[llvm] 35ec06f - [X86][SSE] movmsk-cmp.ll - cleanup check prefixes and add SSE41 target tests
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Fri Jun 19 07:25:11 PDT 2020
Author: Simon Pilgrim
Date: 2020-06-19T15:24:59+01:00
New Revision: 35ec06f9669775044f6ff2f1db1203a6904e5d2a
URL: https://github.com/llvm/llvm-project/commit/35ec06f9669775044f6ff2f1db1203a6904e5d2a
DIFF: https://github.com/llvm/llvm-project/commit/35ec06f9669775044f6ff2f1db1203a6904e5d2a.diff
LOG: [X86][SSE] movmsk-cmp.ll - cleanup check prefixes and add SSE41 target tests
Added:
Modified:
llvm/test/CodeGen/X86/movmsk-cmp.ll
Removed:
################################################################################
diff --git a/llvm/test/CodeGen/X86/movmsk-cmp.ll b/llvm/test/CodeGen/X86/movmsk-cmp.ll
index 1a7599055b57..bc2edf43be38 100644
--- a/llvm/test/CodeGen/X86/movmsk-cmp.ll
+++ b/llvm/test/CodeGen/X86/movmsk-cmp.ll
@@ -1,17 +1,18 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=KNL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+avx512bw | FileCheck %s --check-prefix=SKX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX,AVX512,KNL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+avx512bw | FileCheck %s --check-prefixes=AVX,AVX512,SKX
define i1 @allones_v16i8_sign(<16 x i8> %arg) {
-; SSE2-LABEL: allones_v16i8_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpw $-1, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v16i8_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpw $-1, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX-LABEL: allones_v16i8_sign:
; AVX: # %bb.0:
@@ -19,20 +20,6 @@ define i1 @allones_v16i8_sign(<16 x i8> %arg) {
; AVX-NEXT: cmpw $-1, %ax
; AVX-NEXT: sete %al
; AVX-NEXT: retq
-;
-; KNL-LABEL: allones_v16i8_sign:
-; KNL: # %bb.0:
-; KNL-NEXT: vpmovmskb %xmm0, %eax
-; KNL-NEXT: cmpw $-1, %ax
-; KNL-NEXT: sete %al
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allones_v16i8_sign:
-; SKX: # %bb.0:
-; SKX-NEXT: vpmovmskb %xmm0, %eax
-; SKX-NEXT: cmpw $-1, %ax
-; SKX-NEXT: sete %al
-; SKX-NEXT: retq
%tmp = icmp slt <16 x i8> %arg, zeroinitializer
%tmp1 = bitcast <16 x i1> %tmp to i16
%tmp2 = icmp eq i16 %tmp1, -1
@@ -40,12 +27,12 @@ define i1 @allones_v16i8_sign(<16 x i8> %arg) {
}
define i1 @allzeros_v16i8_sign(<16 x i8> %arg) {
-; SSE2-LABEL: allzeros_v16i8_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testw %ax, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v16i8_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testw %ax, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX-LABEL: allzeros_v16i8_sign:
; AVX: # %bb.0:
@@ -53,20 +40,6 @@ define i1 @allzeros_v16i8_sign(<16 x i8> %arg) {
; AVX-NEXT: testw %ax, %ax
; AVX-NEXT: sete %al
; AVX-NEXT: retq
-;
-; KNL-LABEL: allzeros_v16i8_sign:
-; KNL: # %bb.0:
-; KNL-NEXT: vpmovmskb %xmm0, %eax
-; KNL-NEXT: testw %ax, %ax
-; KNL-NEXT: sete %al
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allzeros_v16i8_sign:
-; SKX: # %bb.0:
-; SKX-NEXT: vpmovmskb %xmm0, %eax
-; SKX-NEXT: testw %ax, %ax
-; SKX-NEXT: sete %al
-; SKX-NEXT: retq
%tmp = icmp slt <16 x i8> %arg, zeroinitializer
%tmp1 = bitcast <16 x i1> %tmp to i16
%tmp2 = icmp eq i16 %tmp1, 0
@@ -74,15 +47,15 @@ define i1 @allzeros_v16i8_sign(<16 x i8> %arg) {
}
define i1 @allones_v32i8_sign(<32 x i8> %arg) {
-; SSE2-LABEL: allones_v32i8_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: cmpl $-1, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v32i8_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: cmpl $-1, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v32i8_sign:
; AVX1: # %bb.0:
@@ -104,21 +77,13 @@ define i1 @allones_v32i8_sign(<32 x i8> %arg) {
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; KNL-LABEL: allones_v32i8_sign:
-; KNL: # %bb.0:
-; KNL-NEXT: vpmovmskb %ymm0, %eax
-; KNL-NEXT: cmpl $-1, %eax
-; KNL-NEXT: sete %al
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allones_v32i8_sign:
-; SKX: # %bb.0:
-; SKX-NEXT: vpmovmskb %ymm0, %eax
-; SKX-NEXT: cmpl $-1, %eax
-; SKX-NEXT: sete %al
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
+; AVX512-LABEL: allones_v32i8_sign:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpmovmskb %ymm0, %eax
+; AVX512-NEXT: cmpl $-1, %eax
+; AVX512-NEXT: sete %al
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%tmp = icmp slt <32 x i8> %arg, zeroinitializer
%tmp1 = bitcast <32 x i1> %tmp to i32
%tmp2 = icmp eq i32 %tmp1, -1
@@ -126,14 +91,14 @@ define i1 @allones_v32i8_sign(<32 x i8> %arg) {
}
define i1 @allzeros_v32i8_sign(<32 x i8> %arg) {
-; SSE2-LABEL: allzeros_v32i8_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v32i8_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v32i8_sign:
; AVX1: # %bb.0:
@@ -154,21 +119,13 @@ define i1 @allzeros_v32i8_sign(<32 x i8> %arg) {
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; KNL-LABEL: allzeros_v32i8_sign:
-; KNL: # %bb.0:
-; KNL-NEXT: vpmovmskb %ymm0, %eax
-; KNL-NEXT: testl %eax, %eax
-; KNL-NEXT: sete %al
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allzeros_v32i8_sign:
-; SKX: # %bb.0:
-; SKX-NEXT: vpmovmskb %ymm0, %eax
-; SKX-NEXT: testl %eax, %eax
-; SKX-NEXT: sete %al
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
+; AVX512-LABEL: allzeros_v32i8_sign:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpmovmskb %ymm0, %eax
+; AVX512-NEXT: testl %eax, %eax
+; AVX512-NEXT: sete %al
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%tmp = icmp slt <32 x i8> %arg, zeroinitializer
%tmp1 = bitcast <32 x i1> %tmp to i32
%tmp2 = icmp eq i32 %tmp1, 0
@@ -176,21 +133,21 @@ define i1 @allzeros_v32i8_sign(<32 x i8> %arg) {
}
define i1 @allones_v64i8_sign(<64 x i8> %arg) {
-; SSE2-LABEL: allones_v64i8_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: pmovmskb %xmm2, %eax
-; SSE2-NEXT: pmovmskb %xmm3, %edx
-; SSE2-NEXT: shll $16, %edx
-; SSE2-NEXT: orl %eax, %edx
-; SSE2-NEXT: shlq $32, %rdx
-; SSE2-NEXT: orq %rcx, %rdx
-; SSE2-NEXT: cmpq $-1, %rdx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v64i8_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: pmovmskb %xmm2, %eax
+; SSE-NEXT: pmovmskb %xmm3, %edx
+; SSE-NEXT: shll $16, %edx
+; SSE-NEXT: orl %eax, %edx
+; SSE-NEXT: shlq $32, %rdx
+; SSE-NEXT: orq %rcx, %rdx
+; SSE-NEXT: cmpq $-1, %rdx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v64i8_sign:
; AVX1: # %bb.0:
@@ -248,20 +205,20 @@ define i1 @allones_v64i8_sign(<64 x i8> %arg) {
}
define i1 @allzeros_v64i8_sign(<64 x i8> %arg) {
-; SSE2-LABEL: allzeros_v64i8_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: pmovmskb %xmm2, %eax
-; SSE2-NEXT: pmovmskb %xmm3, %edx
-; SSE2-NEXT: shll $16, %edx
-; SSE2-NEXT: orl %eax, %edx
-; SSE2-NEXT: shlq $32, %rdx
-; SSE2-NEXT: orq %rcx, %rdx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v64i8_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: pmovmskb %xmm2, %eax
+; SSE-NEXT: pmovmskb %xmm3, %edx
+; SSE-NEXT: shll $16, %edx
+; SSE-NEXT: orl %eax, %edx
+; SSE-NEXT: shlq $32, %rdx
+; SSE-NEXT: orq %rcx, %rdx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v64i8_sign:
; AVX1: # %bb.0:
@@ -316,21 +273,21 @@ define i1 @allzeros_v64i8_sign(<64 x i8> %arg) {
}
define i1 @allones_v8i16_sign(<8 x i16> %arg) {
-; SSE2-LABEL: allones_v8i16_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpb $-1, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allones_v8i16_sign:
-; AVX: # %bb.0:
-; AVX-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: cmpb $-1, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allones_v8i16_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allones_v8i16_sign:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: cmpb $-1, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allones_v8i16_sign:
; KNL: # %bb.0:
@@ -357,19 +314,19 @@ define i1 @allones_v8i16_sign(<8 x i16> %arg) {
}
define i1 @allzeros_v8i16_sign(<8 x i16> %arg) {
