movl %edx, %r13d
- movq %r14, (%rsp)
- /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus) */
- .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
- # LOE rbx r15 r12d r13d
-
-/* Range mask
- * bits check
- */
+ vmovups %ymm9, 32(%rsp)
+ vmovups %ymm0, 64(%rsp)
+ # LOE rbx r12 r13 r14 r15 edx ymm0
+
+ xorl %eax, %eax
+ # LOE rbx r12 r13 r14 r15 eax edx
+
+ vzeroupper
+ movq %r12, 16(%rsp)
+ /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus) */
+ .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
+ movl %eax, %r12d
+ movq %r13, 8(%rsp)
+ /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus) */
+ .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
+ movl %edx, %r13d
+ movq %r14, (%rsp)
+ /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus) */
+ .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
+ # LOE rbx r15 r12d r13d
+
+ /* Range mask
+ * bits check
+ */
L(RANGEMASK_CHECK):
- btl %r12d, %r13d
+ btl %r12d, %r13d
-/* Call scalar math function */
- jc L(SCALAR_MATH_CALL)
- # LOE rbx r15 r12d r13d
+ /* Call scalar math function */
+ jc L(SCALAR_MATH_CALL)
+ # LOE rbx r15 r12d r13d
-/* Special inputs
- * processing loop
- */
+ /* Special inputs
+ * processing loop
+ */
L(SPECIAL_VALUES_LOOP):
- incl %r12d
- cmpl $8, %r12d
-
-/* Check bits in range mask */
- jl L(RANGEMASK_CHECK)
- # LOE rbx r15 r12d r13d
-
- movq 16(%rsp), %r12
- cfi_restore(12)
- movq 8(%rsp), %r13
- cfi_restore(13)
- movq (%rsp), %r14
- cfi_restore(14)
- vmovups 64(%rsp), %ymm0
-
-/* Go to exit */
- jmp L(EXIT)
- /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus) */
- .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
- /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus) */
- .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
- /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus) */
- .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
- # LOE rbx r12 r13 r14 r15 ymm0
-
-/* Scalar math fucntion call
- * to process special input
- */
+ incl %r12d
+ cmpl $8, %r12d
+
+ /* Check bits in range mask */
+ jl L(RANGEMASK_CHECK)
+ # LOE rbx r15 r12d r13d
+
+ movq 16(%rsp), %r12
+ cfi_restore(12)
+ movq 8(%rsp), %r13
+ cfi_restore(13)
+ movq (%rsp), %r14
+ cfi_restore(14)
+ vmovups 64(%rsp), %ymm0
+
+ /* Go to exit */
+ jmp L(EXIT)
+ /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus) */
+ .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
+ /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus) */
+ .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
+ /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus) */
+ .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
+ # LOE rbx r12 r13 r14 r15 ymm0
+
+ /* Scalar math fucntion call
+ * to process special input
+ */
L(SCALAR_MATH_CALL):
- movl %r12d, %r14d
- movss 32(%rsp,%r14,4), %xmm0
- call asinhf@PLT
- # LOE rbx r14 r15 r12d r13d xmm0
+ movl %r12d, %r14d
+ movss 32(%rsp, %r14, 4), %xmm0
+ call asinhf@PLT
+ # LOE rbx r14 r15 r12d r13d xmm0
- movss %xmm0, 64(%rsp,%r14,4)
+ movss %xmm0, 64(%rsp, %r14, 4)
-/* Process special inputs in loop */
- jmp L(SPECIAL_VALUES_LOOP)
- # LOE rbx r15 r12d r13d
+ /* Process special inputs in loop */
+ jmp L(SPECIAL_VALUES_LOOP)
+ # LOE rbx r15 r12d r13d
END(_ZGVdN8v_asinhf_avx2)
- .section .rodata, "a"
- .align 32
+ .section .rodata, "a"
+ .align 32
#ifdef __svml_sasinh_data_internal_typedef
typedef unsigned int VUINT32;
typedef struct {
- __declspec(align(32)) VUINT32 SgnMask[8][1];
- __declspec(align(32)) VUINT32 sOne[8][1];
- __declspec(align(32)) VUINT32 sPoly[8][8][1];
- __declspec(align(32)) VUINT32 iBrkValue[8][1];
