diff options
author | Ulrich Drepper <drepper@redhat.com> | 2005-03-31 10:02:53 +0000 |
---|---|---|
committer | Ulrich Drepper <drepper@redhat.com> | 2005-03-31 10:02:53 +0000 |
commit | ee6189855aab3a9be8f3c2d95ce2b2cd17db4ec2 (patch) | |
tree | cf3e2fe1f9be5b358033fd927a0bd922542e04a1 /sysdeps/ia64/fpu/e_scalb.S | |
parent | 4d6302cf51b16a129addf7687c91490c40a7225c (diff) | |
download | glibc-ee6189855aab3a9be8f3c2d95ce2b2cd17db4ec2.tar.gz glibc-ee6189855aab3a9be8f3c2d95ce2b2cd17db4ec2.tar.xz glibc-ee6189855aab3a9be8f3c2d95ce2b2cd17db4ec2.zip |
* sysdeps/unix/sysv/linux/x86_64/getcontext.S: Use functionally
equivalent, but shorter instructions. * sysdeps/unix/sysv/linux/x86_64/sysdep.h: Likewise. * sysdeps/unix/sysv/linux/x86_64/setcontext.S: Likewise. * sysdeps/unix/sysv/linux/x86_64/clone.S: Likewise. * sysdeps/unix/sysv/linux/x86_64/swapcontext.S: Likewise. * sysdeps/unix/x86_64/sysdep.S: Likewise. * sysdeps/x86_64/strchr.S: Likewise. * sysdeps/x86_64/memset.S: Likewise. * sysdeps/x86_64/strcspn.S: Likewise. * sysdeps/x86_64/strcmp.S: Likewise. * sysdeps/x86_64/elf/start.S: Likewise. * sysdeps/x86_64/strspn.S: Likewise. * sysdeps/x86_64/dl-machine.h: Likewise. * sysdeps/x86_64/bsd-_setjmp.S: Likewise. * sysdeps/x86_64/bsd-setjmp.S: Likewise. * sysdeps/x86_64/strtok.S: Likewise.
Diffstat (limited to 'sysdeps/ia64/fpu/e_scalb.S')
-rw-r--r-- | sysdeps/ia64/fpu/e_scalb.S | 630 |
1 files changed, 341 insertions, 289 deletions
diff --git a/sysdeps/ia64/fpu/e_scalb.S b/sysdeps/ia64/fpu/e_scalb.S index 82e914e259..3d48aab189 100644 --- a/sysdeps/ia64/fpu/e_scalb.S +++ b/sysdeps/ia64/fpu/e_scalb.S @@ -21,60 +21,82 @@ // products derived from this software without specific prior written // permission. -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, -// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, -// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR -// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY +// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING -// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS -// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -// +// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +// // Intel Corporation is the author of this code, and requests that all -// problem reports or change requests be submitted to it directly at +// problem reports or change requests be submitted to it directly at // http://www.intel.com/software/products/opensource/libraries/num.htm. // // History //============================================================== // 02/02/00 Initial version -// 01/26/01 Scalb completely reworked and now standalone version +// 01/26/01 Scalb completely reworked and now standalone version // 05/20/02 