[TLI] Add mapping for various '__<func>_finite' forms of the math routines to SVML routines

[chrischr]

Summary

This is the third of a set of patches to followup on the thread in http://lists.llvm.org/pipermail/llvm-dev/2017-March/111323.html. (Builds on the changes within the patch @ https://reviews.llvm.org/D31787)

This extends the mappings used to map some of the math function calls to the Intel SVML library. This handles the case when the functions have been redirected by the -ffast-math flag to call the '__<func>_finite' alternative routine.

[andrew.w.kaylor]

lib/Analysis/TargetLibraryInfo.cpp

1377 ↗

(On Diff #94444)

Is there a reason not to group these with the corresponding non-finite-only versions?

[chrischr]

lib/Analysis/TargetLibraryInfo.cpp

1377 ↗

(On Diff #94444)

No, and looking back at how the others are ordered/grouped, I think it makes sense to move these to be with those. I'll update it.

[chrischr]

Updated patch to group the new SVML function mappings to be with the existing function groupings.

[andrew.w.kaylor]

lgtm

llvm/trunk/lib/Analysis/TargetLibraryInfo.cpp

const VecDesc VecFuncs[] = {
         {"pow", "__svml_pow2", 2},
         {"pow", "__svml_pow4", 4},
         {"pow", "__svml_pow8", 8},
 
         {"powf", "__svml_powf4", 4},
         {"powf", "__svml_powf8", 8},
         {"powf", "__svml_powf16", 16},
 
+        { "__pow_finite", "__svml_pow2", 2 },
+        { "__pow_finite", "__svml_pow4", 4 },
+        { "__pow_finite", "__svml_pow8", 8 },
+
+        { "__powf_finite", "__svml_powf4", 4 },
+        { "__powf_finite", "__svml_powf8", 8 },
+        { "__powf_finite", "__svml_powf16", 16 },
+
         {"llvm.pow.f64", "__svml_pow2", 2},
         {"llvm.pow.f64", "__svml_pow4", 4},
         {"llvm.pow.f64", "__svml_pow8", 8},
 
         {"llvm.pow.f32", "__svml_powf4", 4},
         {"llvm.pow.f32", "__svml_powf8", 8},
         {"llvm.pow.f32", "__svml_powf16", 16},
 
         {"exp", "__svml_exp2", 2},
         {"exp", "__svml_exp4", 4},
         {"exp", "__svml_exp8", 8},
 
         {"expf", "__svml_expf4", 4},
         {"expf", "__svml_expf8", 8},
         {"expf", "__svml_expf16", 16},
 
+        { "__exp_finite", "__svml_exp2", 2 },
+        { "__exp_finite", "__svml_exp4", 4 },
+        { "__exp_finite", "__svml_exp8", 8 },
+
+        { "__expf_finite", "__svml_expf4", 4 },
+        { "__expf_finite", "__svml_expf8", 8 },
+        { "__expf_finite", "__svml_expf16", 16 },
+
         {"llvm.exp.f64", "__svml_exp2", 2},
         {"llvm.exp.f64", "__svml_exp4", 4},
         {"llvm.exp.f64", "__svml_exp8", 8},
 
         {"llvm.exp.f32", "__svml_expf4", 4},
         {"llvm.exp.f32", "__svml_expf8", 8},
         {"llvm.exp.f32", "__svml_expf16", 16},
 
         {"log", "__svml_log2", 2},
         {"log", "__svml_log4", 4},
         {"log", "__svml_log8", 8},
 
         {"logf", "__svml_logf4", 4},
         {"logf", "__svml_logf8", 8},
         {"logf", "__svml_logf16", 16},
 
+        { "__log_finite", "__svml_log2", 2 },
+        { "__log_finite", "__svml_log4", 4 },
+        { "__log_finite", "__svml_log8", 8 },
+
+        { "__logf_finite", "__svml_logf4", 4 },
+        { "__logf_finite", "__svml_logf8", 8 },
+        { "__logf_finite", "__svml_logf16", 16 },
+
         {"llvm.log.f64", "__svml_log2", 2},
         {"llvm.log.f64", "__svml_log4", 4},
         {"llvm.log.f64", "__svml_log8", 8},
 