-; SSE2-LABEL: allzeros_v8i16_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testl $43690, %eax # imm = 0xAAAA
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allzeros_v8i16_sign:
-; AVX: # %bb.0:
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: testl $43690, %eax # imm = 0xAAAA
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allzeros_v8i16_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testl $43690, %eax # imm = 0xAAAA
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v8i16_sign:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: testl $43690, %eax # imm = 0xAAAA
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v8i16_sign:
; KNL: # %bb.0:
@@ -396,13 +353,13 @@ define i1 @allzeros_v8i16_sign(<8 x i16> %arg) {
}
define i1 @allones_v16i16_sign(<16 x i16> %arg) {
-; SSE2-LABEL: allones_v16i16_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpw $-1, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v16i16_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpw $-1, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v16i16_sign:
; AVX1: # %bb.0:
@@ -449,13 +406,13 @@ define i1 @allones_v16i16_sign(<16 x i16> %arg) {
}
define i1 @allzeros_v16i16_sign(<16 x i16> %arg) {
-; SSE2-LABEL: allzeros_v16i16_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testw %ax, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v16i16_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testw %ax, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v16i16_sign:
; AVX1: # %bb.0:
@@ -500,17 +457,17 @@ define i1 @allzeros_v16i16_sign(<16 x i16> %arg) {
}
define i1 @allones_v32i16_sign(<32 x i16> %arg) {
-; SSE2-LABEL: allones_v32i16_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: packsswb %xmm3, %xmm2
-; SSE2-NEXT: pmovmskb %xmm2, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: cmpl $-1, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v32i16_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: packsswb %xmm3, %xmm2
+; SSE-NEXT: pmovmskb %xmm2, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: cmpl $-1, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v32i16_sign:
; AVX1: # %bb.0:
@@ -569,16 +526,16 @@ define i1 @allones_v32i16_sign(<32 x i16> %arg) {
}
define i1 @allzeros_v32i16_sign(<32 x i16> %arg) {
-; SSE2-LABEL: allzeros_v32i16_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: packsswb %xmm3, %xmm2
-; SSE2-NEXT: pmovmskb %xmm2, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v32i16_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: packsswb %xmm3, %xmm2
+; SSE-NEXT: pmovmskb %xmm2, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v32i16_sign:
; AVX1: # %bb.0:
@@ -635,12 +592,12 @@ define i1 @allzeros_v32i16_sign(<32 x i16> %arg) {
}
define i1 @allones_v4i32_sign(<4 x i32> %arg) {
-; SSE2-LABEL: allones_v4i32_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: cmpb $15, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v4i32_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: cmpb $15, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX-LABEL: allones_v4i32_sign:
; AVX: # %bb.0:
@@ -648,20 +605,6 @@ define i1 @allones_v4i32_sign(<4 x i32> %arg) {
; AVX-NEXT: cmpb $15, %al
; AVX-NEXT: sete %al
; AVX-NEXT: retq
-;
-; KNL-LABEL: allones_v4i32_sign:
-; KNL: # %bb.0:
-; KNL-NEXT: vmovmskps %xmm0, %eax
-; KNL-NEXT: cmpb $15, %al
-; KNL-NEXT: sete %al
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allones_v4i32_sign:
-; SKX: # %bb.0:
-; SKX-NEXT: vmovmskps %xmm0, %eax
-; SKX-NEXT: cmpb $15, %al
-; SKX-NEXT: sete %al
-; SKX-NEXT: retq
%tmp = icmp slt <4 x i32> %arg, zeroinitializer
%tmp1 = bitcast <4 x i1> %tmp to i4
%tmp2 = icmp eq i4 %tmp1, -1
@@ -669,12 +612,12 @@ define i1 @allones_v4i32_sign(<4 x i32> %arg) {
}
define i1 @allzeros_v4i32_sign(<4 x i32> %arg) {
-; SSE2-LABEL: allzeros_v4i32_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: testb %al, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v4i32_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: testb %al, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX-LABEL: allzeros_v4i32_sign:
; AVX: # %bb.0:
@@ -682,20 +625,6 @@ define i1 @allzeros_v4i32_sign(<4 x i32> %arg) {
; AVX-NEXT: testb %al, %al
; AVX-NEXT: sete %al
; AVX-NEXT: retq
-;
-; KNL-LABEL: allzeros_v4i32_sign:
-; KNL: # %bb.0:
-; KNL-NEXT: vmovmskps %xmm0, %eax
-; KNL-NEXT: testb %al, %al
-; KNL-NEXT: sete %al
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allzeros_v4i32_sign:
-; SKX: # %bb.0:
-; SKX-NEXT: vmovmskps %xmm0, %eax
-; SKX-NEXT: testb %al, %al
-; SKX-NEXT: sete %al
-; SKX-NEXT: retq
%tmp = icmp slt <4 x i32> %arg, zeroinitializer
%tmp1 = bitcast <4 x i1> %tmp to i4
%tmp2 = icmp eq i4 %tmp1, 0
@@ -703,14 +632,14 @@ define i1 @allzeros_v4i32_sign(<4 x i32> %arg) {
}
define i1 @allones_v8i32_sign(<8 x i32> %arg) {
-; SSE2-LABEL: allones_v8i32_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpb $-1, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v8i32_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX-LABEL: allones_v8i32_sign:
; AVX: # %bb.0:
@@ -719,22 +648,6 @@ define i1 @allones_v8i32_sign(<8 x i32> %arg) {
; AVX-NEXT: sete %al
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
-;
-; KNL-LABEL: allones_v8i32_sign:
-; KNL: # %bb.0:
-; KNL-NEXT: vmovmskps %ymm0, %eax
-; KNL-NEXT: cmpb $-1, %al
-; KNL-NEXT: sete %al
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allones_v8i32_sign:
-; SKX: # %bb.0:
-; SKX-NEXT: vmovmskps %ymm0, %eax
-; SKX-NEXT: cmpb $-1, %al
-; SKX-NEXT: sete %al
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
%tmp = icmp slt <8 x i32> %arg, zeroinitializer
%tmp1 = bitcast <8 x i1> %tmp to i8
%tmp2 = icmp eq i8 %tmp1, -1
@@ -742,13 +655,13 @@ define i1 @allones_v8i32_sign(<8 x i32> %arg) {
}
define i1 @allzeros_v8i32_sign(<8 x i32> %arg) {
-; SSE2-LABEL: allzeros_v8i32_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testl $43690, %eax # imm = 0xAAAA
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v8i32_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testl $43690, %eax # imm = 0xAAAA
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX-LABEL: allzeros_v8i32_sign:
; AVX: # %bb.0:
@@ -757,22 +670,6 @@ define i1 @allzeros_v8i32_sign(<8 x i32> %arg) {
; AVX-NEXT: sete %al
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
-;
-; KNL-LABEL: allzeros_v8i32_sign:
-; KNL: # %bb.0:
-; KNL-NEXT: vmovmskps %ymm0, %eax
-; KNL-NEXT: testb %al, %al
-; KNL-NEXT: sete %al
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allzeros_v8i32_sign:
-; SKX: # %bb.0:
-; SKX-NEXT: vmovmskps %ymm0, %eax
-; SKX-NEXT: testb %al, %al
-; SKX-NEXT: sete %al
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
%tmp = icmp slt <8 x i32> %arg, zeroinitializer
%tmp1 = bitcast <8 x i1> %tmp to i8
%tmp2 = icmp eq i8 %tmp1, 0
@@ -780,15 +677,15 @@ define i1 @allzeros_v8i32_sign(<8 x i32> %arg) {
}
define i1 @allones_v16i32_sign(<16 x i32> %arg) {
-; SSE2-LABEL: allones_v16i32_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm2, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpw $-1, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v16i32_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm2, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpw $-1, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v16i32_sign:
; AVX1: # %bb.0:
@@ -841,15 +738,15 @@ define i1 @allones_v16i32_sign(<16 x i32> %arg) {
}
define i1 @allzeros_v16i32_sign(<16 x i32> %arg) {
-; SSE2-LABEL: allzeros_v16i32_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm2, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testw %ax, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v16i32_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm2, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testw %ax, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v16i32_sign:
; AVX1: # %bb.0:
@@ -900,13 +797,13 @@ define i1 @allzeros_v16i32_sign(<16 x i32> %arg) {