Cleaned up namespace and sf0 syntax // 02/10/03 Reordered header: .section, .global, .proc, .align +// 08/06/03 Improved performance // // API //============================================================== -// double = scalb (double x, double n) +// double = scalb (double x, double n) // input floating point f8 and floating point f9 // output floating point f8 // +// int_type = 0 if int is 32 bits +// int_type = 1 if int is 64 bits +// // Returns x* 2**n using an fma and detects overflow -// and underflow. +// and underflow. // // +// Strategy: +// Compute biased exponent of result exp_Result = N + exp_X +// Break into ranges: +// exp_Result > 0x103fe -> Certain overflow +// exp_Result = 0x103fe -> Possible overflow +// 0x0fc01 <= exp_Result < 0x103fe -> No over/underflow (main path) +// 0x0fc01 - 52 <= exp_Result < 0x0fc01 -> Possible underflow +// exp_Result < 0x0fc01 - 52 -> Certain underflow +FR_Big = f6 +FR_NBig = f7 FR_Floating_X = f8 FR_Result = f8 FR_Floating_N = f9 FR_Result2 = f9 -FR_Norm_N = f10 -FR_Result3 = f11 -FR_Norm_X = f12 +FR_Result3 = f10 +FR_Norm_X = f11 +FR_Two_N = f12 FR_N_float_int = f13 -FR_Two_N = f14 -FR_Two_to_Big = f15 -FR_Big = f6 -FR_NBig = f7 +FR_Norm_N = f14 +GR_neg_ov_limit= r14 +GR_big_exp = r14 GR_N_Biased = r15 GR_Big = r16 -GR_NBig = r17 -GR_Scratch = r18 -GR_Scratch1 = r19 +GR_exp_Result = r18 +GR_pos_ov_limit= r19 +GR_exp_sure_ou = r19 GR_Bias = r20 GR_N_as_int = r21 +GR_signexp_X = r22 +GR_exp_X = r23 +GR_exp_mask = r24 +GR_max_exp = r25 +GR_min_exp = r26 +GR_min_den_exp = r27 +GR_Scratch = r28 +GR_signexp_N = r29 +GR_exp_N = r30 GR_SAVE_B0 = r32 GR_SAVE_GP = r33 @@ -89,412 +111,442 @@ GLOBAL_IEEE754_ENTRY(scalb) // // Is x NAN, INF, ZERO, +-? +// Build the exponent Bias // { .mfi - alloc r32=ar.pfs,0,3,4,0 - fclass.m.unc p7,p0 = FR_Floating_X, 0xe7 //@snan | @qnan | @inf | @zero - addl GR_Scratch = 0x019C3F,r0 + getf.exp GR_signexp_N = FR_Floating_N // Get signexp of n + fclass.m p6,p0 = FR_Floating_X, 0xe7 // @snan | @qnan | @inf | @zero + mov GR_Bias = 0x0ffff +} +{ .mfi + mov GR_Big = 35000 // If N this big then certain overflow + fcvt.fx.trunc.s1 FR_N_float_int = FR_Floating_N // Get N in significand + nop.i 0 +} +;; + +{ .mfi + getf.exp GR_signexp_X = FR_Floating_X // Get signexp of x + fclass.m p7,p0 = FR_Floating_N, 0x0b // Test for n=unorm + nop.i 0 } // -// Is y a NAN, INF, ZERO, +-? +// Normalize n // { .mfi - nop.m 999 - fclass.m.unc p6,p0 = FR_Floating_N, 0xe7 //@snan | @qnan | @inf | @zero - addl GR_Scratch1 = 0x063BF,r0 + mov GR_exp_mask = 0x1ffff // Exponent mask + fnorm.s1 FR_Norm_N = FR_Floating_N + nop.i 0 } ;; // -// Convert N to a fp integer -// Normalize x +// Is n NAN, INF, ZERO, +-? // { .mfi - nop.m 0 - fnorm.s1 FR_Norm_N = FR_Floating_N - nop.i 999 + mov GR_big_exp = 0x1003e // Exponent at which n is integer + fclass.m p9,p0 = FR_Floating_N, 0xe7 // @snan | @qnan | @inf | @zero + mov GR_max_exp = 0x103fe // Exponent of maximum double } -{ .mfi - nop.m 999 - fnorm.s1 FR_Norm_X = FR_Floating_X - nop.i 999 -};; - // -// Create 2*big -// Create 2**-big // Normalize x -// Branch on special values. // -{ .mib - setf.exp FR_Big = GR_Scratch - nop.i 0 -(p6) br.cond.spnt SCALB_NAN_INF_ZERO +{ .mfb + nop.m 0 + fnorm.s1 FR_Norm_X = FR_Floating_X +(p7) br.cond.spnt SCALB_N_UNORM // Branch if n=unorm } -{ .mib - setf.exp FR_NBig = GR_Scratch1 - nop.i 0 -(p7) br.cond.spnt SCALB_NAN_INF_ZERO -};; +;; -// -// Convert N to a fp integer -// Create -35000 -// +SCALB_COMMON1: +// Main path continues. Also return here from u=unorm path. +// Handle special cases if x = Nan, Inf, Zero +{ .mfb + nop.m 0 + fcmp.lt.s1 p7,p0 = FR_Floating_N, f0 // Test N negative +(p6) br.cond.spnt SCALB_NAN_INF_ZERO +} +;; + +// Handle special cases if n = Nan, Inf, Zero { .mfi - addl GR_Scratch = 1,r0 - fcvt.fx.trunc.s1 FR_N_float_int = FR_Norm_N - addl GR_NBig = -35000,r0 + getf.sig GR_N_as_int = FR_N_float_int // Get n from significand + fclass.m p8,p0 = FR_Floating_X, 0x0b // Test for x=unorm + mov GR_exp_sure_ou = 0x1000e // Exp_N where x*2^N sure over/under +} +{ .mfb + mov GR_min_exp = 0x0fc01 // Exponent of minimum double + fcvt.xf FR_N_float_int = FR_N_float_int // Convert N to FP integer +(p9) br.cond.spnt SCALB_NAN_INF_ZERO } ;; -// -// Put N if a GP register -// Convert N_float_int to floating point value -// Create 35000 -// Build the exponent Bias -// -{ .mii - getf.sig GR_N_as_int = FR_N_float_int - shl GR_Scratch = GR_Scratch,63 - addl GR_Big = 35000,r0 +{ .mmi + and GR_exp_N = GR_exp_mask, GR_signexp_N // Get exponent of N +(p7) sub GR_Big = r0, GR_Big // Limit for N + nop.i 0 } -{ .mfi - addl GR_Bias = 0x0FFFF,r0 - fcvt.xf FR_N_float_int = FR_N_float_int - nop.i 0 -};; +;; -// -// Catch those fp values that are beyond 2**64-1 -// Is N > 35000 -// Is N < -35000 -// -{ .mfi - cmp.ne.unc p9,p10 = GR_N_as_int,GR_Scratch - nop.f 0 - nop.i 0 +{ .mib + cmp.lt p9,p0 = GR_exp_N, GR_big_exp // N possible non-integer? + cmp.ge p6,p0 = GR_exp_N, GR_exp_sure_ou // N certain over/under? +(p8) br.cond.spnt SCALB_X_UNORM // Branch if x=unorm } -{ .mmi - cmp.ge.unc p6, p0 = GR_N_as_int, GR_Big - cmp.le.unc p8, p0 = GR_N_as_int, GR_NBig - nop.i 0 -};; +;; -// -// Is N really an int, only for those non-int indefinites? -// Create exp bias. -// -{ .mfi - add GR_N_Biased = GR_Bias,GR_N_as_int -(p9) fcmp.neq.unc.s1 p7,p0 = FR_Norm_N, FR_N_float_int - nop.i 0 -};; +SCALB_COMMON2: +// Main path continues. Also return here from x=unorm path. +// Create