         {"llvm.log.f32", "__svml_logf4", 4},
         {"llvm.log.f32", "__svml_logf8", 8},
         {"llvm.log.f32", "__svml_logf16", 16},
     };

llvm/trunk/test/Transforms/LoopVectorize/X86/svml-calls-finite.ll

+; RUN: opt -vector-library=SVML -loop-vectorize -S < %s | FileCheck %s
+
+; Test to verify that when math headers are built with
+; __FINITE_MATH_ONLY__ enabled, causing use of __<func>_finite
+; function versions, vectorization can map these to vector versions.
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+declare float @__expf_finite(float) #0
+
+; CHECK-LABEL: @exp_f32
+; CHECK: <4 x float> @__svml_expf4
+; CHECK: ret
+define void @exp_f32(float* nocapture %varray) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %tmp = trunc i64 %indvars.iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call fast float @__expf_finite(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %indvars.iv
+  store float %call, float* %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !1
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+!1 = distinct !{!1, !2, !3}
+!2 = !{!"llvm.loop.vectorize.width", i32 4}
+!3 = !{!"llvm.loop.vectorize.enable", i1 true}
+
+
+declare double @__exp_finite(double) #0
+
+; CHECK-LABEL: @exp_f64
+; CHECK: <4 x double> @__svml_exp4
+; CHECK: ret
+define void @exp_f64(double* nocapture %varray) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %tmp = trunc i64 %indvars.iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call fast double @__exp_finite(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %indvars.iv
+  store double %call, double* %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !11
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+!11 = distinct !{!11, !12, !13}
+!12 = !{!"llvm.loop.vectorize.width", i32 4}
+!13 = !{!"llvm.loop.vectorize.enable", i1 true}
+
+
+
+
+declare float @__logf_finite(float) #0
+
+; CHECK-LABEL: @log_f32
+; CHECK: <4 x float> @__svml_logf4
+; CHECK: ret
+define void @log_f32(float* nocapture %varray) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %tmp = trunc i64 %indvars.iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call fast float @__logf_finite(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %indvars.iv
+  store float %call, float* %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !21
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+!21 = distinct !{!21, !22, !23}
+!22 = !{!"llvm.loop.vectorize.width", i32 4}
+!23 = !{!"llvm.loop.vectorize.enable", i1 true}
+
+
+declare double @__log_finite(double) #0
+
+; CHECK-LABEL: @log_f64
+; CHECK: <4 x double> @__svml_log4
+; CHECK: ret
+define void @log_f64(double* nocapture %varray) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %tmp = trunc i64 %indvars.iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call fast double @__log_finite(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %indvars.iv
+  store double %call, double* %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !31
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+!31 = distinct !{!31, !32, !33}
+!32 = !{!"llvm.loop.vectorize.width", i32 4}
+!33 = !{!"llvm.loop.vectorize.enable", i1 true}
+
+
+declare float @__powf_finite(float, float) #0
+
+; CHECK-LABEL: @pow_f32
+; CHECK: <4 x float> @__svml_powf4
+; CHECK: ret
+define void @pow_f32(float* nocapture %varray, float* nocapture readonly %exp) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %tmp = trunc i64 %indvars.iv to i32
+  %conv = sitofp i32 %tmp to float
+  %arrayidx = getelementptr inbounds float, float* %exp, i64 %indvars.iv
+  %tmp1 = load float, float* %arrayidx, align 4
+  %tmp2 = tail call fast float @__powf_finite(float %conv, float %tmp1)
+  %arrayidx2 = getelementptr inbounds float, float* %varray, i64 %indvars.iv
+  store float %tmp2, float* %arrayidx2, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !41
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+!41 = distinct !{!41, !42, !43}
+!42 = !{!"llvm.loop.vectorize.width", i32 4}
+!43 = !{!"llvm.loop.vectorize.enable", i1 true}
+
+
+declare double @__pow_finite(double, double) #0
+
+; CHECK-LABEL: @pow_f64
+; CHECK: <4 x double> @__svml_pow4
+; CHECK: ret
+define void @pow_f64(double* nocapture %varray, double* nocapture readonly %exp) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %tmp = trunc i64 %indvars.iv to i32
+  %conv = sitofp i32 %tmp to double
+  %arrayidx = getelementptr inbounds double, double* %exp, i64 %indvars.iv
+  %tmp1 = load double, double* %arrayidx, align 4
+  %tmp2 = tail call fast double @__pow_finite(double %conv, double %tmp1)
+  %arrayidx2 = getelementptr inbounds double, double* %varray, i64 %indvars.iv
+  store double %tmp2, double* %arrayidx2, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !51
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+!51 = distinct !{!51, !52, !53}
+!52 = !{!"llvm.loop.vectorize.width", i32 4}
+!53 = !{!"llvm.loop.vectorize.enable", i1 true}