maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/Transforms/InstCombine/pow-4.ll

120 lines
3.5 KiB
LLVM
Raw Normal View History

[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; Test that the pow library call simplifier works correctly.
; RUN: opt -instcombine -S < %s | FileCheck %s
; Function Attrs: nounwind readnone
declare double @llvm.pow.f64(double, double)
declare float @llvm.pow.f32(float, float)
; pow(x, 4.0f)
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
define float @test_simplify_4f(float %x) {
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-LABEL: @test_simplify_4f(
; CHECK-NOT: pow
; CHECK-NEXT: %1 = fmul float %x, %x
; CHECK-NEXT: %2 = fmul float %1, %1
; CHECK-NEXT: ret float %2
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
%1 = call fast float @llvm.pow.f32(float %x, float 4.000000e+00)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
ret float %1
}
; pow(x, 3.0)
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
define double @test_simplify_3(double %x) {
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-LABEL: @test_simplify_3(
; CHECK-NOT: pow
; CHECK-NEXT: %1 = fmul double %x, %x
; CHECK-NEXT: %2 = fmul double %1, %x
; CHECK-NEXT: ret double %2
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
%1 = call fast double @llvm.pow.f64(double %x, double 3.000000e+00)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
ret double %1
}
; pow(x, 4.0)
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
define double @test_simplify_4(double %x) {
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-LABEL: @test_simplify_4(
; CHECK-NOT: pow
; CHECK-NEXT: %1 = fmul double %x, %x
; CHECK-NEXT: %2 = fmul double %1, %1
; CHECK-NEXT: ret double %2
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
%1 = call fast double @llvm.pow.f64(double %x, double 4.000000e+00)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
ret double %1
}
; pow(x, 15.0)
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
define double @test_simplify_15(double %x) {
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-LABEL: @test_simplify_15(
; CHECK-NOT: pow
; CHECK-NEXT: %1 = fmul double %x, %x
; CHECK-NEXT: %2 = fmul double %1, %x
; CHECK-NEXT: %3 = fmul double %2, %2
; CHECK-NEXT: %4 = fmul double %3, %3
; CHECK-NEXT: %5 = fmul double %2, %4
; CHECK-NEXT: ret double %5
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
%1 = call fast double @llvm.pow.f64(double %x, double 1.500000e+01)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
ret double %1
}
; pow(x, -7.0)
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
define double @test_simplify_neg_7(double %x) {
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-LABEL: @test_simplify_neg_7(
; CHECK-NOT: pow
; CHECK-NEXT: %1 = fmul double %x, %x
; CHECK-NEXT: %2 = fmul double %1, %x
; CHECK-NEXT: %3 = fmul double %1, %2
; CHECK-NEXT: %4 = fmul double %1, %3
; CHECK-NEXT: %5 = fdiv double 1.000000e+00, %4
; CHECK-NEXT: ret double %5
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
%1 = call fast double @llvm.pow.f64(double %x, double -7.000000e+00)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
ret double %1
}
; pow(x, -19.0)
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
define double @test_simplify_neg_19(double %x) {
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-LABEL: @test_simplify_neg_19(
; CHECK-NOT: pow
; CHECK-NEXT: %1 = fmul double %x, %x
; CHECK-NEXT: %2 = fmul double %1, %1
; CHECK-NEXT: %3 = fmul double %2, %2
; CHECK-NEXT: %4 = fmul double %3, %3
; CHECK-NEXT: %5 = fmul double %1, %4
; CHECK-NEXT: %6 = fmul double %5, %x
; CHECK-NEXT: %7 = fdiv double 1.000000e+00, %6
; CHECK-NEXT: ret double %7
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
%1 = call fast double @llvm.pow.f64(double %x, double -1.900000e+01)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
ret double %1
}
; pow(x, 11.23)
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
define double @test_simplify_11_23(double %x) {
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-LABEL: @test_simplify_11_23(
; CHECK-NOT: fmul
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
; CHECK-NEXT: %1 = call fast double @llvm.pow.f64(double %x, double 1.123000e+01)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-NEXT: ret double %1
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
%1 = call fast double @llvm.pow.f64(double %x, double 1.123000e+01)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
ret double %1
}
; pow(x, 32.0)
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
define double @test_simplify_32(double %x) {
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-LABEL: @test_simplify_32(
; CHECK-NOT: pow
; CHECK-NEXT: %1 = fmul double %x, %x
; CHECK-NEXT: %2 = fmul double %1, %1
; CHECK-NEXT: %3 = fmul double %2, %2
; CHECK-NEXT: %4 = fmul double %3, %3
; CHECK-NEXT: %5 = fmul double %4, %4
; CHECK-NEXT: ret double %5
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
%1 = call fast double @llvm.pow.f64(double %x, double 3.200000e+01)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
ret double %1
}
; pow(x, 33.0)
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
define double @test_simplify_33(double %x) {
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-LABEL: @test_simplify_33(
; CHECK-NOT: fmul
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
; CHECK-NEXT: %1 = call fast double @llvm.pow.f64(double %x, double 3.300000e+01)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
; CHECK-NEXT: ret double %1
[LibCallSimplifier] use instruction-level fast-math-flags to transform pow(x, [small integer]) calls This is a continuation of adding FMF to call instructions: http://reviews.llvm.org/rL255555 As with D15937, the intent of the patch is to preserve the current behavior of the transform except that we use the pow call's 'fast' attribute as a trigger rather than a function-level attribute. The TODO comment notes a potential follow-on patch that would propagate FMF to the new instructions. Differential Revision: http://reviews.llvm.org/D16122 llvm-svn: 258153
2016-01-20 02:15:12 +08:00
%1 = call fast double @llvm.pow.f64(double %x, double 3.300000e+01)
[SimplifyLibCalls] Optimization for pow(x, n) where n is some constant Summary: In order to avoid calling pow function we generate repeated fmul when n is a positive or negative whole number. For each exponent we pre-compute Addition Chains in order to minimize the no. of fmuls. Refer: http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html We pre-compute addition chains for exponents upto 32 (which results in a max of 7 fmuls). For eg: 4 = 2+2 5 = 2+3 6 = 3+3 and so on Hence, pow(x, 4.0) ==> y = fmul x, x x = fmul y, y ret x For negative exponents, we simply compute the reciprocal of the final result. Note: This transformation is only enabled under fast-math. Patch by Mandeep Singh Grang <mgrang@codeaurora.org> Reviewers: weimingz, majnemer, escha, davide, scanon, joerg Subscribers: probinson, escha, llvm-commits Differential Revision: http://reviews.llvm.org/D13994 llvm-svn: 254776
2015-12-05 06:00:47 +08:00
ret double %1
}