; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=-bmi,+sse2 | FileCheck %s -check-prefixes=CHECK,SSE2,NOBMI
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+bmi,+sse2 | FileCheck %s -check-prefixes=CHECK,SSE2,BMI
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=-bmi,+sse2 | FileCheck %s -check-prefixes=CHECK,NOBMI
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+bmi,+sse2 | FileCheck %s -check-prefixes=CHECK,BMI,SSE2
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+bmi,+sse4.2 | FileCheck %s -check-prefixes=CHECK,BMI,SSE42
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+bmi,+avx | FileCheck %s -check-prefixes=CHECK,BMI,AVX1
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+bmi,+avx2 | FileCheck %s -check-prefixes=CHECK,BMI,AVX2
; NOBMI-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v2i64'
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v2i64u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v2i64u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v4i64'
; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v4i64u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v4i64u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v8i64'
; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v8i64u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v8i64u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v4i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v4i32u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v4i32u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v8i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v8i32u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v8i32u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v16i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v16i32u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v16i32u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
}
define <16 x i16> @var_cttz_v16i16(<16 x i16> %a) {
+; NOBMI-LABEL: 'var_cttz_v16i16'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %cttz
+;
; SSE2-LABEL: 'var_cttz_v16i16'
; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %cttz
}
define <16 x i16> @var_cttz_v16i16u(<16 x i16> %a) {
+; NOBMI-LABEL: 'var_cttz_v16i16u'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %cttz
+;
; SSE2-LABEL: 'var_cttz_v16i16u'
; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %cttz
}
define <32 x i16> @var_cttz_v32i16(<32 x i16> %a) {
+; NOBMI-LABEL: 'var_cttz_v32i16'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %cttz
+;
; SSE2-LABEL: 'var_cttz_v32i16'
; SSE2-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %cttz
}
define <32 x i16> @var_cttz_v32i16u(<32 x i16> %a) {
+; NOBMI-LABEL: 'var_cttz_v32i16u'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %cttz
+;
; SSE2-LABEL: 'var_cttz_v32i16u'
; SSE2-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %cttz
}
define <16 x i8> @var_cttz_v16i8(<16 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v16i8'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v16i8'
; SSE2-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %cttz
}
define <16 x i8> @var_cttz_v16i8u(<16 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v16i8u'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v16i8u'
; SSE2-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %cttz
}
define <32 x i8> @var_cttz_v32i8(<32 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v32i8'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v32i8'
; SSE2-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %cttz
}
define <32 x i8> @var_cttz_v32i8u(<32 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v32i8u'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v32i8u'
; SSE2-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %cttz
}
define <64 x i8> @var_cttz_v64i8(<64 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v64i8'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v64i8'
; SSE2-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %cttz
}
define <64 x i8> @var_cttz_v64i8u(<64 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v64i8u'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v64i8u'
; SSE2-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %cttz
; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=-bmi,+sse2 | FileCheck %s -check-prefixes=CHECK,SSE2,NOBMI
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+bmi,+sse2 | FileCheck %s -check-prefixes=CHECK,SSE2,BMI
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=-bmi,+sse2 | FileCheck %s -check-prefixes=CHECK,NOBMI
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+bmi,+sse2 | FileCheck %s -check-prefixes=CHECK,BMI,SSE2
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+bmi,+sse4.2 | FileCheck %s -check-prefixes=CHECK,BMI,SSE42
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+bmi,+avx | FileCheck %s -check-prefixes=CHECK,BMI,AVX1
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+bmi,+avx2 | FileCheck %s -check-prefixes=CHECK,BMI,AVX2
; NOBMI-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v2i64'
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v2i64u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v2i64u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v4i64'
; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v4i64u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v4i64u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v8i64'
; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
;
+; SSE2-LABEL: 'var_cttz_v8i64u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
+;
; SSE42-LABEL: 'var_cttz_v8i64u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v4i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v4i32u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v4i32u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v8i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v8i32u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v8i32u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v16i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
; NOBMI-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
;
+; SSE2-LABEL: 'var_cttz_v16i32u'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
+;
; SSE42-LABEL: 'var_cttz_v16i32u'
; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %cttz
}
define <16 x i16> @var_cttz_v16i16(<16 x i16> %a) {
+; NOBMI-LABEL: 'var_cttz_v16i16'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %cttz
+;
; SSE2-LABEL: 'var_cttz_v16i16'
; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %cttz
}
define <16 x i16> @var_cttz_v16i16u(<16 x i16> %a) {
+; NOBMI-LABEL: 'var_cttz_v16i16u'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %cttz
+;
; SSE2-LABEL: 'var_cttz_v16i16u'
; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %cttz
}
define <32 x i16> @var_cttz_v32i16(<32 x i16> %a) {
+; NOBMI-LABEL: 'var_cttz_v32i16'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %cttz
+;
; SSE2-LABEL: 'var_cttz_v32i16'
; SSE2-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %cttz
}
define <32 x i16> @var_cttz_v32i16u(<32 x i16> %a) {
+; NOBMI-LABEL: 'var_cttz_v32i16u'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %cttz
+;
; SSE2-LABEL: 'var_cttz_v32i16u'
; SSE2-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %cttz
}
define <16 x i8> @var_cttz_v16i8(<16 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v16i8'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v16i8'
; SSE2-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %cttz
}
define <16 x i8> @var_cttz_v16i8u(<16 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v16i8u'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v16i8u'
; SSE2-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %cttz
}
define <32 x i8> @var_cttz_v32i8(<32 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v32i8'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v32i8'
; SSE2-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %cttz
}
define <32 x i8> @var_cttz_v32i8u(<32 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v32i8u'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v32i8u'
; SSE2-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %cttz
}
define <64 x i8> @var_cttz_v64i8(<64 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v64i8'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v64i8'
; SSE2-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %cttz
}
define <64 x i8> @var_cttz_v64i8u(<64 x i8> %a) {
+; NOBMI-LABEL: 'var_cttz_v64i8u'
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
+; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %cttz
+;
; SSE2-LABEL: 'var_cttz_v64i8u'
; SSE2-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %cttz