}
define i1 @allones_v4i64_sign(<4 x i64> %arg) {
-; SSE2-LABEL: allones_v4i64_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: cmpb $15, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v4i64_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: cmpb $15, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX-LABEL: allones_v4i64_sign:
; AVX: # %bb.0:
@@ -915,22 +812,6 @@ define i1 @allones_v4i64_sign(<4 x i64> %arg) {
; AVX-NEXT: sete %al
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
-;
-; KNL-LABEL: allones_v4i64_sign:
-; KNL: # %bb.0:
-; KNL-NEXT: vmovmskpd %ymm0, %eax
-; KNL-NEXT: cmpb $15, %al
-; KNL-NEXT: sete %al
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allones_v4i64_sign:
-; SKX: # %bb.0:
-; SKX-NEXT: vmovmskpd %ymm0, %eax
-; SKX-NEXT: cmpb $15, %al
-; SKX-NEXT: sete %al
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
%tmp = icmp slt <4 x i64> %arg, zeroinitializer
%tmp1 = bitcast <4 x i1> %tmp to i4
%tmp2 = icmp eq i4 %tmp1, -1
@@ -938,13 +819,13 @@ define i1 @allones_v4i64_sign(<4 x i64> %arg) {
}
define i1 @allzeros_v4i64_sign(<4 x i64> %arg) {
-; SSE2-LABEL: allzeros_v4i64_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: testb %al, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v4i64_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: testb %al, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX-LABEL: allzeros_v4i64_sign:
; AVX: # %bb.0:
@@ -953,22 +834,6 @@ define i1 @allzeros_v4i64_sign(<4 x i64> %arg) {
; AVX-NEXT: sete %al
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
-;
-; KNL-LABEL: allzeros_v4i64_sign:
-; KNL: # %bb.0:
-; KNL-NEXT: vmovmskpd %ymm0, %eax
-; KNL-NEXT: testb %al, %al
-; KNL-NEXT: sete %al
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allzeros_v4i64_sign:
-; SKX: # %bb.0:
-; SKX-NEXT: vmovmskpd %ymm0, %eax
-; SKX-NEXT: testb %al, %al
-; SKX-NEXT: sete %al
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
%tmp = icmp slt <4 x i64> %arg, zeroinitializer
%tmp1 = bitcast <4 x i1> %tmp to i4
%tmp2 = icmp eq i4 %tmp1, 0
@@ -976,16 +841,16 @@ define i1 @allzeros_v4i64_sign(<4 x i64> %arg) {
}
define i1 @allones_v8i64_sign(<8 x i64> %arg) {
-; SSE2-LABEL: allones_v8i64_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packssdw %xmm2, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpb $-1, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v8i64_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packssdw %xmm2, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v8i64_sign:
; AVX1: # %bb.0:
@@ -1036,15 +901,15 @@ define i1 @allones_v8i64_sign(<8 x i64> %arg) {
}
define i1 @allzeros_v8i64_sign(<8 x i64> %arg) {
-; SSE2-LABEL: allzeros_v8i64_sign:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packssdw %xmm2, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testl $43690, %eax # imm = 0xAAAA
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v8i64_sign:
+; SSE: # %bb.0:
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packssdw %xmm2, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testl $43690, %eax # imm = 0xAAAA
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v8i64_sign:
; AVX1: # %bb.0:
@@ -1095,21 +960,21 @@ define i1 @allzeros_v8i64_sign(<8 x i64> %arg) {
}
define i1 @allones_v16i8_and1(<16 x i8> %arg) {
-; SSE2-LABEL: allones_v16i8_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $7, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpw $-1, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allones_v16i8_and1:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllw $7, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: cmpw $-1, %ax
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allones_v16i8_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $7, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpw $-1, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allones_v16i8_and1:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllw $7, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: cmpw $-1, %ax
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allones_v16i8_and1:
; KNL: # %bb.0:
@@ -1133,23 +998,23 @@ define i1 @allones_v16i8_and1(<16 x i8> %arg) {
}
define i1 @allzeros_v16i8_not(<16 x i8> %a0) {
-; SSE2-LABEL: allzeros_v16i8_not:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pxor %xmm1, %xmm1
-; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
-; SSE2-NEXT: pmovmskb %xmm1, %eax
-; SSE2-NEXT: cmpw $-1, %ax
-; SSE2-NEXT: setne %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allzeros_v16i8_not:
-; AVX: # %bb.0:
-; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: cmpw $-1, %ax
-; AVX-NEXT: setne %al
-; AVX-NEXT: retq
+; SSE-LABEL: allzeros_v16i8_not:
+; SSE: # %bb.0:
+; SSE-NEXT: pxor %xmm1, %xmm1
+; SSE-NEXT: pcmpeqb %xmm0, %xmm1
+; SSE-NEXT: pmovmskb %xmm1, %eax
+; SSE-NEXT: cmpw $-1, %ax
+; SSE-NEXT: setne %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v16i8_not:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; AVX1OR2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: cmpw $-1, %ax
+; AVX1OR2-NEXT: setne %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v16i8_not:
; KNL: # %bb.0:
@@ -1184,14 +1049,23 @@ define i1 @allzeros_v2i64_not(<2 x i64> %a0) {
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
-; AVX-LABEL: allzeros_v2i64_not:
-; AVX: # %bb.0:
-; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vmovmskpd %xmm0, %eax
-; AVX-NEXT: cmpb $3, %al
-; AVX-NEXT: setne %al
-; AVX-NEXT: retq
+; SSE41-LABEL: allzeros_v2i64_not:
+; SSE41: # %bb.0:
+; SSE41-NEXT: pxor %xmm1, %xmm1
+; SSE41-NEXT: pcmpeqq %xmm0, %xmm1
+; SSE41-NEXT: movmskpd %xmm1, %eax
+; SSE41-NEXT: cmpb $3, %al
+; SSE41-NEXT: setne %al
+; SSE41-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v2i64_not:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; AVX1OR2-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
+; AVX1OR2-NEXT: cmpb $3, %al
+; AVX1OR2-NEXT: setne %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v2i64_not:
; KNL: # %bb.0:
@@ -1218,17 +1092,17 @@ define i1 @allzeros_v2i64_not(<2 x i64> %a0) {
}
define i1 @allzeros_v8i32_not(<8 x i32> %a0) {
-; SSE2-LABEL: allzeros_v8i32_not:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pxor %xmm2, %xmm2
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm1
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpb $-1, %al
-; SSE2-NEXT: setne %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v8i32_not:
+; SSE: # %bb.0:
+; SSE-NEXT: pxor %xmm2, %xmm2
+; SSE-NEXT: pcmpeqd %xmm2, %xmm1
+; SSE-NEXT: pcmpeqd %xmm2, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: setne %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v8i32_not:
; AVX1: # %bb.0:
@@ -1301,6 +1175,22 @@ define i1 @allzeros_v8i64_not(<8 x i64> %a0) {
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
+; SSE41-LABEL: allzeros_v8i64_not:
+; SSE41: # %bb.0:
+; SSE41-NEXT: pxor %xmm4, %xmm4
+; SSE41-NEXT: pcmpeqq %xmm4, %xmm3
+; SSE41-NEXT: pcmpeqq %xmm4, %xmm2
+; SSE41-NEXT: packssdw %xmm3, %xmm2
+; SSE41-NEXT: pcmpeqq %xmm4, %xmm1
+; SSE41-NEXT: pcmpeqq %xmm4, %xmm0
+; SSE41-NEXT: packssdw %xmm1, %xmm0
+; SSE41-NEXT: packssdw %xmm2, %xmm0
+; SSE41-NEXT: packsswb %xmm0, %xmm0
+; SSE41-NEXT: pmovmskb %xmm0, %eax
+; SSE41-NEXT: cmpb $-1, %al
+; SSE41-NEXT: setne %al
+; SSE41-NEXT: retq
+;
; AVX1-LABEL: allzeros_v8i64_not:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
@@ -1354,21 +1244,21 @@ define i1 @allzeros_v8i64_not(<8 x i64> %a0) {
}
define i1 @allzeros_v16i8_and1(<16 x i8> %arg) {
-; SSE2-LABEL: allzeros_v16i8_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $7, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testw %ax, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allzeros_v16i8_and1:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllw $7, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: testw %ax, %ax