- __declspec(align(32)) VUINT32 iOffExpoMask[8][1];
- __declspec(align(32)) VUINT32 sBigThreshold[8][1];
- __declspec(align(32)) VUINT32 sC2[8][1];
- __declspec(align(32)) VUINT32 sC3[8][1];
- __declspec(align(32)) VUINT32 sHalf[8][1];
- __declspec(align(32)) VUINT32 sLargestFinite[8][1];
- __declspec(align(32)) VUINT32 sLittleThreshold[8][1];
- __declspec(align(32)) VUINT32 sSign[8][1];
- __declspec(align(32)) VUINT32 sThirtyOne[8][1];
- __declspec(align(32)) VUINT32 sTopMask8[8][1];
- __declspec(align(32)) VUINT32 XScale[8][1];
- __declspec(align(32)) VUINT32 sLn2[8][1];
+ __declspec(align(32)) VUINT32 SgnMask[8][1];
+ __declspec(align(32)) VUINT32 sOne[8][1];
+ __declspec(align(32)) VUINT32 sPoly[8][8][1];
+ __declspec(align(32)) VUINT32 iBrkValue[8][1];
+ __declspec(align(32)) VUINT32 iOffExpoMask[8][1];
+ __declspec(align(32)) VUINT32 sBigThreshold[8][1];
+ __declspec(align(32)) VUINT32 sC2[8][1];
+ __declspec(align(32)) VUINT32 sC3[8][1];
+ __declspec(align(32)) VUINT32 sHalf[8][1];
+ __declspec(align(32)) VUINT32 sLargestFinite[8][1];
+ __declspec(align(32)) VUINT32 sLittleThreshold[8][1];
+ __declspec(align(32)) VUINT32 sSign[8][1];
+ __declspec(align(32)) VUINT32 sThirtyOne[8][1];
+ __declspec(align(32)) VUINT32 sTopMask8[8][1];
+ __declspec(align(32)) VUINT32 XScale[8][1];
+ __declspec(align(32)) VUINT32 sLn2[8][1];
} __svml_sasinh_data_internal;
#endif
__svml_sasinh_data_internal:
- /*== SgnMask ==*/
- .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff
- /*== sOne = SP 1.0 ==*/
- .align 32
- .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000
- /*== sPoly[] = SP polynomial ==*/
- .align 32
- .long 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000 /* -5.0000000000000000000000000e-01 P0 */
- .long 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94 /* 3.3333265781402587890625000e-01 P1 */
- .long 0xbe80058e, 0xbe80058e, 0xbe80058e, 0xbe80058e, 0xbe80058e, 0xbe80058e, 0xbe80058e, 0xbe80058e /* -2.5004237890243530273437500e-01 P2 */
- .long 0x3e4ce190, 0x3e4ce190, 0x3e4ce190, 0x3e4ce190, 0x3e4ce190, 0x3e4ce190, 0x3e4ce190, 0x3e4ce190 /* 2.0007920265197753906250000e-01 P3 */
- .long 0xbe28ad37, 0xbe28ad37, 0xbe28ad37, 0xbe28ad37, 0xbe28ad37, 0xbe28ad37, 0xbe28ad37, 0xbe28ad37 /* -1.6472326219081878662109375e-01 P4 */
- .long 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12 /* 1.4042308926582336425781250e-01 P5 */
- .long 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3 /* -1.5122179687023162841796875e-01 P6 */
- .long 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed /* 1.3820238411426544189453125e-01 P7 */
- /*== iBrkValue = SP 2/3 ==*/
- .align 32
- .long 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab
- /*== iOffExpoMask = SP significand mask ==*/
- .align 32
- .long 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff
- /*== sBigThreshold ==*/
- .align 32
- .long 0x4E800000, 0x4E800000, 0x4E800000, 0x4E800000, 0x4E800000, 0x4E800000, 0x4E800000, 0x4E800000
- /*== sC2 ==*/
- .align 32
- .long 0x3EC00000, 0x3EC00000, 0x3EC00000, 0x3EC00000, 0x3EC00000, 0x3EC00000, 0x3EC00000, 0x3EC00000
- /*== sC3 ==*/
- .align 32
- .long 0x3EA00000, 0x3EA00000, 0x3EA00000, 0x3EA00000, 0x3EA00000, 0x3EA00000, 0x3EA00000, 0x3EA00000
- /*== sHalf ==*/
- .align 32
- .long 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000
- /*== sLargestFinite ==*/
- .align 32
- .long 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF
- /*== sLittleThreshold ==*/
- .align 32
- .long 0x3D800000, 0x3D800000, 0x3D800000, 0x3D800000, 0x3D800000, 0x3D800000, 0x3D800000, 0x3D800000
- /*== sSign ==*/
- .align 32
- .long 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000
