summaryrefslogtreecommitdiff
path: root/test/Analysis/CostModel
diff options
context:
space:
mode:
Diffstat (limited to 'test/Analysis/CostModel')
-rw-r--r--test/Analysis/CostModel/X86/ctlz.ll156
-rw-r--r--test/Analysis/CostModel/X86/ctpop.ll63
-rw-r--r--test/Analysis/CostModel/X86/cttz.ll131
3 files changed, 338 insertions, 12 deletions
diff --git a/test/Analysis/CostModel/X86/ctlz.ll b/test/Analysis/CostModel/X86/ctlz.ll
index 2c97da15aee5..769d73915e36 100644
--- a/test/Analysis/CostModel/X86/ctlz.ll
+++ b/test/Analysis/CostModel/X86/ctlz.ll
@@ -1,9 +1,12 @@
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=pentium4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE2 -check-prefix=NOPOPCNT
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE42 -check-prefix=POPCNT
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7-avx -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1 -check-prefix=POPCNT
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=core-avx2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2 -check-prefix=POPCNT
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1 -check-prefix=POPCNT
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2 -check-prefix=POPCNT
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=pentium4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7-avx -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=core-avx2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=knl -mattr=-avx512cd -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX512 -check-prefix=AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=skx -mattr=-avx512cd -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX512 -check-prefix=AVX512BW
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=skx -mattr=+avx512cd -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX512CD
; Verify the cost of scalar leading zero count instructions.
@@ -80,11 +83,18 @@ declare <8 x i32> @llvm.ctlz.v8i32(<8 x i32>, i1)
declare <16 x i16> @llvm.ctlz.v16i16(<16 x i16>, i1)
declare <32 x i8> @llvm.ctlz.v32i8(<32 x i8>, i1)
+declare <8 x i64> @llvm.ctlz.v8i64(<8 x i64>, i1)
+declare <16 x i32> @llvm.ctlz.v16i32(<16 x i32>, i1)
+declare <32 x i16> @llvm.ctlz.v32i16(<32 x i16>, i1)
+declare <64 x i8> @llvm.ctlz.v64i8(<64 x i8>, i1)
+
define <2 x i64> @var_ctlz_v2i64(<2 x i64> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v2i64':
; SSE2: Found an estimated cost of 25 for instruction: %ctlz
; SSE42: Found an estimated cost of 23 for instruction: %ctlz
; AVX: Found an estimated cost of 23 for instruction: %ctlz
+; AVX512: Found an estimated cost of 23 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
%ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 0)
ret <2 x i64> %ctlz
}
@@ -94,6 +104,8 @@ define <2 x i64> @var_ctlz_v2i64u(<2 x i64> %a) {
; SSE2: Found an estimated cost of 25 for instruction: %ctlz
; SSE42: Found an estimated cost of 23 for instruction: %ctlz
; AVX: Found an estimated cost of 23 for instruction: %ctlz
+; AVX512: Found an estimated cost of 23 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
%ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 1)
ret <2 x i64> %ctlz
}
@@ -104,6 +116,8 @@ define <4 x i64> @var_ctlz_v4i64(<4 x i64> %a) {
; SSE42: Found an estimated cost of 46 for instruction: %ctlz
; AVX1: Found an estimated cost of 48 for instruction: %ctlz
; AVX2: Found an estimated cost of 23 for instruction: %ctlz
+; AVX512: Found an estimated cost of 23 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
%ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 0)
ret <4 x i64> %ctlz
}
@@ -114,15 +128,45 @@ define <4 x i64> @var_ctlz_v4i64u(<4 x i64> %a) {
; SSE42: Found an estimated cost of 46 for instruction: %ctlz
; AVX1: Found an estimated cost of 48 for instruction: %ctlz
; AVX2: Found an estimated cost of 23 for instruction: %ctlz
+; AVX512: Found an estimated cost of 23 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
%ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 1)
ret <4 x i64> %ctlz
}
+define <8 x i64> @var_ctlz_v8i64(<8 x i64> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v8i64':
+; SSE2: Found an estimated cost of 100 for instruction: %ctlz
+; SSE42: Found an estimated cost of 92 for instruction: %ctlz
+; AVX1: Found an estimated cost of 96 for instruction: %ctlz
+; AVX2: Found an estimated cost of 46 for instruction: %ctlz
+; AVX512F: Found an estimated cost of 29 for instruction: %ctlz
+; AVX512BW: Found an estimated cost of 23 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
+ %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 0)
+ ret <8 x i64> %ctlz
+}
+
+define <8 x i64> @var_ctlz_v8i64u(<8 x i64> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v8i64u':
+; SSE2: Found an estimated cost of 100 for instruction: %ctlz
+; SSE42: Found an estimated cost of 92 for instruction: %ctlz
+; AVX1: Found an estimated cost of 96 for instruction: %ctlz
+; AVX2: Found an estimated cost of 46 for instruction: %ctlz