biased exponent for 2**N +{ .mmi +(p6) mov GR_N_as_int = GR_Big // Limit N +;; + add GR_N_Biased = GR_Bias,GR_N_as_int + nop.i 0 +} +;; -// -// Branch and return if N is not an int. -// Main path, create 2**N -// { .mfi - setf.exp FR_Two_N = GR_N_Biased - nop.i 999 + setf.exp FR_Two_N = GR_N_Biased // Form 2**N +(p9) fcmp.neq.unc.s1 p9,p0 = FR_Norm_N, FR_N_float_int // Test if N an integer + and GR_exp_X = GR_exp_mask, GR_signexp_X // Get exponent of X } -{ .mfb - nop.m 0 -(p7) frcpa.s0 f8,p11 = f0,f0 -(p7) br.ret.spnt b0 -};; +;; // -// Set denormal on denormal input x and denormal input N +// Compute biased result exponent +// Branch if N is not an integer // -{ .mfi - nop.m 999 -(p10)fcmp.ge.s1 p6,p8 = FR_Norm_N,f0 - nop.i 0 -};; -{ .mfi - nop.m 999 - fcmp.ge.s0 p0,p11 = FR_Floating_X,f0 - nop.i 999 +{ .mib + add GR_exp_Result = GR_exp_X, GR_N_as_int + mov GR_min_den_exp = 0x0fc01 - 52 // Exponent of min denorm dble +(p9) br.cond.spnt SCALB_N_NOT_INT } -{ .mfi - nop.m 999 - fcmp.ge.s0 p12,p13 = FR_Floating_N,f0 - nop.i 0 -};; +;; // -// Adjust 2**N if N was very small or very large +// Raise Denormal operand flag with compare +// Do final operation // - { .mfi - nop.m 0 -(p6) fma.s1 FR_Two_N = FR_Big,f1,f0 - nop.i 0 + cmp.lt p7,p6 = GR_exp_Result, GR_max_exp // Test no overflow + fcmp.ge.s0 p0,p11 = FR_Floating_X,FR_Floating_N // Dummy to set denorm + cmp.lt p9,p0 = GR_exp_Result, GR_min_den_exp // Test sure underflow } -{ .mlx - nop.m 999 - movl GR_Scratch = 0x00000000000303FF -};; -{ .mfi - nop.m 0 -(p8) fma.s1 FR_Two_N = FR_NBig,f1,f0 - nop.i 0 +{ .mfb + nop.m 0 + fma.d.s0 FR_Result = FR_Two_N,FR_Norm_X,f0 +(p9) br.cond.spnt SCALB_UNDERFLOW // Branch if certain underflow } -{ .mlx - nop.m 999 - movl GR_Scratch1= 0x00000000000103FF -};; +;; + +{ .mib +(p6) cmp.gt.unc p6,p8 = GR_exp_Result, GR_max_exp // Test sure overflow +(p7) cmp.ge.unc p7,p9 = GR_exp_Result, GR_min_exp // Test no over/underflow +(p7) br.ret.sptk b0 // Return from main path +} +;; -// Set up necessary status fields +{ .bbb +(p6) br.cond.spnt SCALB_OVERFLOW // Branch if certain overflow +(p8) br.cond.spnt SCALB_POSSIBLE_OVERFLOW // Branch if possible overflow +(p9) br.cond.spnt SCALB_POSSIBLE_UNDERFLOW // Branch if possible underflow +} +;; + +// Here if possible underflow. +// Resulting exponent: 0x0fc01-52 <= exp_Result < 0x0fc01 +SCALB_POSSIBLE_UNDERFLOW: +// +// Here if possible overflow. +// Resulting exponent: 0x103fe = exp_Result +SCALB_POSSIBLE_OVERFLOW: + +// Set up necessary status fields // // S0 user supplied status // S2 user supplied status + WRE + TD (Overflows) // S3 user supplied status + FZ + TD (Underflows) // { .mfi - nop.m 999 - fsetc.s3 0x7F,0x41 - nop.i 999 + mov GR_pos_ov_limit = 