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allzeros_v16i8_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $7, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testw %ax, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v16i8_and1:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllw $7, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: testw %ax, %ax
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v16i8_and1:
; KNL: # %bb.0:
@@ -1392,17 +1282,17 @@ define i1 @allzeros_v16i8_and1(<16 x i8> %arg) {
}
define i1 @allones_v32i8_and1(<32 x i8> %arg) {
-; SSE2-LABEL: allones_v32i8_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $7, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $7, %xmm1
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: cmpl $-1, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v32i8_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $7, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $7, %xmm1
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: cmpl $-1, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v32i8_and1:
; AVX1: # %bb.0:
@@ -1451,16 +1341,16 @@ define i1 @allones_v32i8_and1(<32 x i8> %arg) {
}
define i1 @allzeros_v32i8_and1(<32 x i8> %arg) {
-; SSE2-LABEL: allzeros_v32i8_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $7, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $7, %xmm1
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v32i8_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $7, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $7, %xmm1
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v32i8_and1:
; AVX1: # %bb.0:
@@ -1508,25 +1398,25 @@ define i1 @allzeros_v32i8_and1(<32 x i8> %arg) {
}
define i1 @allones_v64i8_and1(<64 x i8> %arg) {
-; SSE2-LABEL: allones_v64i8_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $7, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $7, %xmm1
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: psllw $7, %xmm2
-; SSE2-NEXT: pmovmskb %xmm2, %eax
-; SSE2-NEXT: psllw $7, %xmm3
-; SSE2-NEXT: pmovmskb %xmm3, %edx
-; SSE2-NEXT: shll $16, %edx
-; SSE2-NEXT: orl %eax, %edx
-; SSE2-NEXT: shlq $32, %rdx
-; SSE2-NEXT: orq %rcx, %rdx
-; SSE2-NEXT: cmpq $-1, %rdx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v64i8_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $7, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $7, %xmm1
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: psllw $7, %xmm2
+; SSE-NEXT: pmovmskb %xmm2, %eax
+; SSE-NEXT: psllw $7, %xmm3
+; SSE-NEXT: pmovmskb %xmm3, %edx
+; SSE-NEXT: shll $16, %edx
+; SSE-NEXT: orl %eax, %edx
+; SSE-NEXT: shlq $32, %rdx
+; SSE-NEXT: orq %rcx, %rdx
+; SSE-NEXT: cmpq $-1, %rdx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v64i8_and1:
; AVX1: # %bb.0:
@@ -1593,24 +1483,24 @@ define i1 @allones_v64i8_and1(<64 x i8> %arg) {
}
define i1 @allzeros_v64i8_and1(<64 x i8> %arg) {
-; SSE2-LABEL: allzeros_v64i8_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $7, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $7, %xmm1
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: psllw $7, %xmm2
-; SSE2-NEXT: pmovmskb %xmm2, %eax
-; SSE2-NEXT: psllw $7, %xmm3
-; SSE2-NEXT: pmovmskb %xmm3, %edx
-; SSE2-NEXT: shll $16, %edx
-; SSE2-NEXT: orl %eax, %edx
-; SSE2-NEXT: shlq $32, %rdx
-; SSE2-NEXT: orq %rcx, %rdx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v64i8_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $7, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $7, %xmm1
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: psllw $7, %xmm2
+; SSE-NEXT: pmovmskb %xmm2, %eax
+; SSE-NEXT: psllw $7, %xmm3
+; SSE-NEXT: pmovmskb %xmm3, %edx
+; SSE-NEXT: shll $16, %edx
+; SSE-NEXT: orl %eax, %edx
+; SSE-NEXT: shlq $32, %rdx
+; SSE-NEXT: orq %rcx, %rdx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v64i8_and1:
; AVX1: # %bb.0:
@@ -1674,23 +1564,23 @@ define i1 @allzeros_v64i8_and1(<64 x i8> %arg) {
}
define i1 @allones_v8i16_and1(<8 x i16> %arg) {
-; SSE2-LABEL: allones_v8i16_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $15, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpb $-1, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allones_v8i16_and1:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllw $15, %xmm0, %xmm0
-; AVX-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: cmpb $-1, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allones_v8i16_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $15, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allones_v8i16_and1:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllw $15, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: cmpb $-1, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allones_v8i16_and1:
; KNL: # %bb.0:
@@ -1718,21 +1608,21 @@ define i1 @allones_v8i16_and1(<8 x i16> %arg) {
}
define i1 @allzeros_v8i16_and1(<8 x i16> %arg) {
-; SSE2-LABEL: allzeros_v8i16_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $15, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testl $43690, %eax # imm = 0xAAAA
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allzeros_v8i16_and1:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllw $15, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: testl $43690, %eax # imm = 0xAAAA
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allzeros_v8i16_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $15, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testl $43690, %eax # imm = 0xAAAA
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v8i16_and1:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllw $15, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: testl $43690, %eax # imm = 0xAAAA
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v8i16_and1:
; KNL: # %bb.0:
@@ -1760,15 +1650,15 @@ define i1 @allzeros_v8i16_and1(<8 x i16> %arg) {
}
define i1 @allones_v16i16_and1(<16 x i16> %arg) {
-; SSE2-LABEL: allones_v16i16_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $15, %xmm1
-; SSE2-NEXT: psllw $15, %xmm0
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpw $-1, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v16i16_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $15, %xmm1
+; SSE-NEXT: psllw $15, %xmm0
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpw $-1, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v16i16_and1:
; AVX1: # %bb.0:
@@ -1819,21 +1709,21 @@ define i1 @allones_v16i16_and1(<16 x i16> %arg) {
}
define i1 @allones_v32i16_and1(<32 x i16> %arg) {
-; SSE2-LABEL: allones_v32i16_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $15, %xmm1
-; SSE2-NEXT: psllw $15, %xmm0
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $15, %xmm3
-; SSE2-NEXT: psllw $15, %xmm2
-; SSE2-NEXT: packsswb %xmm3, %xmm2
-; SSE2-NEXT: pmovmskb %xmm2, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: cmpl $-1, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v32i16_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $15, %xmm1
+; SSE-NEXT: psllw $15, %xmm0
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $15, %xmm3
+; SSE-NEXT: psllw $15, %xmm2
+; SSE-NEXT: packsswb %xmm3, %xmm2
+; SSE-NEXT: pmovmskb %xmm2, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: cmpl $-1, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v32i16_and1:
; AVX1: # %bb.0:
@@ -1900,20 +1790,20 @@ define i1 @allones_v32i16_and1(<32 x i16> %arg) {
}
define i1 @allzeros_v32i16_and1(<32 x i16> %arg) {
-; SSE2-LABEL: allzeros_v32i16_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $15, %xmm1
-; SSE2-NEXT: psllw $15, %xmm0
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $15, %xmm3
-; SSE2-NEXT: psllw $15, %xmm2
-; SSE2-NEXT: packsswb %xmm3, %xmm2
-; SSE2-NEXT: pmovmskb %xmm2, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v32i16_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $15, %xmm1