+; AVX512F: Found an estimated cost of 29 for instruction: %ctlz
+; AVX512BW: Found an estimated cost of 23 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
+ %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 1)
+ ret <8 x i64> %ctlz
+}
+
define <4 x i32> @var_ctlz_v4i32(<4 x i32> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v4i32':
; SSE2: Found an estimated cost of 26 for instruction: %ctlz
; SSE42: Found an estimated cost of 18 for instruction: %ctlz
; AVX: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
%ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 0)
ret <4 x i32> %ctlz
}
@@ -132,6 +176,8 @@ define <4 x i32> @var_ctlz_v4i32u(<4 x i32> %a) {
; SSE2: Found an estimated cost of 26 for instruction: %ctlz
; SSE42: Found an estimated cost of 18 for instruction: %ctlz
; AVX: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
%ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 1)
ret <4 x i32> %ctlz
}
@@ -142,6 +188,8 @@ define <8 x i32> @var_ctlz_v8i32(<8 x i32> %a) {
; SSE42: Found an estimated cost of 36 for instruction: %ctlz
; AVX1: Found an estimated cost of 38 for instruction: %ctlz
; AVX2: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
%ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 0)
ret <8 x i32> %ctlz
}
@@ -152,15 +200,45 @@ define <8 x i32> @var_ctlz_v8i32u(<8 x i32> %a) {
; SSE42: Found an estimated cost of 36 for instruction: %ctlz
; AVX1: Found an estimated cost of 38 for instruction: %ctlz
; AVX2: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
%ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 1)
ret <8 x i32> %ctlz
}
+define <16 x i32> @var_ctlz_v16i32(<16 x i32> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v16i32':
+; SSE2: Found an estimated cost of 104 for instruction: %ctlz
+; SSE42: Found an estimated cost of 72 for instruction: %ctlz
+; AVX1: Found an estimated cost of 76 for instruction: %ctlz
+; AVX2: Found an estimated cost of 36 for instruction: %ctlz
+; AVX512F: Found an estimated cost of 35 for instruction: %ctlz
+; AVX512BW: Found an estimated cost of 22 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
+ %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 0)
+ ret <16 x i32> %ctlz
+}
+
+define <16 x i32> @var_ctlz_v16i32u(<16 x i32> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v16i32u':
+; SSE2: Found an estimated cost of 104 for instruction: %ctlz
+; SSE42: Found an estimated cost of 72 for instruction: %ctlz
+; AVX1: Found an estimated cost of 76 for instruction: %ctlz
+; AVX2: Found an estimated cost of 36 for instruction: %ctlz
+; AVX512F: Found an estimated cost of 35 for instruction: %ctlz
+; AVX512BW: Found an estimated cost of 22 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 1 for instruction: %ctlz
+ %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 1)
+ ret <16 x i32> %ctlz
+}
+
define <8 x i16> @var_ctlz_v8i16(<8 x i16> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v8i16':
; SSE2: Found an estimated cost of 20 for instruction: %ctlz
; SSE42: Found an estimated cost of 14 for instruction: %ctlz
; AVX: Found an estimated cost of 14 for instruction: %ctlz
+; AVX512: Found an estimated cost of 14 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 4 for instruction: %ctlz
%ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 0)
ret <8 x i16> %ctlz
}
@@ -170,6 +248,8 @@ define <8 x i16> @var_ctlz_v8i16u(<8 x i16> %a) {
; SSE2: Found an estimated cost of 20 for instruction: %ctlz
; SSE42: Found an estimated cost of 14 for instruction: %ctlz
; AVX: Found an estimated cost of 14 for instruction: %ctlz
+; AVX512: Found an estimated cost of 14 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 4 for instruction: %ctlz
%ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 1)
ret <8 x i16> %ctlz
}
@@ -180,6 +260,8 @@ define <16 x i16> @var_ctlz_v16i16(<16 x i16> %a) {
; SSE42: Found an estimated cost of 28 for instruction: %ctlz
; AVX1: Found an estimated cost of 30 for instruction: %ctlz
; AVX2: Found an estimated cost of 14 for instruction: %ctlz
+; AVX512: Found an estimated cost of 14 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 4 for instruction: %ctlz
%ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 0)
ret <16 x i16> %ctlz
}
@@ -190,15 +272,45 @@ define <16 x i16> @var_ctlz_v16i16u(<16 x i16> %a) {
; SSE42: Found an estimated cost of 28 for instruction: %ctlz
; AVX1: Found an estimated cost of 30 for instruction: %ctlz
; AVX2: Found an estimated cost of 14 for instruction: %ctlz
+; AVX512: Found an estimated cost of 14 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 4 for instruction: %ctlz
%ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 1)
ret <16 x i16> %ctlz
}
+define <32 x i16> @var_ctlz_v32i16(<32 x i16> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v32i16':
+; SSE2: Found an estimated cost of 80 for instruction: %ctlz
+; SSE42: Found an estimated cost of 56 for instruction: %ctlz