0x103ff // Exponent for positive overflow + fsetc.s3 0x7F,0x41 + nop.i 0 } { .mfi - nop.m 999 - fsetc.s2 0x7F,0x42 - nop.i 999 -};; + mov GR_neg_ov_limit = 0x303ff // Exponent for negative overflow + fsetc.s2 0x7F,0x42 + nop.i 0 +} +;; // -// Do final operation +// Do final operation with s2 and s3 // { .mfi - setf.exp FR_NBig = GR_Scratch - fma.d.s0 FR_Result = FR_Two_N,FR_Norm_X,f0 - nop.i 999 + setf.exp FR_NBig = GR_neg_ov_limit + fma.d.s3 FR_Result3 = FR_Two_N,FR_Norm_X,f0 + nop.i 0 } { .mfi - nop.m 999 - fma.d.s3 FR_Result3 = FR_Two_N,FR_Norm_X,f0 - nop.i 999 -};; -{ .mfi - setf.exp FR_Big = GR_Scratch1 - fma.d.s2 FR_Result2 = FR_Two_N,FR_Norm_X,f0 - nop.i 999 -};; + setf.exp FR_Big = GR_pos_ov_limit + fma.d.s2 FR_Result2 = FR_Two_N,FR_Norm_X,f0 + nop.i 0 +} +;; // Check for overflow or underflow. -// -// S0 user supplied status -// S2 user supplied status + WRE + TD (Overflow) -// S3 user supplied status + FZ + TD (Underflow) -// -// // Restore s3 // Restore s2 // { .mfi - nop.m 0 - fsetc.s3 0x7F,0x40 - nop.i 999 + nop.m 0 + fsetc.s3 0x7F,0x40 + nop.i 0 } { .mfi - nop.m 0 - fsetc.s2 0x7F,0x40 - nop.i 999 -};; + nop.m 0 + fsetc.s2 0x7F,0x40 + nop.i 0 +} +;; // // Is the result zero? // { .mfi - nop.m 999 - fclass.m.unc p6, p0 = FR_Result3, 0x007 - nop.i 999 -} + nop.m 0 + fclass.m p6, p0 = FR_Result3, 0x007 + nop.i 0 +} { .mfi - addl GR_Tag = 53, r0 - fcmp.ge.unc.s1 p7, p8 = FR_Result2 , FR_Big - nop.i 0 -};; + nop.m 0 + fcmp.ge.s1 p7, p8 = FR_Result2 , FR_Big + nop.i 0 +} +;; // // Detect masked underflow - Tiny + Inexact Only // { .mfi - nop.m 999 + nop.m 0 (p6) fcmp.neq.unc.s1 p6, p0 = FR_Result , FR_Result2 - nop.i 999 -};; + nop.i 0 +} +;; // // Is result bigger the allowed range? // Branch out for underflow // { .mfb -(p6) addl GR_Tag = 54, r0 + nop.m 0 (p8) fcmp.le.unc.s1 p9, p10 = FR_Result2 , FR_NBig -(p6) br.cond.spnt SCALB_UNDERFLOW -};; +(p6) br.cond.spnt SCALB_UNDERFLOW +} +;; // // Branch out for overflow // -{ .mbb - nop.m 0 -(p7) br.cond.spnt SCALB_OVERFLOW -(p9) br.cond.spnt SCALB_OVERFLOW -};; +{ .bbb +(p7) br.cond.spnt SCALB_OVERFLOW +(p9) br.cond.spnt SCALB_OVERFLOW + br.ret.sptk b0 // Return from main path. +} +;; -// -// Return from main path. -// -{ .mfb - nop.m 999 - nop.f 0 - br.ret.sptk b0;; +// Here if result overflows +SCALB_OVERFLOW: +{ .mib + alloc r32=ar.pfs,3,0,4,0 + addl GR_Tag = 53, r0 // Set error tag for overflow + br.cond.sptk __libm_error_region // Call error support for overflow +} +;; + +// Here if result underflows +SCALB_UNDERFLOW: +{ .mib + alloc r32=ar.pfs,3,0,4,0 + addl GR_Tag = 54, r0 // Set error tag for underflow + br.cond.sptk __libm_error_region // Call error support for underflow } +;; -SCALB_NAN_INF_ZERO: +SCALB_NAN_INF_ZERO: // -// Convert N to a fp integer -// +// Before entry, N has been converted to a fp integer in significand of +// FR_N_float_int +// +// Convert N_float_int to floating point value +// { .mfi - addl GR_Scratch = 1,r0 - fcvt.fx.trunc.s1 FR_N_float_int = FR_Norm_N - nop.i 999 + getf.sig GR_N_as_int = FR_N_float_int + fclass.m p6,p0 = FR_Floating_N, 0xc3 //@snan | @qnan + nop.i 0 } { .mfi - nop.m 0 - fclass.m.unc p6,p0 = FR_Floating_N, 0xc3 //@snan | @qnan - nop.i 0 -};; + addl GR_Scratch = 1,r0 + fcvt.xf FR_N_float_int = FR_N_float_int + nop.i 0 +} +;; + { .mfi - nop.m 0 - fclass.m.unc p7,p0 = FR_Floating_X, 0xc3 //@snan | @qnan - shl GR_Scratch = GR_Scratch,63 -};; + nop.m 0 + fclass.m p7,p0 = FR_Floating_X, 0xc3 //@snan | @qnan + shl GR_Scratch = GR_Scratch,63 +} +;; + { .mfi - nop.m 0 - fclass.m.unc p8,p0 = FR_Floating_N, 0x21 // @inf - nop.i 0 -} - { .mfi - nop.m 0 - fclass.m.unc p9,p0 = FR_Floating_N, 0x22 // @-inf - nop.i 0 -};; + nop.m 0 + fclass.m p8,p0 = FR_Floating_N, 0x21 // @inf + nop.i 0 +} +{ .mfi + nop.m 0 + fclass.m p9,p0 = FR_Floating_N, 0x22 // @-inf + nop.i 0 +} +;; // // Either X or N is a Nan, return result and possible raise invalid. // { .mfb - nop.m 0 -(p6) fma.d.s0 FR_Result = FR_Floating_N,FR_Floating_X,f0 + nop.m 0 +(p6) fma.d.s0 FR_Result = FR_Floating_N,FR_Floating_X,f0 (p6) br.ret.spnt b0 -};; +} +;; + { .mfb - getf.sig GR_N_as_int = FR_N_float_int -(p7) fma.d.s0 FR_Result = FR_Floating_N,FR_Floating_X,f0 + nop.m 0 +(p7) fma.d.s0 FR_Result = FR_Floating_N,FR_Floating_X,f0 (p7) br.ret.spnt b0 -};; +} +;; // // If N + Inf do something special // For N = -Inf, create Int // { .mfb - nop.m 0 -(p8) fma.d.s0 FR_Result = FR_Floating_X, FR_Floating_N,f0 -(p8) br.ret.spnt b0 + nop.m 0 +(p8) fma.d.s0 FR_Result = FR_Floating_X, FR_Floating_N,f0 +(p8) br.ret.spnt b0 } { .mfi - nop.m 0 -(p9) fnma.d.s0 FR_Floating_N = FR_Floating_N, f1, f0 - nop.i 0 -};; + nop.m 0 +(p9) fnma.d.s0 FR_Floating_N = FR_Floating_N, f1, f0 + nop.i 0 +} +;; // // If N==-Inf,return x/(-N) // { .mfb - nop.m 0 -(p9) frcpa.s0 FR_Result,p6 = FR_Floating_X,FR_Floating_N -(p9) br.ret.spnt b0 -};; - -// -// Convert N_float_int to floating point value -// -{ .mfi - cmp.ne.unc p9,p0 = GR_N_as_int,GR_Scratch - fcvt.xf FR_N_float_int = FR_N_float_int - nop.i 0 -};; + cmp.ne p7,p0 = GR_N_as_int,GR_Scratch +(p9) frcpa.s0 FR_Result,p0 = FR_Floating_X,FR_Floating_N +(p9) br.ret.spnt b0 +} +;; // // Is N an integer. // { .mfi - nop.m 0 -(p9) fcmp.neq.unc.s1 p7,p0 = FR_Norm_N, FR_N_float_int - nop.i 0 -};; + nop.m 0 +(p7) fcmp.neq.unc.s1 p7,p0 = FR_Norm_N, FR_N_float_int + nop.i 0 +} +;; // // If N not an int, return NaN and raise invalid. // { .mfb - nop.m 0 -(p7) frcpa.s0 FR_Result,p6 = f0,f0 -(p7) br.ret.spnt