+; SSE-NEXT: psllw $15, %xmm0
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $15, %xmm3
+; SSE-NEXT: psllw $15, %xmm2
+; SSE-NEXT: packsswb %xmm3, %xmm2
+; SSE-NEXT: pmovmskb %xmm2, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v32i16_and1:
; AVX1: # %bb.0:
@@ -1978,15 +1868,15 @@ define i1 @allzeros_v32i16_and1(<32 x i16> %arg) {
}
define i1 @allzeros_v16i16_and1(<16 x i16> %arg) {
-; SSE2-LABEL: allzeros_v16i16_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $15, %xmm1
-; SSE2-NEXT: psllw $15, %xmm0
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testw %ax, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v16i16_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $15, %xmm1
+; SSE-NEXT: psllw $15, %xmm0
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testw %ax, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v16i16_and1:
; AVX1: # %bb.0:
@@ -2035,21 +1925,21 @@ define i1 @allzeros_v16i16_and1(<16 x i16> %arg) {
}
define i1 @allones_v4i32_and1(<4 x i32> %arg) {
-; SSE2-LABEL: allones_v4i32_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $31, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: cmpb $15, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allones_v4i32_and1:
-; AVX: # %bb.0:
-; AVX-NEXT: vpslld $31, %xmm0, %xmm0
-; AVX-NEXT: vmovmskps %xmm0, %eax
-; AVX-NEXT: cmpb $15, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allones_v4i32_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $31, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: cmpb $15, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allones_v4i32_and1:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpslld $31, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskps %xmm0, %eax
+; AVX1OR2-NEXT: cmpb $15, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allones_v4i32_and1:
; KNL: # %bb.0:
@@ -2077,21 +1967,21 @@ define i1 @allones_v4i32_and1(<4 x i32> %arg) {
}
define i1 @allzeros_v4i32_and1(<4 x i32> %arg) {
-; SSE2-LABEL: allzeros_v4i32_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $31, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: testb %al, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allzeros_v4i32_and1:
-; AVX: # %bb.0:
-; AVX-NEXT: vpslld $31, %xmm0, %xmm0
-; AVX-NEXT: vmovmskps %xmm0, %eax
-; AVX-NEXT: testb %al, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allzeros_v4i32_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $31, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: testb %al, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v4i32_and1:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpslld $31, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskps %xmm0, %eax
+; AVX1OR2-NEXT: testb %al, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v4i32_and1:
; KNL: # %bb.0:
@@ -2117,16 +2007,16 @@ define i1 @allzeros_v4i32_and1(<4 x i32> %arg) {
}
define i1 @allones_v8i32_and1(<8 x i32> %arg) {
-; SSE2-LABEL: allones_v8i32_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $31, %xmm1
-; SSE2-NEXT: pslld $31, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpb $-1, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v8i32_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $31, %xmm1
+; SSE-NEXT: pslld $31, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v8i32_and1:
; AVX1: # %bb.0:
@@ -2174,15 +2064,15 @@ define i1 @allones_v8i32_and1(<8 x i32> %arg) {
}
define i1 @allzeros_v8i32_and1(<8 x i32> %arg) {
-; SSE2-LABEL: allzeros_v8i32_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $31, %xmm1
-; SSE2-NEXT: pslld $31, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testl $43690, %eax # imm = 0xAAAA
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v8i32_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $31, %xmm1
+; SSE-NEXT: pslld $31, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testl $43690, %eax # imm = 0xAAAA
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v8i32_and1:
; AVX1: # %bb.0:
@@ -2230,19 +2120,19 @@ define i1 @allzeros_v8i32_and1(<8 x i32> %arg) {
}
define i1 @allones_v16i32_and1(<16 x i32> %arg) {
-; SSE2-LABEL: allones_v16i32_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $31, %xmm3
-; SSE2-NEXT: pslld $31, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: pslld $31, %xmm1
-; SSE2-NEXT: pslld $31, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm2, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpw $-1, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v16i32_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $31, %xmm3
+; SSE-NEXT: pslld $31, %xmm2
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: pslld $31, %xmm1
+; SSE-NEXT: pslld $31, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm2, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpw $-1, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v16i32_and1:
; AVX1: # %bb.0:
@@ -2277,21 +2167,13 @@ define i1 @allones_v16i32_and1(<16 x i32> %arg) {
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; KNL-LABEL: allones_v16i32_and1:
-; KNL: # %bb.0:
-; KNL-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
-; KNL-NEXT: kortestw %k0, %k0
-; KNL-NEXT: setb %al
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allones_v16i32_and1:
-; SKX: # %bb.0:
-; SKX-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
-; SKX-NEXT: kortestw %k0, %k0
-; SKX-NEXT: setb %al
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
+; AVX512-LABEL: allones_v16i32_and1:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
+; AVX512-NEXT: kortestw %k0, %k0
+; AVX512-NEXT: setb %al
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%tmp = and <16 x i32> %arg, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
%tmp1 = icmp ne <16 x i32> %tmp, zeroinitializer
%tmp2 = bitcast <16 x i1> %tmp1 to i16
@@ -2300,19 +2182,19 @@ define i1 @allones_v16i32_and1(<16 x i32> %arg) {
}
define i1 @allzeros_v16i32_and1(<16 x i32> %arg) {
-; SSE2-LABEL: allzeros_v16i32_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $31, %xmm3
-; SSE2-NEXT: pslld $31, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: pslld $31, %xmm1
-; SSE2-NEXT: pslld $31, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm2, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testw %ax, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v16i32_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $31, %xmm3
+; SSE-NEXT: pslld $31, %xmm2
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: pslld $31, %xmm1
+; SSE-NEXT: pslld $31, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm2, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testw %ax, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v16i32_and1:
; AVX1: # %bb.0:
@@ -2345,21 +2227,13 @@ define i1 @allzeros_v16i32_and1(<16 x i32> %arg) {
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; KNL-LABEL: allzeros_v16i32_and1:
-; KNL: # %bb.0:
-; KNL-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
-; KNL-NEXT: kortestw %k0, %k0
-; KNL-NEXT: sete %al
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allzeros_v16i32_and1:
-; SKX: # %bb.0:
-; SKX-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
-; SKX-NEXT: kortestw %k0, %k0
-; SKX-NEXT: sete %al
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
+; AVX512-LABEL: allzeros_v16i32_and1:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
+; AVX512-NEXT: kortestw %k0, %k0
+; AVX512-NEXT: sete %al
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%tmp = and <16 x i32> %arg, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
%tmp1 = icmp ne <16 x i32> %tmp, zeroinitializer
%tmp2 = bitcast <16 x i1> %tmp1 to i16
@@ -2368,21 +2242,21 @@ define i1 @allzeros_v16i32_and1(<16 x i32> %arg) {
}
define i1 @allones_v2i64_and1(<2 x i64> %arg) {
-; SSE2-LABEL: allones_v2i64_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $63, %xmm0