+; AVX1: Found an estimated cost of 60 for instruction: %ctlz
+; AVX2: Found an estimated cost of 28 for instruction: %ctlz
+; AVX512F: Found an estimated cost of 28 for instruction: %ctlz
+; AVX512BW: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 8 for instruction: %ctlz
+ %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 0)
+ ret <32 x i16> %ctlz
+}
+
+define <32 x i16> @var_ctlz_v32i16u(<32 x i16> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v32i16u':
+; SSE2: Found an estimated cost of 80 for instruction: %ctlz
+; SSE42: Found an estimated cost of 56 for instruction: %ctlz
+; AVX1: Found an estimated cost of 60 for instruction: %ctlz
+; AVX2: Found an estimated cost of 28 for instruction: %ctlz
+; AVX512F: Found an estimated cost of 28 for instruction: %ctlz
+; AVX512BW: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 8 for instruction: %ctlz
+ %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 1)
+ ret <32 x i16> %ctlz
+}
+
define <16 x i8> @var_ctlz_v16i8(<16 x i8> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v16i8':
; SSE2: Found an estimated cost of 17 for instruction: %ctlz
; SSE42: Found an estimated cost of 9 for instruction: %ctlz
; AVX: Found an estimated cost of 9 for instruction: %ctlz
+; AVX512: Found an estimated cost of 9 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 4 for instruction: %ctlz
%ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 0)
ret <16 x i8> %ctlz
}
@@ -208,6 +320,8 @@ define <16 x i8> @var_ctlz_v16i8u(<16 x i8> %a) {
; SSE2: Found an estimated cost of 17 for instruction: %ctlz
; SSE42: Found an estimated cost of 9 for instruction: %ctlz
; AVX: Found an estimated cost of 9 for instruction: %ctlz
+; AVX512: Found an estimated cost of 9 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 4 for instruction: %ctlz
%ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 1)
ret <16 x i8> %ctlz
}
@@ -218,6 +332,8 @@ define <32 x i8> @var_ctlz_v32i8(<32 x i8> %a) {
; SSE42: Found an estimated cost of 18 for instruction: %ctlz
; AVX1: Found an estimated cost of 20 for instruction: %ctlz
; AVX2: Found an estimated cost of 9 for instruction: %ctlz
+; AVX512: Found an estimated cost of 9 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 10 for instruction: %ctlz
%ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 0)
ret <32 x i8> %ctlz
}
@@ -228,6 +344,34 @@ define <32 x i8> @var_ctlz_v32i8u(<32 x i8> %a) {
; SSE42: Found an estimated cost of 18 for instruction: %ctlz
; AVX1: Found an estimated cost of 20 for instruction: %ctlz
; AVX2: Found an estimated cost of 9 for instruction: %ctlz
+; AVX512: Found an estimated cost of 9 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 10 for instruction: %ctlz
%ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 1)
ret <32 x i8> %ctlz
}
+
+define <64 x i8> @var_ctlz_v64i8(<64 x i8> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v64i8':
+; SSE2: Found an estimated cost of 68 for instruction: %ctlz
+; SSE42: Found an estimated cost of 36 for instruction: %ctlz
+; AVX1: Found an estimated cost of 40 for instruction: %ctlz
+; AVX2: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512F: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512BW: Found an estimated cost of 17 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 20 for instruction: %ctlz
+ %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 0)
+ ret <64 x i8> %ctlz
+}
+
+define <64 x i8> @var_ctlz_v64i8u(<64 x i8> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v64i8u':
+; SSE2: Found an estimated cost of 68 for instruction: %ctlz
+; SSE42: Found an estimated cost of 36 for instruction: %ctlz
+; AVX1: Found an estimated cost of 40 for instruction: %ctlz
+; AVX2: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512F: Found an estimated cost of 18 for instruction: %ctlz
+; AVX512BW: Found an estimated cost of 17 for instruction: %ctlz
+; AVX512CD: Found an estimated cost of 20 for instruction: %ctlz
+ %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 1)
+ ret <64 x i8> %ctlz
+}
diff --git a/test/Analysis/CostModel/X86/ctpop.ll b/test/Analysis/CostModel/X86/ctpop.ll
index f072cbaec492..e6a14e98e37a 100644
--- a/test/Analysis/CostModel/X86/ctpop.ll
+++ b/test/Analysis/CostModel/X86/ctpop.ll
@@ -4,6 +4,8 @@
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=core-avx2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2 -check-prefix=POPCNT
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1 -check-prefix=POPCNT
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2 -check-prefix=POPCNT
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=knl -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX512 -check-prefix=AVX512F -check-prefix=POPCNT
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=skx -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX512 -check-prefix=AVX512BW -check-prefix=POPCNT
; Verify the cost of scalar population count instructions.