- /*== sThirtyOne ==*/
- .align 32
- .long 0x41F80000, 0x41F80000, 0x41F80000, 0x41F80000, 0x41F80000, 0x41F80000, 0x41F80000, 0x41F80000
- /*== sTopMask8 ==*/
- .align 32
- .long 0xFFFF0000, 0xFFFF0000, 0xFFFF0000, 0xFFFF0000, 0xFFFF0000, 0xFFFF0000, 0xFFFF0000, 0xFFFF0000
- /*== XScale ==*/
- .align 32
- .long 0x30800000, 0x30800000, 0x30800000, 0x30800000, 0x30800000, 0x30800000, 0x30800000, 0x30800000
- /*== sLn2 = SP ln(2) ==*/
- .align 32
- .long 0x3f317218, 0x3f317218, 0x3f317218, 0x3f317218, 0x3f317218, 0x3f317218, 0x3f317218, 0x3f317218
- .align 32
- .type __svml_sasinh_data_internal,@object
- .size __svml_sasinh_data_internal,.-__svml_sasinh_data_internal
+ /* SgnMask */
+ .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff
+ /* sOne = SP 1.0 */
+ .align 32
+ .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000
+ /* sPoly[] = SP polynomial */
+ .align 32
+ .long 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000 /* -5.0000000000000000000000000e-01 P0 */
+ .long 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94, 0x3eaaaa94 /* 3.3333265781402587890625000e-01 P1 */
+ .long 0xbe80058e, 0xbe80058e, 0xbe80058e, 0xbe80058e, 0xbe80058e, 0xbe80058e, 0xbe80058e, 0xbe80058e /* -2.5004237890243530273437500e-01 P2 */
+ .long 0x3e4ce190, 0x3e4ce190, 0x3e4ce190, 0x3e4ce190, 0x3e4ce190, 0x3e4ce190, 0x3e4ce190, 0x3e4ce190 /* 2.0007920265197753906250000e-01 P3 */
+ .long 0xbe28ad37, 0xbe28ad37, 0xbe28ad37, 0xbe28ad37, 0xbe28ad37, 0xbe28ad37, 0xbe28ad37, 0xbe28ad37 /* -1.6472326219081878662109375e-01 P4 */
+ .long 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12, 0x3e0fcb12 /* 1.4042308926582336425781250e-01 P5 */
+ .long 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3, 0xbe1ad9e3 /* -1.5122179687023162841796875e-01 P6 */
+ .long 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed, 0x3e0d84ed /* 1.3820238411426544189453125e-01 P7 */
+ /* iBrkValue = SP 2/3 */
+ .align 32
+ .long 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab
+ /* iOffExpoMask = SP significand mask */
+ .align 32
+ .long 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff
+ /* sBigThreshold */
+ .align 32
+ .long 0x4E800000, 0x4E800000, 0x4E800000, 0x4E800000, 0x4E800000, 0x4E800000, 0x4E800000, 0x4E800000
+ /* sC2 */
+ .align 32
+ .long 0x3EC00000, 0x3EC00000, 0x3EC00000, 0x3EC00000, 0x3EC00000, 0x3EC00000, 0x3EC00000, 0x3EC00000
+ /* sC3 */
+ .align 32
+ .long 0x3EA00000, 0x3EA00000, 0x3EA00000, 0x3EA00000, 0x3EA00000, 0x3EA00000, 0x3EA00000, 0x3EA00000
+ /* sHalf */
+ .align 32
+ .long 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000
+ /* sLargestFinite */
+ .align 32
+ .long 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF, 0x7F7FFFFF
+ /* sLittleThreshold */
+ .align 32
+ .long 0x3D800000, 0x3D800000, 0x3D800000, 0x3D800000, 0x3D800000, 0x3D800000, 0x3D800000, 0x3D800000
+ /* sSign */
+ .align 32
+ .long 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000
+ /* sThirtyOne */
+ .align 32
+ .long 0x41F80000, 0x41F80000, 0x41F80000, 0x41F80000, 0x41F80000, 0x41F80000, 0x41F80000, 0x41F80000
+ /* sTopMask8 */
+ .align 32
+ .long 0xFFFF0000, 0xFFFF0000, 0xFFFF0000, 0xFFFF0000, 0xFFFF0000, 0xFFFF0000, 0xFFFF0000, 0xFFFF0000
+ /* XScale */
+ .align 32
+ .long 0x30800000, 0x30800000, 0x30800000, 0x30800000, 0x30800000, 0x30800000, 0x30800000, 0x30800000
+ /* sLn2 = SP ln(2) */
+ .align 32
+ .long 0x3f317218, 0x3f317218, 0x3f317218, 0x3f317218, 0x3f317218, 0x3f317218, 0x3f317218, 0x3f317218
+ .align 32
+ .type __svml_sasinh_data_internal, @object
+ .size __svml_sasinh_data_internal, .-__svml_sasinh_data_internal