From a0983ed3d2c825ed959536b70e944e551fd5fc13 Mon Sep 17 00:00:00 2001
From: laith sakka <laith.s.sakka@gmail.com>
Date: Thu, 2 Apr 2020 21:10:54 -0700
Subject: [PATCH] Handle exp2 with proper vectorization and lowering to SVML
 calls

Summary:
Add mapping from exp2 math functions
to corresponding SVML calls.

This is a follow up and extension for llvm diff
https://reviews.llvm.org/D19544

Test Plan:
- update test case and run ninja check.
- run tests locally

Reviewers: wenlei, hoyFB, mmasten, mzolotukhin, spatel

Reviewed By: spatel

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D77114
---
 llvm/include/llvm/Analysis/VecFuncs.def            |  23 ++++
 .../Transforms/LoopVectorize/X86/svml-calls.ll     | 145 ++++++++++++++++++++-
 2 files changed, 167 insertions(+), 1 deletion(-)

diff --git a/llvm/include/llvm/Analysis/VecFuncs.def b/llvm/include/llvm/Analysis/VecFuncs.def
index 86bec0b..2f64b0f 100644
--- a/llvm/include/llvm/Analysis/VecFuncs.def
+++ b/llvm/include/llvm/Analysis/VecFuncs.def
@@ -245,6 +245,29 @@ TLI_DEFINE_VECFUNC("llvm.log.f32", "__svml_logf4", 4)
 TLI_DEFINE_VECFUNC("llvm.log.f32", "__svml_logf8", 8)
 TLI_DEFINE_VECFUNC("llvm.log.f32", "__svml_logf16", 16)
 
+TLI_DEFINE_VECFUNC("exp2", "__svml_exp22", 2)
+TLI_DEFINE_VECFUNC("exp2", "__svml_exp24", 4)
+TLI_DEFINE_VECFUNC("exp2", "__svml_exp28", 8)
+
+TLI_DEFINE_VECFUNC("exp2f", "__svml_exp2f4", 4)
+TLI_DEFINE_VECFUNC("exp2f", "__svml_exp2f8", 8)
+TLI_DEFINE_VECFUNC("exp2f", "__svml_exp2f16", 16)
+
+TLI_DEFINE_VECFUNC("llvm.exp2.f64", "__svml_exp22", 2)
+TLI_DEFINE_VECFUNC("llvm.exp2.f64", "__svml_exp24", 4)
+TLI_DEFINE_VECFUNC("llvm.exp2.f64", "__svml_exp28", 8)
+
+TLI_DEFINE_VECFUNC("llvm.exp2.f32", "__svml_exp2f4", 4)
+TLI_DEFINE_VECFUNC("llvm.exp2.f32", "__svml_exp2f8", 8)
+TLI_DEFINE_VECFUNC("llvm.exp2.f32", "__svml_exp2f16", 16)
+
+TLI_DEFINE_VECFUNC("__exp2_finite", "__svml_exp22", 2)
+TLI_DEFINE_VECFUNC("__exp2_finite", "__svml_exp24", 4)
+TLI_DEFINE_VECFUNC("__exp2_finite", "__svml_exp28", 8)
+
+TLI_DEFINE_VECFUNC("__exp2f_finite", "__svml_exp2f4", 4)
+TLI_DEFINE_VECFUNC("__exp2f_finite", "__svml_exp2f8", 8)
+TLI_DEFINE_VECFUNC("__exp2f_finite", "__svml_exp2f16", 16)
 
 #else
 #error "Must choose which vector library functions are to be defined."
diff --git a/llvm/test/Transforms/LoopVectorize/X86/svml-calls.ll b/llvm/test/Transforms/LoopVectorize/X86/svml-calls.ll
index 8ff62f1..c1af2cb 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/svml-calls.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/svml-calls.ll
@@ -28,6 +28,12 @@ declare float @logf(float) #0
 declare double @llvm.log.f64(double) #0
 declare float @llvm.log.f32(float) #0
 
+declare double @exp2(double) #0
+declare float @exp2f(float) #0
+declare double @llvm.exp2.f64(double) #0
+declare float @llvm.exp2.f32(float) #0
+declare double @__exp2_finite(double) #0
+declare float @__exp2f_finite(float) #0
 
 define void @sin_f64(double* nocapture %varray) {
 ; CHECK-LABEL: @sin_f64(
@@ -497,5 +503,142 @@ for.end:
   ret void
 }
 
-attributes #0 = { nounwind readnone }
+define void @exp2_f64(double* nocapture %varray) {
+; CHECK-LABEL: @exp2_f64(
+; CHECK:    [[TMP5:%.*]] = call <4 x double> @__svml_exp24(<4 x double> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call double @exp2(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %iv
+  store double %call, double* %arrayidx, align 4
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:
+  ret void
+}
+
+define void @exp2_f32(float* nocapture %varray) {
+; CHECK-LABEL: @exp2_f32(
+; CHECK:    [[TMP5:%.*]] = call <4 x float> @__svml_exp2f4(<4 x float> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call float @exp2f(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %iv
+  store float %call, float* %arrayidx, align 4
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body
 
+for.end:
+  ret void
+}
+
+define void @exp2_f64_intrinsic(double* nocapture %varray) {
+; CHECK-LABEL: @exp2_f64_intrinsic(
+; CHECK:    [[TMP5:%.*]] = call <4 x double> @__svml_exp24(<4 x double> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call double @llvm.exp2.f64(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %iv
+  store double %call, double* %arrayidx, align 4
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:
+  ret void
+}
+
+define void @exp2_f32_intrinsic(float* nocapture %varray) {
+; CHECK-LABEL: @exp2_f32_intrinsic(
+; CHECK:    [[TMP5:%.*]] = call <4 x float> @__svml_exp2f4(<4 x float> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call float @llvm.exp2.f32(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %iv
+  store float %call, float* %arrayidx, align 4
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:
+  ret void
+}
+
+define void @exp2f_finite(float* nocapture %varray) {
+; CHECK-LABEL: @exp2f_finite(
+; CHECK:    [[TMP5:%.*]] = call <4 x float> @__svml_exp2f4(<4 x float> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call float @__exp2f_finite(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %iv
+  store float %call, float* %arrayidx, align 4
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:
+  ret void
+}
+
+define void @exp2_finite(double* nocapture %varray) {
+; CHECK-LABEL: @exp2_finite(
+; CHECK:    [[TMP5:%.*]] = call <4 x double> @__svml_exp24(<4 x double> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call double @__exp2_finite(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %iv
+  store double %call, double* %arrayidx, align 4
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:
+  ret void
+}
+
+attributes #0 = { nounwind readnone }
-- 
2.7.4