85 lines
3.6 KiB
LLVM
85 lines
3.6 KiB
LLVM
; RUN: opt -S -mtriple=x86_64-apple-darwin -mcpu=core2 -passes="print<cost-model>" 2>&1 -disable-output < %s | FileCheck %s -check-prefix=CORE2
|
|
; RUN: opt -S -mtriple=x86_64-apple-darwin -mcpu=corei7 -passes="print<cost-model>" 2>&1 -disable-output < %s | FileCheck %s -check-prefix=COREI7
|
|
|
|
; If SSE4.1 roundps instruction is available it is cheap to lower, otherwise
|
|
; it'll be scalarized into calls which are expensive.
|
|
define void @test1(ptr nocapture %f) nounwind {
|
|
vector.ph:
|
|
br label %vector.body
|
|
|
|
vector.body: ; preds = %vector.body, %vector.ph
|
|
%index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
|
|
%0 = getelementptr inbounds float, ptr %f, i64 %index
|
|
%wide.load = load <4 x float>, ptr %0, align 4
|
|
%1 = call <4 x float> @llvm.ceil.v4f32(<4 x float> %wide.load)
|
|
store <4 x float> %1, ptr %0, align 4
|
|
%index.next = add i64 %index, 4
|
|
%2 = icmp eq i64 %index.next, 1024
|
|
br i1 %2, label %for.end, label %vector.body
|
|
|
|
for.end: ; preds = %vector.body
|
|
ret void
|
|
|
|
; CORE2: function 'test1'
|
|
; CORE2: Cost Model: Found an estimated cost of 46 for instruction: %1 = call <4 x float> @llvm.ceil.v4f32(<4 x float> %wide.load)
|
|
|
|
; COREI7: function 'test1'
|
|
; COREI7: Cost Model: Found an estimated cost of 1 for instruction: %1 = call <4 x float> @llvm.ceil.v4f32(<4 x float> %wide.load)
|
|
|
|
}
|
|
|
|
declare <4 x float> @llvm.ceil.v4f32(<4 x float>) nounwind readnone
|
|
|
|
define void @test2(ptr nocapture %f) nounwind {
|
|
vector.ph:
|
|
br label %vector.body
|
|
|
|
vector.body: ; preds = %vector.body, %vector.ph
|
|
%index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
|
|
%0 = getelementptr inbounds float, ptr %f, i64 %index
|
|
%wide.load = load <4 x float>, ptr %0, align 4
|
|
%1 = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %wide.load)
|
|
store <4 x float> %1, ptr %0, align 4
|
|
%index.next = add i64 %index, 4
|
|
%2 = icmp eq i64 %index.next, 1024
|
|
br i1 %2, label %for.end, label %vector.body
|
|
|
|
for.end: ; preds = %vector.body
|
|
ret void
|
|
|
|
; CORE2: function 'test2'
|
|
; CORE2: Cost Model: Found an estimated cost of 46 for instruction: %1 = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %wide.load)
|
|
|
|
; COREI7: function 'test2'
|
|
; COREI7: Cost Model: Found an estimated cost of 1 for instruction: %1 = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %wide.load)
|
|
|
|
}
|
|
|
|
declare <4 x float> @llvm.nearbyint.v4f32(<4 x float>) nounwind readnone
|
|
|
|
define void @test3(ptr nocapture %f, <4 x float> %b, <4 x float> %c) nounwind {
|
|
vector.ph:
|
|
br label %vector.body
|
|
|
|
vector.body: ; preds = %vector.body, %vector.ph
|
|
%index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
|
|
%0 = getelementptr inbounds float, ptr %f, i64 %index
|
|
%wide.load = load <4 x float>, ptr %0, align 4
|
|
%1 = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %wide.load, <4 x float> %b, <4 x float> %c)
|
|
store <4 x float> %1, ptr %0, align 4
|
|
%index.next = add i64 %index, 4
|
|
%2 = icmp eq i64 %index.next, 1024
|
|
br i1 %2, label %for.end, label %vector.body
|
|
|
|
for.end: ; preds = %vector.body
|
|
ret void
|
|
|
|
; CORE2: function 'test3'
|
|
; CORE2: Cost Model: Found an estimated cost of 4 for instruction: %1 = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %wide.load, <4 x float> %b, <4 x float> %c)
|
|
|
|
; COREI7: function 'test3'
|
|
; COREI7: Cost Model: Found an estimated cost of 2 for instruction: %1 = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %wide.load, <4 x float> %b, <4 x float> %c)
|
|
|
|
}
|
|
|
|
declare <4 x float> @llvm.fmuladd.v4f32(<4 x float>, <4 x float>, <4 x float>) nounwind readnone
|