diff options
Diffstat (limited to 'sysdeps/x86_64/fpu')
-rw-r--r-- | sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S | 478 |
1 files changed, 238 insertions, 240 deletions
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S index 9062a1475e..b7c7ec9cb7 100644 --- a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S @@ -22,7 +22,7 @@ * exp10(x) = 2^x/log10(2) = 2^n * (1 + T[j]) * (1 + P(y)) * where * x = m*log10(2)/K + y, y in [-log10(2)/K..log10(2)/K] - * m = n*K + j, m,n,j - signed integer, j in [-K/2..K/2] + * m = n*K + j, m, n,j - signed integer, j in [-K/2..K/2] * * values of 2^j/K are tabulated * @@ -43,269 +43,267 @@ /* Offsets for data table __svml_sexp10_data_internal */ -#define _sT 0 -#define _sLg2_10 128 -#define _sShifter 144 -#define _sInvLg2_10hi 160 -#define _sInvLg2_10lo 176 -#define _sPC0 192 -#define _sPC1 208 -#define _sPC2 224 -#define _iIndexMask 240 -#define _iAbsMask 256 -#define _iDomainRange 272 +#define _sT 0 +#define _sLg2_10 128 +#define _sShifter 144 +#define _sInvLg2_10hi 160 +#define _sInvLg2_10lo 176 +#define _sPC0 192 +#define _sPC1 208 +#define _sPC2 224 +#define _iIndexMask 240 +#define _iAbsMask 256 +#define _iDomainRange 272 #include <sysdep.h> - .text - .section .text.sse4,"ax",@progbits + .section .text.sse4, "ax", @progbits ENTRY(_ZGVbN4v_exp10f_sse4) - subq $72, %rsp - cfi_def_cfa_offset(80) - movaps %xmm0, %xmm4 - -/* Load arument */ - movups _sLg2_10+__svml_sexp10_data_internal(%rip), %xmm2 - lea __svml_sexp10_data_internal(%rip), %r8 - mulps %xmm4, %xmm2 - movups _sShifter+__svml_sexp10_data_internal(%rip), %xmm5 - -/* R */ - movups _sInvLg2_10hi+__svml_sexp10_data_internal(%rip), %xmm14 - addps %xmm5, %xmm2 - movaps %xmm2, %xmm1 - movups _sInvLg2_10lo+__svml_sexp10_data_internal(%rip), %xmm15 - subps %xmm5, %xmm1 - mulps %xmm1, %xmm14 - movaps %xmm4, %xmm5 - mulps %xmm1, %xmm15 - subps %xmm14, %xmm5 - -/* - * Polynomial - * exp10 = 2^N*(Tj+Tj*poly) - * poly(sN) = {1+later} a0+a1*sR - */ - movups _sPC2+__svml_sexp10_data_internal(%rip), %xmm1 - subps %xmm15, %xmm5 - mulps %xmm5, %xmm1 - movdqu _iIndexMask+__svml_sexp10_data_internal(%rip), %xmm3 - -/* Index and lookup */ - movdqa %xmm3, %xmm10 - -/* remove index bits */ - pandn %xmm2, %xmm3 - pand %xmm2, %xmm10 - -/* 2^N */ - pslld $18, %xmm3 - -/* iIndex *= sizeof(S); */ - pslld $2, %xmm10 - addps _sPC1+__svml_sexp10_data_internal(%rip), %xmm1 - movd %xmm10, %edx - pshufd $1, %xmm10, %xmm7 - pshufd $2, %xmm10, %xmm9 - pshufd $3, %xmm10, %xmm11 - movd %xmm7, %ecx - movd %xmm9, %esi - movd %xmm11, %edi - -/* Check for overflow\underflow */ - movdqu _iAbsMask+__svml_sexp10_data_internal(%rip), %xmm6 - pand %xmm4, %xmm6 - mulps %xmm1, %xmm5 - movslq %edx, %rdx - addps _sPC0+__svml_sexp10_data_internal(%rip), %xmm5 - movslq %ecx, %rcx - movslq %esi, %rsi - movslq %edi, %rdi - movd (%r8,%rdx), %xmm0 - movd (%r8,%rcx), %xmm8 - movd (%r8,%rsi), %xmm13 - movd (%r8,%rdi), %xmm12 - punpckldq %xmm8, %xmm0 - punpckldq %xmm12, %xmm13 - punpcklqdq %xmm13, %xmm0 - -/* Tj_l+Tj_h*poly */ - mulps %xmm0, %xmm5 - pcmpgtd _iDomainRange+__svml_sexp10_data_internal(%rip), %xmm6 - addps %xmm5, %xmm0 - movmskps %xmm6, %eax - -/* quick mul 2^N */ - paddd %xmm3, %xmm0 - -/* Finish */ - testl %eax, %eax - -/* Go to special inputs processing branch */ - jne L(SPECIAL_VALUES_BRANCH) - # LOE rbx rbp r12 r13 r14 r15 eax xmm0 xmm4 - -/* Restore registers - * and exit the function - */ + subq $72, %rsp + cfi_def_cfa_offset(80) + movaps %xmm0, %xmm4 + + /* Load arument */ + movups _sLg2_10+__svml_sexp10_data_internal(%rip), %xmm2 + lea __svml_sexp10_data_internal(%rip), %r8 + mulps %xmm4, %xmm2 + movups _sShifter+__svml_sexp10_data_internal(%rip), %xmm5 + + /* R */ + movups _sInvLg2_10hi+__svml_sexp10_data_internal(%rip), %xmm14 + addps %xmm5, %xmm2 + movaps %xmm2, %xmm1 + movups _sInvLg2_10lo+__svml_sexp10_data_internal(%rip), %xmm15 + subps %xmm5, %xmm1 + mulps %xmm1, %xmm14 + movaps %xmm4, %xmm5 + mulps %xmm1, %xmm15 + subps %xmm14, %xmm5 + + /* + * Polynomial + * exp10 = 2^N*(Tj+Tj*poly) + * poly(sN) = {1+later} a0+a1*sR + */ + movups _sPC2+__svml_sexp10_data_internal(%rip), %xmm1 + subps %xmm15, %xmm5 + mulps %xmm5, %xmm1 + movdqu _iIndexMask+__svml_sexp10_data_internal(%rip), %xmm3 + + /* Index and lookup */ + movdqa %xmm3, %xmm10 + + /* remove index bits */ + pandn %xmm2, %xmm3 + pand %xmm2, %xmm10 + + /* 2^N */ + pslld $18, %xmm3 + + /* iIndex *= sizeof(S); */ + pslld $2, %xmm10 + addps _sPC1+__svml_sexp10_data_internal(%rip), %xmm1 + movd %xmm10, %edx + pshufd $1, %xmm10, %xmm7 + pshufd $2, %xmm10, %xmm9 + pshufd $3, %xmm10, %xmm11 + movd %xmm7, %ecx + movd %xmm9, %esi + movd %xmm11, %edi + + /* Check for overflow\underflow */ + movdqu _iAbsMask+__svml_sexp10_data_internal(%rip), %xmm6 + pand %xmm4, %xmm6 + mulps %xmm1, %xmm5 + movslq %edx, %rdx + addps _sPC0+__svml_sexp10_data_internal(%rip), %xmm5 + movslq %ecx, %rcx + movslq %esi, %rsi + movslq %edi, %rdi + movd (%r8, %rdx), %xmm0 + movd (%r8, %rcx), %xmm8 + movd (%r8, %rsi), %xmm13 + movd (%r8, %rdi), %xmm12 + punpckldq %xmm8, %xmm0 + punpckldq %xmm12, %xmm13 + punpcklqdq %xmm13, %xmm0 + + /* Tj_l+Tj_h*poly */ + mulps %xmm0, %xmm5 + pcmpgtd _iDomainRange+__svml_sexp10_data_internal(%rip), %xmm6 + addps %xmm5, %xmm0 + movmskps %xmm6, %eax + + /* quick mul 2^N */ + paddd %xmm3, %xmm0 + + /* Finish */ + testl %eax, %eax + + /* Go to special inputs processing branch */ + jne L(SPECIAL_VALUES_BRANCH) + # LOE rbx rbp r12 r13 r14 r15 eax xmm0 xmm4 + + /* Restore registers + * and exit the function + */ L(EXIT): - addq $72, %rsp - cfi_def_cfa_offset(8) - ret - cfi_def_cfa_offset(80) + addq $72, %rsp + cfi_def_cfa_offset(8) + ret + cfi_def_cfa_offset(80) -/* Branch to process - * special inputs - */ + /* Branch to process + * special inputs + */ L(SPECIAL_VALUES_BRANCH): - movups %xmm4, 32(%rsp) - movups %xmm0, 48(%rsp) - # LOE rbx rbp r12 r13 r14 r15 eax - - xorl %edx, %edx - movq %r12, 16(%rsp) - cfi_offset(12, -64) - movl %edx, %r12d - movq %r13, 8(%rsp) - cfi_offset(13, -72) - movl %eax, %r13d - movq %r14, (%rsp) - cfi_offset(14, -80) - # LOE rbx rbp r15 r12d r13d - -/* Range mask - * bits check - */ + movups %xmm4, 32(%rsp) + movups %xmm0, 48(%rsp) + # LOE rbx rbp r12 r13 r14 r15 eax + + xorl %edx, %edx + movq %r12, 16(%rsp) + cfi_offset(12, -64) + movl %edx, %r12d + movq %r13, 8(%rsp) + cfi_offset(13, -72) + movl %eax, %r13d + movq %r14, (%rsp) + cfi_offset(14, -80) + # LOE rbx rbp 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 rbp r15 r12d r13d + /* Call scalar math function */ + jc L(SCALAR_MATH_CALL) + # LOE rbx rbp r15 r12d r13d -/* Special inputs - * processing loop - */ + /* Special inputs + * processing loop + */ L(SPECIAL_VALUES_LOOP): - incl %r12d - cmpl $4, %r12d - -/* Check bits in range mask */ - jl L(RANGEMASK_CHECK) - # LOE rbx rbp r15 r12d r13d - - movq 16(%rsp), %r12 - cfi_restore(12) - movq 8(%rsp), %r13 - cfi_restore(13) - movq (%rsp), %r14 - cfi_restore(14) - movups 48(%rsp), %xmm0 - -/* Go to exit */ - jmp L(EXIT) - cfi_offset(12, -64) - cfi_offset(13, -72) - cfi_offset(14, -80) - # LOE rbx rbp r12 r13 r14 r15 xmm0 - -/* Scalar math fucntion call - * to process special input - */ + incl %r12d + cmpl $4, %r12d + + /* Check bits in range mask */ + jl L(RANGEMASK_CHECK) + # LOE rbx rbp r15 r12d r13d + + movq 16(%rsp), %r12 + cfi_restore(12) + movq 8(%rsp), %r13 + cfi_restore(13) + movq (%rsp), %r14 + cfi_restore(14) + movups 48(%rsp), %xmm0 + + /* Go to exit */ + jmp L(EXIT) + cfi_offset(12, -64) + cfi_offset(13, -72) + cfi_offset(14, -80) + # LOE rbx rbp r12 r13 r14 r15 xmm0 + + /* Scalar math fucntion call + * to process special input + */ L(SCALAR_MATH_CALL): - movl %r12d, %r14d - movss 32(%rsp,%r14,4), %xmm0 - call exp10f@PLT - # LOE rbx rbp r14 r15 r12d r13d xmm0 + movl %r12d, %r14d + movss 32(%rsp, %r14, 4), %xmm0 + call exp10f@PLT + # LOE rbx rbp r14 r15 r12d r13d xmm0 - movss %xmm0, 48(%rsp,%r14,4) + movss %xmm0, 48(%rsp, %r14, 4) -/* Process special inputs in loop */ - jmp L(SPECIAL_VALUES_LOOP) - # LOE rbx rbp r15 r12d r13d + /* Process special inputs in loop */ + jmp L(SPECIAL_VALUES_LOOP) + # LOE rbx rbp r15 r12d r13d END(_ZGVbN4v_exp10f_sse4) - .section .rodata, "a" - .align 16 + .section .rodata, "a" + .align 16 #ifdef __svml_sexp10_data_internal_typedef typedef unsigned int VUINT32; -typedef struct -{ - __declspec(align(16)) VUINT32 _sT[(1<<5)][1]; - __declspec(align(16)) VUINT32 _sLg2_10[4][1]; - __declspec(align(16)) VUINT32 _sShifter[4][1]; - __declspec(align(16)) VUINT32 _sInvLg2_10hi[4][1]; - __declspec(align(16)) VUINT32 _sInvLg2_10lo[4][1]; - __declspec(align(16)) VUINT32 _sPC0[4][1]; - __declspec(align(16)) VUINT32 _sPC1[4][1]; - __declspec(align(16)) VUINT32 _sPC2[4][1]; - __declspec(align(16)) VUINT32 _iIndexMask[4][1]; - __declspec(align(16)) VUINT32 _iAbsMask[4][1]; - __declspec(align(16)) VUINT32 _iDomainRange[4][1]; +typedef struct { + __declspec(align(16)) VUINT32 _sT[(1<<5)][1]; + __declspec(align(16)) VUINT32 _sLg2_10[4][1]; + __declspec(align(16)) VUINT32 _sShifter[4][1]; + __declspec(align(16)) VUINT32 _sInvLg2_10hi[4][1]; + __declspec(align(16)) VUINT32 _sInvLg2_10lo[4][1]; + __declspec(align(16)) VUINT32 _sPC0[4][1]; + __declspec(align(16)) VUINT32 _sPC1[4][1]; + __declspec(align(16)) VUINT32 _sPC2[4][1]; + __declspec(align(16)) VUINT32 _iIndexMask[4][1]; + __declspec(align(16)) VUINT32 _iAbsMask[4][1]; + __declspec(align(16)) VUINT32 _iDomainRange[4][1]; } __svml_sexp10_data_internal; #endif __svml_sexp10_data_internal: - /*== _sT ==*/ - .long 0x3f800000 // 2^( 0 /32 ) - .long 0x3f82cd87 // 2^( 1 /32 ) - .long 0x3f85aac3 // 2^( 2 /32 ) - .long 0x3f88980f // 2^( 3 /32 ) - .long 0x3f8b95c2 // 2^( 4 /32 ) - .long 0x3f8ea43a // 2^( 5 /32 ) - .long 0x3f91c3d3 // 2^( 6 /32 ) - .long 0x3f94f4f0 // 2^( 7 /32 ) - .long 0x3f9837f0 // 2^( 8 /32 ) - .long 0x3f9b8d3a // 2^( 9 /32 ) - .long 0x3f9ef532 // 2^( 10/32 ) - .long 0x3fa27043 // 2^( 11/32 ) - .long 0x3fa5fed7 // 2^( 12/32 ) - .long 0x3fa9a15b // 2^( 13/32 ) - .long 0x3fad583f // 2^( 14/32 ) - .long 0x3fb123f6 // 2^( 15/32 ) - .long 0x3fb504f3 // 2^( 16/32 ) - .long 0x3fb8fbaf // 2^( 17/32 ) - .long 0x3fbd08a4 // 2^( 18/32 ) - .long 0x3fc12c4d // 2^( 19/32 ) - .long 0x3fc5672a // 2^( 20/32 ) - .long 0x3fc9b9be // 2^( 21/32 ) - .long 0x3fce248c // 2^( 22/32 ) - .long 0x3fd2a81e // 2^( 23/32 ) - .long 0x3fd744fd // 2^( 24/32 ) - .long 0x3fdbfbb8 // 2^( 25/32 ) - .long 0x3fe0ccdf // 2^( 26/32 ) - .long 0x3fe5b907 // 2^( 27/32 ) - .long 0x3feac0c7 // 2^( 28/32 ) - .long 0x3fefe4ba // 2^( 29/32 ) - .long 0x3ff5257d // 2^( 30/32 ) - .long 0x3ffa83b3 // 2^( 31/32 ) - .align 16 - .long 0x42d49a78, 0x42d49a78, 0x42d49a78, 0x42d49a78 /* _sLg2_10*2^K */ - .align 16 - .long 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000 /* _sShifter) */ - .align 16 - .long 0x3c1a2000, 0x3c1a2000, 0x3c1a2000, 0x3c1a2000 /* _sInvLg2_10hi/2^K hi (24-K-7) bits*/ - .align 16 - .long 0x341a84fc, 0x341a84fc, 0x341a84fc, 0x341a84fc /* _sInvLg2_10lo/2^K lo bits */ - // otherwise exp10(0) won't produce exact 1.0 - .align 16 - .long 0x2fecc868, 0x2fecc868, 0x2fecc868, 0x2fecc868 /* _sPC0 */ - .align 16 - .long 0x40135e1b, 0x40135e1b, 0x40135e1b, 0x40135e1b /* _sPC1 */ - .align 16 - .long 0x4029a8d2, 0x4029a8d2, 0x4029a8d2, 0x4029a8d2 /* _sPC2 */ - .align 16 - .long 0x0000001f, 0x0000001f, 0x0000001f, 0x0000001f /* _iIndexMask =(2^K-1)*/ - //common - .align 16 - .