@@ -56,11 +58,17 @@ declare <8 x i32> @llvm.ctpop.v8i32(<8 x i32>)
declare <16 x i16> @llvm.ctpop.v16i16(<16 x i16>)
declare <32 x i8> @llvm.ctpop.v32i8(<32 x i8>)
+declare <8 x i64> @llvm.ctpop.v8i64(<8 x i64>)
+declare <16 x i32> @llvm.ctpop.v16i32(<16 x i32>)
+declare <32 x i16> @llvm.ctpop.v32i16(<32 x i16>)
+declare <64 x i8> @llvm.ctpop.v64i8(<64 x i8>)
+
define <2 x i64> @var_ctpop_v2i64(<2 x i64> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_ctpop_v2i64':
; SSE2: Found an estimated cost of 12 for instruction: %ctpop
; SSE42: Found an estimated cost of 7 for instruction: %ctpop
; AVX: Found an estimated cost of 7 for instruction: %ctpop
+; AVX512: Found an estimated cost of 7 for instruction: %ctpop
%ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
ret <2 x i64> %ctpop
}
@@ -71,15 +79,29 @@ define <4 x i64> @var_ctpop_v4i64(<4 x i64> %a) {
; SSE42: Found an estimated cost of 14 for instruction: %ctpop
; AVX1: Found an estimated cost of 16 for instruction: %ctpop
; AVX2: Found an estimated cost of 7 for instruction: %ctpop
+; AVX512: Found an estimated cost of 7 for instruction: %ctpop
%ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %a)
ret <4 x i64> %ctpop
}
+define <8 x i64> @var_ctpop_v8i64(<8 x i64> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctpop_v8i64':
+; SSE2: Found an estimated cost of 48 for instruction: %ctpop
+; SSE42: Found an estimated cost of 28 for instruction: %ctpop
+; AVX1: Found an estimated cost of 32 for instruction: %ctpop
+; AVX2: Found an estimated cost of 14 for instruction: %ctpop
+; AVX512F: Found an estimated cost of 16 for instruction: %ctpop
+; AVX512BW: Found an estimated cost of 7 for instruction: %ctpop
+ %ctpop = call <8 x i64> @llvm.ctpop.v8i64(<8 x i64> %a)
+ ret <8 x i64> %ctpop
+}
+
define <4 x i32> @var_ctpop_v4i32(<4 x i32> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_ctpop_v4i32':
; SSE2: Found an estimated cost of 15 for instruction: %ctpop
; SSE42: Found an estimated cost of 11 for instruction: %ctpop
; AVX: Found an estimated cost of 11 for instruction: %ctpop
+; AVX512: Found an estimated cost of 11 for instruction: %ctpop
%ctpop = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %a)
ret <4 x i32> %ctpop
}
@@ -90,15 +112,29 @@ define <8 x i32> @var_ctpop_v8i32(<8 x i32> %a) {
; SSE42: Found an estimated cost of 22 for instruction: %ctpop
; AVX1: Found an estimated cost of 24 for instruction: %ctpop
; AVX2: Found an estimated cost of 11 for instruction: %ctpop
+; AVX512: Found an estimated cost of 11 for instruction: %ctpop
%ctpop = call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> %a)
ret <8 x i32> %ctpop
}
+define <16 x i32> @var_ctpop_v16i32(<16 x i32> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctpop_v16i32':
+; SSE2: Found an estimated cost of 60 for instruction: %ctpop
+; SSE42: Found an estimated cost of 44 for instruction: %ctpop
+; AVX1: Found an estimated cost of 48 for instruction: %ctpop
+; AVX2: Found an estimated cost of 22 for instruction: %ctpop
+; AVX512F: Found an estimated cost of 24 for instruction: %ctpop
+; AVX512BW: Found an estimated cost of 11 for instruction: %ctpop
+ %ctpop = call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %a)
+ ret <16 x i32> %ctpop
+}
+
define <8 x i16> @var_ctpop_v8i16(<8 x i16> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_ctpop_v8i16':
; SSE2: Found an estimated cost of 13 for instruction: %ctpop
; SSE42: Found an estimated cost of 9 for instruction: %ctpop
; AVX: Found an estimated cost of 9 for instruction: %ctpop
+; AVX512: Found an estimated cost of 9 for instruction: %ctpop
%ctpop = call <8 x i16> @llvm.ctpop.v8i16(<8 x i16> %a)
ret <8 x i16> %ctpop
}
@@ -109,15 +145,29 @@ define <16 x i16> @var_ctpop_v16i16(<16 x i16> %a) {