b0 -};; + nop.m 0 +(p7) frcpa.s0 FR_Result,p0 = f0,f0 +(p7) br.ret.spnt b0 +} +;; // -// Always return x in other path. +// Always return x in other path. // { .mfb - nop.m 0 - fma.d.s0 FR_Result = FR_Floating_X,f1,f0 - br.ret.sptk b0 -};; + nop.m 0 + fma.d.s0 FR_Result = FR_Floating_X,f1,f0 + br.ret.sptk b0 +} +;; -GLOBAL_IEEE754_END(scalb) -__libm_error_region: +// Here if n not int +// Return NaN and raise invalid. +SCALB_N_NOT_INT: +{ .mfb + nop.m 0 + frcpa.s0 FR_Result,p0 = f0,f0 + br.ret.sptk b0 +} +;; + +// Here if n=unorm +SCALB_N_UNORM: +{ .mfb + getf.exp GR_signexp_N = FR_Norm_N // Get signexp of normalized n + fcvt.fx.trunc.s1 FR_N_float_int = FR_Norm_N // Get N in significand + br.cond.sptk SCALB_COMMON1 // Return to main path +} +;; + +// Here if x=unorm +SCALB_X_UNORM: +{ .mib + getf.exp GR_signexp_X = FR_Norm_X // Get signexp of normalized x + nop.i 0 + br.cond.sptk SCALB_COMMON2 // Return to main path +} +;; -SCALB_OVERFLOW: -SCALB_UNDERFLOW: +GLOBAL_IEEE754_END(scalb) +LOCAL_LIBM_ENTRY(__libm_error_region) // // Get stack address of N // .prologue { .mfi - add GR_Parameter_Y=-32,sp + add GR_Parameter_Y=-32,sp nop.f 0 .save ar.pfs,GR_SAVE_PFS - mov GR_SAVE_PFS=ar.pfs + mov GR_SAVE_PFS=ar.pfs } // -// Adjust sp +// Adjust sp // { .mfi .fframe 64 - add sp=-64,sp + add sp=-64,sp nop.f 0 - mov GR_SAVE_GP=gp + mov GR_SAVE_GP=gp };; // -// Store N on stack in correct position +// Store N on stack in correct position // Locate the address of x on stack // { .mmi - stfd [GR_Parameter_Y] = FR_Norm_N,16 - add GR_Parameter_X = 16,sp + stfd [GR_Parameter_Y] = FR_Norm_N,16 + add GR_Parameter_X = 16,sp .save b0, GR_SAVE_B0 - mov GR_SAVE_B0=b0 + mov GR_SAVE_B0=b0 };; // @@ -503,42 +555,42 @@ SCALB_UNDERFLOW: // .body { .mib - stfd [GR_Parameter_X] = FR_Norm_X - add GR_Parameter_RESULT = 0,GR_Parameter_Y + stfd [GR_Parameter_X] = FR_Norm_X + add GR_Parameter_RESULT = 0,GR_Parameter_Y nop.b 0 } { .mib - stfd [GR_Parameter_Y] = FR_Result + stfd [GR_Parameter_Y] = FR_Result add GR_Parameter_Y = -16,GR_Parameter_Y - br.call.sptk b0=__libm_error_support# + br.call.sptk b0=__libm_error_support# };; // // Get location of result on stack // { .mmi + add GR_Parameter_RESULT = 48,sp nop.m 0 - nop.m 0 - add GR_Parameter_RESULT = 48,sp + nop.i 0 };; // -// Get the new result +// Get the new result // { .mmi - ldfd FR_Result = [GR_Parameter_RESULT] + ldfd FR_Result = [GR_Parameter_RESULT] .restore sp - add sp = 64,sp - mov b0 = GR_SAVE_B0 + add sp = 64,sp + mov b0 = GR_SAVE_B0 };; // // Restore gp, ar.pfs and return // { .mib - mov gp = GR_SAVE_GP - mov ar.pfs = GR_SAVE_PFS - br.ret.sptk b0 + mov gp = GR_SAVE_GP + mov ar.pfs = GR_SAVE_PFS + br.ret.sptk b0 };; LOCAL_LIBM_END(__libm_error_region) |