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _iAbsMask */ - .align 16 - .long 0x4217b818, 0x4217b818, 0x4217b818, 0x4217b818 /* _iDomainRange=-log10(max_denormal=0x007fffff) RZ */ - .align 16 - .type __svml_sexp10_data_internal,@object - .size __svml_sexp10_data_internal,.-__svml_sexp10_data_internal + /* _sT */ + .long 0x3f800000 // 2^( 0 /32 ) + .long 0x3f82cd87 // 2^( 1 /32 ) + .long 0x3f85aac3 // 2^( 2 /32 ) + .long 0x3f88980f // 2^( 3 /32 ) + .long 0x3f8b95c2 // 2^( 4 /32 ) + .long 0x3f8ea43a // 2^( 5 /32 ) + .long 0x3f91c3d3 // 2^( 6 /32 ) + .long 0x3f94f4f0 // 2^( 7 /32 ) + .long 0x3f9837f0 // 2^( 8 /32 ) + .long 0x3f9b8d3a // 2^( 9 /32 ) + .long 0x3f9ef532 // 2^( 10/32 ) + .long 0x3fa27043 // 2^( 11/32 ) + .long 0x3fa5fed7 // 2^( 12/32 ) + .long 0x3fa9a15b // 2^( 13/32 ) + .long 0x3fad583f // 2^( 14/32 ) + .long 0x3fb123f6 // 2^( 15/32 ) + .long 0x3fb504f3 // 2^( 16/32 ) + .long 0x3fb8fbaf // 2^( 17/32 ) + .long 0x3fbd08a4 // 2^( 18/32 ) + .long 0x3fc12c4d // 2^( 19/32 ) + .long 0x3fc5672a // 2^( 20/32 ) + .long 0x3fc9b9be // 2^( 21/32 ) + .long 0x3fce248c // 2^( 22/32 ) + .long 0x3fd2a81e // 2^( 23/32 ) + .long 0x3fd744fd // 2^( 24/32 ) + .long 0x3fdbfbb8 // 2^( 25/32 ) + .long 0x3fe0ccdf // 2^( 26/32 ) + .long 0x3fe5b907 // 2^( 27/32 ) + .long 0x3feac0c7 // 2^( 28/32 ) + .long 0x3fefe4ba // 2^( 29/32 ) + .long 0x3ff5257d // 2^( 30/32 ) + .long 0x3ffa83b3 // 2^( 31/32 ) + .align 16 + .long 0x42d49a78, 0x42d49a78, 0x42d49a78, 0x42d49a78 /* _sLg2_10*2^K */ + .align 16 + .long 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000 /* _sShifter) */ + .align 16 + .long 0x3c1a2000, 0x3c1a2000, 0x3c1a2000, 0x3c1a2000 /* _sInvLg2_10hi/2^K hi (24-K-7) bits */ + .align 16 + .long 0x341a84fc, 0x341a84fc, 0x341a84fc, 0x341a84fc /* _sInvLg2_10lo/2^K lo bits */ + // otherwise exp10(0) won't produce exact 1.0 + .align 16 + .long 0x2fecc868, 0x2fecc868, 0x2fecc868, 0x2fecc868 /* _sPC0 */ + .align 16 + .long 0x40135e1b, 0x40135e1b, 0x40135e1b, 0x40135e1b /* _sPC1 */ + .align 16 + .long 0x4029a8d2, 0x4029a8d2, 0x4029a8d2, 0x4029a8d2 /* _sPC2 */ + .align 16 + .long 0x0000001f, 0x0000001f, 0x0000001f, 0x0000001f /* _iIndexMask =(2^K-1) */ + //common + .align 16 + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _iAbsMask */ + .align 16 + .long 0x4217b818, 0x4217b818, 0x4217b818, 0x4217b818 /* _iDomainRange=-log10(max_denormal=0x007fffff) RZ */ + .align 16 + .type __svml_sexp10_data_internal, @object + .size __svml_sexp10_data_internal, .-__svml_sexp10_data_internal |