; SSE42: Found an estimated cost of 18 for instruction: %ctpop
; AVX1: Found an estimated cost of 20 for instruction: %ctpop
; AVX2: Found an estimated cost of 9 for instruction: %ctpop
+; AVX512: Found an estimated cost of 9 for instruction: %ctpop
%ctpop = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> %a)
ret <16 x i16> %ctpop
}
+define <32 x i16> @var_ctpop_v32i16(<32 x i16> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctpop_v32i16':
+; SSE2: Found an estimated cost of 52 for instruction: %ctpop
+; SSE42: Found an estimated cost of 36 for instruction: %ctpop
+; AVX1: Found an estimated cost of 40 for instruction: %ctpop
+; AVX2: Found an estimated cost of 18 for instruction: %ctpop
+; AVX512F: Found an estimated cost of 18 for instruction: %ctpop
+; AVX512BW: Found an estimated cost of 9 for instruction: %ctpop
+ %ctpop = call <32 x i16> @llvm.ctpop.v32i16(<32 x i16> %a)
+ ret <32 x i16> %ctpop
+}
+
define <16 x i8> @var_ctpop_v16i8(<16 x i8> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_ctpop_v16i8':
; SSE2: Found an estimated cost of 10 for instruction: %ctpop
; SSE42: Found an estimated cost of 6 for instruction: %ctpop
; AVX: Found an estimated cost of 6 for instruction: %ctpop
+; AVX512: Found an estimated cost of 6 for instruction: %ctpop
%ctpop = call <16 x i8> @llvm.ctpop.v16i8(<16 x i8> %a)
ret <16 x i8> %ctpop
}
@@ -128,6 +178,19 @@ define <32 x i8> @var_ctpop_v32i8(<32 x i8> %a) {
; SSE42: Found an estimated cost of 12 for instruction: %ctpop
; AVX1: Found an estimated cost of 14 for instruction: %ctpop
; AVX2: Found an estimated cost of 6 for instruction: %ctpop
+; AVX512: Found an estimated cost of 6 for instruction: %ctpop
%ctpop = call <32 x i8> @llvm.ctpop.v32i8(<32 x i8> %a)
ret <32 x i8> %ctpop
}
+
+define <64 x i8> @var_ctpop_v64i8(<64 x i8> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_ctpop_v64i8':
+; SSE2: Found an estimated cost of 40 for instruction: %ctpop
+; SSE42: Found an estimated cost of 24 for instruction: %ctpop
+; AVX1: Found an estimated cost of 28 for instruction: %ctpop
+; AVX2: Found an estimated cost of 12 for instruction: %ctpop
+; AVX512F: Found an estimated cost of 12 for instruction: %ctpop
+; AVX512BW: Found an estimated cost of 6 for instruction: %ctpop
+ %ctpop = call <64 x i8> @llvm.ctpop.v64i8(<64 x i8> %a)
+ ret <64 x i8> %ctpop
+}
diff --git a/test/Analysis/CostModel/X86/cttz.ll b/test/Analysis/CostModel/X86/cttz.ll
index 5d3c59b60232..e7a39781385e 100644
--- a/test/Analysis/CostModel/X86/cttz.ll
+++ b/test/Analysis/CostModel/X86/cttz.ll
@@ -1,9 +1,11 @@
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=pentium4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE2 -check-prefix=NOPOPCNT
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE42 -check-prefix=POPCNT
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7-avx -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1 -check-prefix=POPCNT
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=core-avx2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2 -check-prefix=POPCNT
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1 -check-prefix=POPCNT
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2 -check-prefix=POPCNT
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=pentium4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7-avx -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=core-avx2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver4 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=knl -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX512 -check-prefix=AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=skx -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX512 -check-prefix=AVX512BW
; Verify the cost of scalar trailing zero count instructions.