-; SSE2-NEXT: movmskpd %xmm0, %eax
-; SSE2-NEXT: cmpb $3, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allones_v2i64_and1:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllq $63, %xmm0, %xmm0
-; AVX-NEXT: vmovmskpd %xmm0, %eax
-; AVX-NEXT: cmpb $3, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allones_v2i64_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $63, %xmm0
+; SSE-NEXT: movmskpd %xmm0, %eax
+; SSE-NEXT: cmpb $3, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allones_v2i64_and1:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllq $63, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
+; AVX1OR2-NEXT: cmpb $3, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allones_v2i64_and1:
; KNL: # %bb.0:
@@ -2411,21 +2285,21 @@ define i1 @allones_v2i64_and1(<2 x i64> %arg) {
}
define i1 @allzeros_v2i64_and1(<2 x i64> %arg) {
-; SSE2-LABEL: allzeros_v2i64_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $63, %xmm0
-; SSE2-NEXT: movmskpd %xmm0, %eax
-; SSE2-NEXT: testb %al, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allzeros_v2i64_and1:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllq $63, %xmm0, %xmm0
-; AVX-NEXT: vmovmskpd %xmm0, %eax
-; AVX-NEXT: testb %al, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allzeros_v2i64_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $63, %xmm0
+; SSE-NEXT: movmskpd %xmm0, %eax
+; SSE-NEXT: testb %al, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v2i64_and1:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllq $63, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
+; AVX1OR2-NEXT: testb %al, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v2i64_and1:
; KNL: # %bb.0:
@@ -2452,15 +2326,15 @@ define i1 @allzeros_v2i64_and1(<2 x i64> %arg) {
}
define i1 @allones_v4i64_and1(<4 x i64> %arg) {
-; SSE2-LABEL: allones_v4i64_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $63, %xmm1
-; SSE2-NEXT: psllq $63, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: cmpb $15, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v4i64_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $63, %xmm1
+; SSE-NEXT: psllq $63, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: cmpb $15, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v4i64_and1:
; AVX1: # %bb.0:
@@ -2510,15 +2384,15 @@ define i1 @allones_v4i64_and1(<4 x i64> %arg) {
}
define i1 @allzeros_v4i64_and1(<4 x i64> %arg) {
-; SSE2-LABEL: allzeros_v4i64_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $63, %xmm1
-; SSE2-NEXT: psllq $63, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: testb %al, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v4i64_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $63, %xmm1
+; SSE-NEXT: psllq $63, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: testb %al, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v4i64_and1:
; AVX1: # %bb.0:
@@ -2566,20 +2440,20 @@ define i1 @allzeros_v4i64_and1(<4 x i64> %arg) {
}
define i1 @allones_v8i64_and1(<8 x i64> %arg) {
-; SSE2-LABEL: allones_v8i64_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $63, %xmm3
-; SSE2-NEXT: psllq $63, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: psllq $63, %xmm1
-; SSE2-NEXT: psllq $63, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packssdw %xmm2, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpb $-1, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v8i64_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $63, %xmm3
+; SSE-NEXT: psllq $63, %xmm2
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: psllq $63, %xmm1
+; SSE-NEXT: psllq $63, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packssdw %xmm2, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v8i64_and1:
; AVX1: # %bb.0:
@@ -2636,19 +2510,19 @@ define i1 @allones_v8i64_and1(<8 x i64> %arg) {
}
define i1 @allzeros_v8i64_and1(<8 x i64> %arg) {
-; SSE2-LABEL: allzeros_v8i64_and1:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $63, %xmm3
-; SSE2-NEXT: psllq $63, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: psllq $63, %xmm1
-; SSE2-NEXT: psllq $63, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packssdw %xmm2, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testl $43690, %eax # imm = 0xAAAA
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v8i64_and1:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $63, %xmm3
+; SSE-NEXT: psllq $63, %xmm2
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: psllq $63, %xmm1
+; SSE-NEXT: psllq $63, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packssdw %xmm2, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testl $43690, %eax # imm = 0xAAAA
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v8i64_and1:
; AVX1: # %bb.0:
@@ -2705,21 +2579,21 @@ define i1 @allzeros_v8i64_and1(<8 x i64> %arg) {
}
define i1 @allones_v16i8_and4(<16 x i8> %arg) {
-; SSE2-LABEL: allones_v16i8_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $5, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpw $-1, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allones_v16i8_and4:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllw $5, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: cmpw $-1, %ax
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allones_v16i8_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $5, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpw $-1, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allones_v16i8_and4:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllw $5, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: cmpw $-1, %ax
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allones_v16i8_and4:
; KNL: # %bb.0:
@@ -2735,29 +2609,29 @@ define i1 @allones_v16i8_and4(<16 x i8> %arg) {
; SKX-NEXT: kortestw %k0, %k0
; SKX-NEXT: setb %al
; SKX-NEXT: retq
- %tmp = and <16 x i8> %arg, <i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4>
- %tmp1 = icmp ne <16 x i8> %tmp, zeroinitializer
- %tmp2 = bitcast <16 x i1> %tmp1 to i16
- %tmp3 = icmp eq i16 %tmp2, -1
- ret i1 %tmp3
-}
-
-define i1 @allzeros_v16i8_and4(<16 x i8> %arg) {
-; SSE2-LABEL: allzeros_v16i8_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $5, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testw %ax, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allzeros_v16i8_and4:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllw $5, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: testw %ax, %ax
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+ %tmp = and <16 x i8> %arg, <i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4>
+ %tmp1 = icmp ne <16 x i8> %tmp, zeroinitializer
+ %tmp2 = bitcast <16 x i1> %tmp1 to i16
+ %tmp3 = icmp eq i16 %tmp2, -1
+ ret i1 %tmp3
+}
+
+define i1 @allzeros_v16i8_and4(<16 x i8> %arg) {
+; SSE-LABEL: allzeros_v16i8_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $5, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testw %ax, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v16i8_and4:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllw $5, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: testw %ax, %ax
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v16i8_and4:
; KNL: # %bb.0:
@@ -2781,17 +2655,17 @@ define i1 @allzeros_v16i8_and4(<16 x i8> %arg) {
}
define i1 @allones_v32i8_and4(<32 x i8> %arg) {
-; SSE2-LABEL: allones_v32i8_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $5, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $5, %xmm1
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: cmpl $-1, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v32i8_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $5, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $5, %xmm1
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: cmpl $-1, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v32i8_and4:
; AVX1: # %bb.0:
@@ -2840,16 +2714,16 @@ define i1 @allones_v32i8_and4(<32 x i8> %arg) {