@@ -80,11 +82,17 @@ declare <8 x i32> @llvm.cttz.v8i32(<8 x i32>, i1)
declare <16 x i16> @llvm.cttz.v16i16(<16 x i16>, i1)
declare <32 x i8> @llvm.cttz.v32i8(<32 x i8>, i1)
+declare <8 x i64> @llvm.cttz.v8i64(<8 x i64>, i1)
+declare <16 x i32> @llvm.cttz.v16i32(<16 x i32>, i1)
+declare <32 x i16> @llvm.cttz.v32i16(<32 x i16>, i1)
+declare <64 x i8> @llvm.cttz.v64i8(<64 x i8>, i1)
+
define <2 x i64> @var_cttz_v2i64(<2 x i64> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_cttz_v2i64':
; SSE2: Found an estimated cost of 14 for instruction: %cttz
; SSE42: Found an estimated cost of 10 for instruction: %cttz
; AVX: Found an estimated cost of 10 for instruction: %cttz
+; AVX512: Found an estimated cost of 10 for instruction: %cttz
%cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 0)
ret <2 x i64> %cttz
}
@@ -94,6 +102,7 @@ define <2 x i64> @var_cttz_v2i64u(<2 x i64> %a) {
; SSE2: Found an estimated cost of 14 for instruction: %cttz
; SSE42: Found an estimated cost of 10 for instruction: %cttz
; AVX: Found an estimated cost of 10 for instruction: %cttz
+; AVX512: Found an estimated cost of 10 for instruction: %cttz
%cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 1)
ret <2 x i64> %cttz
}
@@ -104,6 +113,7 @@ define <4 x i64> @var_cttz_v4i64(<4 x i64> %a) {
; SSE42: Found an estimated cost of 20 for instruction: %cttz
; AVX1: Found an estimated cost of 22 for instruction: %cttz
; AVX2: Found an estimated cost of 10 for instruction: %cttz
+; AVX512: Found an estimated cost of 10 for instruction: %cttz
%cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 0)
ret <4 x i64> %cttz
}
@@ -114,15 +124,41 @@ define <4 x i64> @var_cttz_v4i64u(<4 x i64> %a) {
; SSE42: Found an estimated cost of 20 for instruction: %cttz
; AVX1: Found an estimated cost of 22 for instruction: %cttz
; AVX2: Found an estimated cost of 10 for instruction: %cttz
+; AVX512: Found an estimated cost of 10 for instruction: %cttz
%cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 1)
ret <4 x i64> %cttz
}
+define <8 x i64> @var_cttz_v8i64(<8 x i64> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_cttz_v8i64':
+; SSE2: Found an estimated cost of 56 for instruction: %cttz
+; SSE42: Found an estimated cost of 40 for instruction: %cttz
+; AVX1: Found an estimated cost of 44 for instruction: %cttz
+; AVX2: Found an estimated cost of 20 for instruction: %cttz
+; AVX512F: Found an estimated cost of 20 for instruction: %cttz
+; AVX512BW: Found an estimated cost of 10 for instruction: %cttz
+ %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 0)
+ ret <8 x i64> %cttz
+}
+
+define <8 x i64> @var_cttz_v8i64u(<8 x i64> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_cttz_v8i64u':
+; SSE2: Found an estimated cost of 56 for instruction: %cttz
+; SSE42: Found an estimated cost of 40 for instruction: %cttz
+; AVX1: Found an estimated cost of 44 for instruction: %cttz
+; AVX2: Found an estimated cost of 20 for instruction: %cttz
+; AVX512F: Found an estimated cost of 20 for instruction: %cttz
+; AVX512BW: Found an estimated cost of 10 for instruction: %cttz
+ %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 1)
+ ret <8 x i64> %cttz
+}
+
define <4 x i32> @var_cttz_v4i32(<4 x i32> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_cttz_v4i32':
; SSE2: Found an estimated cost of 18 for instruction: %cttz
; SSE42: Found an estimated cost of 14 for instruction: %cttz
; AVX: Found an estimated cost of 14 for instruction: %cttz
+; AVX512: Found an estimated cost of 14 for instruction: %cttz
%cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 0)
ret <4 x i32> %cttz
}
@@ -132,6 +168,7 @@ define <4 x i32> @var_cttz_v4i32u(<4 x i32> %a) {
; SSE2: Found an estimated cost of 18 for instruction: %cttz
; SSE42: Found an estimated cost of 14 for instruction: %cttz
; AVX: Found an estimated cost of 14 for instruction: %cttz
+; AVX512: Found an estimated cost of 14 for instruction: %cttz
%cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 1)
ret <4 x i32> %cttz
}
@@ -142,6 +179,7 @@ define <8 x i32> @var_cttz_v8i32(<8 x i32> %a) {
; SSE42: Found an estimated cost of 28 for instruction: %cttz
; AVX1: Found an estimated cost of 30 for instruction: %cttz
; AVX2: Found an estimated cost of 14 for instruction: %cttz
+; AVX512: Found an estimated cost of 14 for instruction: %cttz
%cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 0)
ret <8 x i32> %cttz
}
@@ -152,15 +190,41 @@ define <8 x i32> @var_cttz_v8i32u(<8 x i32> %a) {
; SSE42: Found an estimated cost of 28 for instruction: %cttz
; AVX1: Found an estimated cost of 30 for instruction: %cttz
; AVX2: Found an estimated cost of 14 for instruction: %cttz
+; AVX512: Found an estimated cost of 14 for instruction: %cttz
%cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 1)
ret <8 x i32> %cttz
}
+define <16 x i32> @var_cttz_v16i32(<16 x i32> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_cttz_v16i32':
+; SSE2: Found an estimated cost of 72 for instruction: %cttz
+; SSE42: Found an estimated cost of 56 for instruction: %cttz
+; AVX1: Found an estimated cost of 60 for instruction: %cttz
+; AVX2: Found an estimated cost of 28 for instruction: %cttz
+; AVX512F: Found an estimated cost of 28 for instruction: %cttz
+; AVX512BW: Found an estimated cost of 14 for instruction: %cttz
+ %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 0)
+ ret <16 x i32> %cttz
+}
+
+define <16 x i32> @var_cttz_v16i32u(<16 x i32> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_cttz_v16i32u':
+; SSE2: Found an estimated cost of 72 for instruction: %cttz
+; SSE42: Found an estimated cost of 56 for instruction: %cttz
+; AVX1: Found an estimated cost of 60 for instruction: %cttz
+; AVX2: Found an estimated cost of 28 for instruction: %cttz
+; AVX512F: Found an estimated cost of 28 for instruction: %cttz
+; AVX512BW: Found an estimated cost of 14 for instruction: %cttz
+ %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 1)
+ ret <16 x i32> %cttz
+}
+
define <8 x i16> @var_cttz_v8i16(<8 x i16> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_cttz_v8i16':
; SSE2: Found an estimated cost of 16 for instruction: %cttz
; SSE42: Found an estimated cost of 12 for instruction: %cttz
; AVX: Found an estimated cost of 12 for instruction: %cttz
+; AVX512: Found an estimated cost of 12 for instruction: %cttz
%cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 0)
ret <8 x i16> %cttz
}
@@ -170,6 +234,7 @@ define <8 x i16> @var_cttz_v8i16u(<8 x i16> %a) {
; SSE2: Found an estimated cost of 16 for instruction: %cttz
; SSE42: Found an estimated cost of 12 for instruction: %cttz
; AVX: Found an estimated cost of 12 for instruction: %cttz
+; AVX512: Found an estimated cost of 12 for instruction: %cttz
%cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 1)
ret <8 x i16> %cttz
}
@@ -180,6 +245,7 @@ define <16 x i16> @var_cttz_v16i16(<16 x i16> %a) {
; SSE42: Found an estimated cost of 24 for instruction: %cttz
; AVX1: Found an estimated cost of 26 for instruction: %cttz
; AVX2: Found an estimated cost of 12 for instruction: %cttz
+; AVX512: Found an estimated cost of 12 for instruction: %cttz
%cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 0)
ret <16 x i16> %cttz
}
@@ -190,15 +256,41 @@ define <16 x i16> @var_cttz_v16i16u(<16 x i16> %a) {
; SSE42: Found an estimated cost of 24 for instruction: %cttz
; AVX1: Found an estimated cost of 26 for instruction: %cttz
; AVX2: Found an estimated cost of 12 for instruction: %cttz
+; AVX512: Found an estimated cost of 12 for instruction: %cttz
%cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 1)
ret <16 x i16> %cttz
}
+define <32 x i16> @var_cttz_v32i16(<32 x i16> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_cttz_v32i16':
+; SSE2: Found an estimated cost of 64 for instruction: %cttz
+; SSE42: Found an estimated cost of 48 for instruction: %cttz
+; AVX1: Found an estimated cost of 52 for instruction: %cttz
+; AVX2: Found an estimated cost of 24 for instruction: %cttz
+; AVX512F: Found an estimated cost of 24 for instruction: %cttz
+; AVX512BW: Found an estimated cost of 12 for instruction: %cttz
+ %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 0)
+ ret <32 x i16> %cttz
+}
+
+define <32 x i16> @var_cttz_v32i16u(<32 x i16> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_cttz_v32i16u':
+; SSE2: Found an estimated cost of 64 for instruction: %cttz
+; SSE42: Found an estimated cost of 48 for instruction: %cttz
+; AVX1: Found an estimated cost of 52 for instruction: %cttz
+; AVX2: Found an estimated cost of 24 for instruction: %cttz
+; AVX512F: Found an estimated cost of 24 for instruction: %cttz
+; AVX512BW: Found an estimated cost of 12 for instruction: %cttz
+ %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 1)
+ ret <32 x i16> %cttz
+}
+
define <16 x i8> @var_cttz_v16i8(<16 x i8> %a) {
; CHECK: 'Cost Model Analysis' for function 'var_cttz_v16i8':
; SSE2: Found an estimated cost of 13 for instruction: %cttz
; SSE42: Found an estimated cost of 9 for instruction: %cttz
; AVX: Found an estimated cost of 9 for instruction: %cttz
+; AVX512: Found an estimated cost of 9 for instruction: %cttz
%cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 0)
ret <16 x i8> %cttz
}
@@ -208,6 +300,7 @@ define <16 x i8> @var_cttz_v16i8u(<16 x i8> %a) {
; SSE2: Found an estimated cost of 13 for instruction: %cttz
; SSE42: Found an estimated cost of 9 for instruction: %cttz
; AVX: Found an estimated cost of 9 for instruction: %cttz
+; AVX512: Found an estimated cost of 9 for instruction: %cttz
%cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 1)
ret <16 x i8> %cttz
}
@@ -218,6 +311,7 @@ define <32 x i8> @var_cttz_v32i8(<32 x i8> %a) {
; SSE42: Found an estimated cost of 18 for instruction: %cttz
; AVX1: Found an estimated cost of 20 for instruction: %cttz
; AVX2: Found an estimated cost of 9 for instruction: %cttz
+; AVX512: Found an estimated cost of 9 for instruction: %cttz
%cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 0)
ret <32 x i8> %cttz
}
@@ -228,6 +322,31 @@ define <32 x i8> @var_cttz_v32i8u(<32 x i8> %a) {
; SSE42: Found an estimated cost of 18 for instruction: %cttz
; AVX1: Found an estimated cost of 20 for instruction: %cttz
; AVX2: Found an estimated cost of 9 for instruction: %cttz
+; AVX512: Found an estimated cost of 9 for instruction: %cttz
%cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 1)
ret <32 x i8> %cttz
}
+
+define <64 x i8> @var_cttz_v64i8(<64 x i8> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_cttz_v64i8':
+; SSE2: Found an estimated cost of 52 for instruction: %cttz
+; SSE42: Found an estimated cost of 36 for instruction: %cttz
+; AVX1: Found an estimated cost of 40 for instruction: %cttz
+; AVX2: Found an estimated cost of 18 for instruction: %cttz
+; AVX512F: Found an estimated cost of 18 for instruction: %cttz
+; AVX512BW: Found an estimated cost of 9 for instruction: %cttz
+ %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 0)
+ ret <64 x i8> %cttz
+}
+
+define <64 x i8> @var_cttz_v64i8u(<64 x i8> %a) {
+; CHECK: 'Cost Model Analysis' for function 'var_cttz_v64i8u':
+; SSE2: Found an estimated cost of 52 for instruction: %cttz
+; SSE42: Found an estimated cost of 36 for instruction: %cttz
+; AVX1: Found an estimated cost of 40 for instruction: %cttz
+; AVX2: Found an estimated cost of 18 for instruction: %cttz
+; AVX512F: Found an estimated cost of 18 for instruction: %cttz
+; AVX512BW: Found an estimated cost of 9 for instruction: %cttz
+ %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 1)
+ ret <64 x i8> %cttz
+}
generated by cgit v1.3 (git 2.53.0) at 2026-07-25 07:31:58 +0000