}
define i1 @allzeros_v32i8_and4(<32 x i8> %arg) {
-; SSE2-LABEL: allzeros_v32i8_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $5, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $5, %xmm1
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v32i8_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $5, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $5, %xmm1
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v32i8_and4:
; AVX1: # %bb.0:
@@ -2897,25 +2771,25 @@ define i1 @allzeros_v32i8_and4(<32 x i8> %arg) {
}
define i1 @allones_v64i8_and4(<64 x i8> %arg) {
-; SSE2-LABEL: allones_v64i8_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $5, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $5, %xmm1
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: psllw $5, %xmm2
-; SSE2-NEXT: pmovmskb %xmm2, %eax
-; SSE2-NEXT: psllw $5, %xmm3
-; SSE2-NEXT: pmovmskb %xmm3, %edx
-; SSE2-NEXT: shll $16, %edx
-; SSE2-NEXT: orl %eax, %edx
-; SSE2-NEXT: shlq $32, %rdx
-; SSE2-NEXT: orq %rcx, %rdx
-; SSE2-NEXT: cmpq $-1, %rdx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v64i8_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $5, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $5, %xmm1
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: psllw $5, %xmm2
+; SSE-NEXT: pmovmskb %xmm2, %eax
+; SSE-NEXT: psllw $5, %xmm3
+; SSE-NEXT: pmovmskb %xmm3, %edx
+; SSE-NEXT: shll $16, %edx
+; SSE-NEXT: orl %eax, %edx
+; SSE-NEXT: shlq $32, %rdx
+; SSE-NEXT: orq %rcx, %rdx
+; SSE-NEXT: cmpq $-1, %rdx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v64i8_and4:
; AVX1: # %bb.0:
@@ -2982,24 +2856,24 @@ define i1 @allones_v64i8_and4(<64 x i8> %arg) {
}
define i1 @allzeros_v64i8_and4(<64 x i8> %arg) {
-; SSE2-LABEL: allzeros_v64i8_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $5, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $5, %xmm1
-; SSE2-NEXT: pmovmskb %xmm1, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: psllw $5, %xmm2
-; SSE2-NEXT: pmovmskb %xmm2, %eax
-; SSE2-NEXT: psllw $5, %xmm3
-; SSE2-NEXT: pmovmskb %xmm3, %edx
-; SSE2-NEXT: shll $16, %edx
-; SSE2-NEXT: orl %eax, %edx
-; SSE2-NEXT: shlq $32, %rdx
-; SSE2-NEXT: orq %rcx, %rdx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v64i8_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $5, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $5, %xmm1
+; SSE-NEXT: pmovmskb %xmm1, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: psllw $5, %xmm2
+; SSE-NEXT: pmovmskb %xmm2, %eax
+; SSE-NEXT: psllw $5, %xmm3
+; SSE-NEXT: pmovmskb %xmm3, %edx
+; SSE-NEXT: shll $16, %edx
+; SSE-NEXT: orl %eax, %edx
+; SSE-NEXT: shlq $32, %rdx
+; SSE-NEXT: orq %rcx, %rdx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v64i8_and4:
; AVX1: # %bb.0:
@@ -3063,23 +2937,23 @@ define i1 @allzeros_v64i8_and4(<64 x i8> %arg) {
}
define i1 @allones_v8i16_and4(<8 x i16> %arg) {
-; SSE2-LABEL: allones_v8i16_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $13, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpb $-1, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allones_v8i16_and4:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllw $13, %xmm0, %xmm0
-; AVX-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: cmpb $-1, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allones_v8i16_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $13, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allones_v8i16_and4:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllw $13, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: cmpb $-1, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allones_v8i16_and4:
; KNL: # %bb.0:
@@ -3107,21 +2981,21 @@ define i1 @allones_v8i16_and4(<8 x i16> %arg) {
}
define i1 @allzeros_v8i16_and4(<8 x i16> %arg) {
-; SSE2-LABEL: allzeros_v8i16_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $13, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testl $43690, %eax # imm = 0xAAAA
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allzeros_v8i16_and4:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllw $13, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: testl $43690, %eax # imm = 0xAAAA
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allzeros_v8i16_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $13, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testl $43690, %eax # imm = 0xAAAA
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v8i16_and4:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllw $13, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: testl $43690, %eax # imm = 0xAAAA
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v8i16_and4:
; KNL: # %bb.0:
@@ -3149,15 +3023,15 @@ define i1 @allzeros_v8i16_and4(<8 x i16> %arg) {
}
define i1 @allones_v16i16_and4(<16 x i16> %arg) {
-; SSE2-LABEL: allones_v16i16_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $13, %xmm1
-; SSE2-NEXT: psllw $13, %xmm0
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpw $-1, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v16i16_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $13, %xmm1
+; SSE-NEXT: psllw $13, %xmm0
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpw $-1, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v16i16_and4:
; AVX1: # %bb.0:
@@ -3208,21 +3082,21 @@ define i1 @allones_v16i16_and4(<16 x i16> %arg) {
}
define i1 @allones_v32i16_and4(<32 x i16> %arg) {
-; SSE2-LABEL: allones_v32i16_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $13, %xmm1
-; SSE2-NEXT: psllw $13, %xmm0
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $13, %xmm3
-; SSE2-NEXT: psllw $13, %xmm2
-; SSE2-NEXT: packsswb %xmm3, %xmm2
-; SSE2-NEXT: pmovmskb %xmm2, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: cmpl $-1, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v32i16_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $13, %xmm1
+; SSE-NEXT: psllw $13, %xmm0
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $13, %xmm3
+; SSE-NEXT: psllw $13, %xmm2
+; SSE-NEXT: packsswb %xmm3, %xmm2
+; SSE-NEXT: pmovmskb %xmm2, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: cmpl $-1, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v32i16_and4:
; AVX1: # %bb.0:
@@ -3289,20 +3163,20 @@ define i1 @allones_v32i16_and4(<32 x i16> %arg) {
}
define i1 @allzeros_v32i16_and4(<32 x i16> %arg) {
-; SSE2-LABEL: allzeros_v32i16_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $13, %xmm1
-; SSE2-NEXT: psllw $13, %xmm0
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: psllw $13, %xmm3
-; SSE2-NEXT: psllw $13, %xmm2
-; SSE2-NEXT: packsswb %xmm3, %xmm2
-; SSE2-NEXT: pmovmskb %xmm2, %ecx
-; SSE2-NEXT: shll $16, %ecx
-; SSE2-NEXT: orl %eax, %ecx
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v32i16_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $13, %xmm1
+; SSE-NEXT: psllw $13, %xmm0
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: psllw $13, %xmm3
+; SSE-NEXT: psllw $13, %xmm2
+; SSE-NEXT: packsswb %xmm3, %xmm2
+; SSE-NEXT: pmovmskb %xmm2, %ecx
+; SSE-NEXT: shll $16, %ecx
+; SSE-NEXT: orl %eax, %ecx
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v32i16_and4:
; AVX1: # %bb.0:
@@ -3367,15 +3241,15 @@ define i1 @allzeros_v32i16_and4(<32 x i16> %arg) {
}
define i1 @allzeros_v16i16_and4(<16 x i16> %arg) {
-; SSE2-LABEL: allzeros_v16i16_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllw $13, %xmm1
-; SSE2-NEXT: psllw $13, %xmm0
-; SSE2-NEXT: packsswb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testw %ax, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v16i16_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllw $13, %xmm1
+; SSE-NEXT: psllw $13, %xmm0
+; SSE-NEXT: packsswb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testw %ax, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v16i16_and4:
; AVX1: # %bb.0:
@@ -3424,21 +3298,21 @@ define i1 @allzeros_v16i16_and4(<16 x i16> %arg) {
}
define i1 @allones_v4i32_and4(<4 x i32> %arg) {
-; SSE2-LABEL: allones_v4i32_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $29, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: cmpb $15, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allones_v4i32_and4:
-; AVX: # %bb.0:
-; AVX-NEXT: vpslld $29, %xmm0, %xmm0
-; AVX-NEXT: vmovmskps %xmm0, %eax
-; AVX-NEXT: cmpb $15, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allones_v4i32_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $29, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: cmpb $15, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allones_v4i32_and4:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpslld $29, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskps %xmm0, %eax
+; AVX1OR2-NEXT: cmpb $15, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allones_v4i32_and4:
; KNL: # %bb.0:
@@ -3466,21 +3340,21 @@ define i1 @allones_v4i32_and4(<4 x i32> %arg) {
}
define i1 @allzeros_v4i32_and4(<4 x i32> %arg) {
-; SSE2-LABEL: allzeros_v4i32_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $29, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: testb %al, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allzeros_v4i32_and4:
-; AVX: # %bb.0:
-; AVX-NEXT: vpslld $29, %xmm0, %xmm0
-; AVX-NEXT: vmovmskps %xmm0, %eax
-; AVX-NEXT: testb %al, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allzeros_v4i32_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $29, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: testb %al, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v4i32_and4:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpslld $29, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskps %xmm0, %eax
+; AVX1OR2-NEXT: testb %al, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v4i32_and4:
; KNL: # %bb.0:
@@ -3506,16 +3380,16 @@ define i1 @allzeros_v4i32_and4(<4 x i32> %arg) {
}
define i1 @allones_v8i32_and4(<8 x i32> %arg) {
-; SSE2-LABEL: allones_v8i32_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $29, %xmm1
-; SSE2-NEXT: pslld $29, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpb $-1, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v8i32_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $29, %xmm1
+; SSE-NEXT: pslld $29, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v8i32_and4:
; AVX1: # %bb.0:
@@ -3563,15 +3437,15 @@ define i1 @allones_v8i32_and4(<8 x i32> %arg) {
}
define i1 @allzeros_v8i32_and4(<8 x i32> %arg) {
-; SSE2-LABEL: allzeros_v8i32_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $29, %xmm1
-; SSE2-NEXT: pslld $29, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testl $43690, %eax # imm = 0xAAAA
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v8i32_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $29, %xmm1
+; SSE-NEXT: pslld $29, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testl $43690, %eax # imm = 0xAAAA
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v8i32_and4:
; AVX1: # %bb.0:
@@ -3619,19 +3493,19 @@ define i1 @allzeros_v8i32_and4(<8 x i32> %arg) {
}
define i1 @allones_v16i32_and4(<16 x i32> %arg) {
-; SSE2-LABEL: allones_v16i32_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $29, %xmm3
-; SSE2-NEXT: pslld $29, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: pslld $29, %xmm1
-; SSE2-NEXT: pslld $29, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm2, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpw $-1, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v16i32_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $29, %xmm3
+; SSE-NEXT: pslld $29, %xmm2
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: pslld $29, %xmm1
+; SSE-NEXT: pslld $29, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm2, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpw $-1, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v16i32_and4:
; AVX1: # %bb.0:
@@ -3666,21 +3540,13 @@ define i1 @allones_v16i32_and4(<16 x i32> %arg) {
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; KNL-LABEL: allones_v16i32_and4:
-; KNL: # %bb.0:
-; KNL-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
-; KNL-NEXT: kortestw %k0, %k0
-; KNL-NEXT: setb %al
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allones_v16i32_and4:
-; SKX: # %bb.0:
-; SKX-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
-; SKX-NEXT: kortestw %k0, %k0
-; SKX-NEXT: setb %al
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
+; AVX512-LABEL: allones_v16i32_and4:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
+; AVX512-NEXT: kortestw %k0, %k0
+; AVX512-NEXT: setb %al
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%tmp = and <16 x i32> %arg, <i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4>
%tmp1 = icmp ne <16 x i32> %tmp, zeroinitializer
%tmp2 = bitcast <16 x i1> %tmp1 to i16
@@ -3689,19 +3555,19 @@ define i1 @allones_v16i32_and4(<16 x i32> %arg) {
}
define i1 @allzeros_v16i32_and4(<16 x i32> %arg) {
-; SSE2-LABEL: allzeros_v16i32_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pslld $29, %xmm3
-; SSE2-NEXT: pslld $29, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: pslld $29, %xmm1
-; SSE2-NEXT: pslld $29, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm2, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testw %ax, %ax
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v16i32_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: pslld $29, %xmm3
+; SSE-NEXT: pslld $29, %xmm2
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: pslld $29, %xmm1
+; SSE-NEXT: pslld $29, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm2, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testw %ax, %ax
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v16i32_and4:
; AVX1: # %bb.0:
@@ -3734,21 +3600,13 @@ define i1 @allzeros_v16i32_and4(<16 x i32> %arg) {
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; KNL-LABEL: allzeros_v16i32_and4:
-; KNL: # %bb.0:
-; KNL-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
-; KNL-NEXT: kortestw %k0, %k0
-; KNL-NEXT: sete %al
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: allzeros_v16i32_and4:
-; SKX: # %bb.0:
-; SKX-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
-; SKX-NEXT: kortestw %k0, %k0
-; SKX-NEXT: sete %al
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
+; AVX512-LABEL: allzeros_v16i32_and4:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
+; AVX512-NEXT: kortestw %k0, %k0
+; AVX512-NEXT: sete %al
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%tmp = and <16 x i32> %arg, <i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4>
%tmp1 = icmp ne <16 x i32> %tmp, zeroinitializer
%tmp2 = bitcast <16 x i1> %tmp1 to i16
@@ -3757,21 +3615,21 @@ define i1 @allzeros_v16i32_and4(<16 x i32> %arg) {
}
define i1 @allones_v2i64_and4(<2 x i64> %arg) {
-; SSE2-LABEL: allones_v2i64_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $61, %xmm0
-; SSE2-NEXT: movmskpd %xmm0, %eax
-; SSE2-NEXT: cmpb $3, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allones_v2i64_and4:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllq $61, %xmm0, %xmm0
-; AVX-NEXT: vmovmskpd %xmm0, %eax
-; AVX-NEXT: cmpb $3, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allones_v2i64_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $61, %xmm0
+; SSE-NEXT: movmskpd %xmm0, %eax
+; SSE-NEXT: cmpb $3, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allones_v2i64_and4:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllq $61, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
+; AVX1OR2-NEXT: cmpb $3, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allones_v2i64_and4:
; KNL: # %bb.0:
@@ -3800,21 +3658,21 @@ define i1 @allones_v2i64_and4(<2 x i64> %arg) {
}
define i1 @allzeros_v2i64_and4(<2 x i64> %arg) {
-; SSE2-LABEL: allzeros_v2i64_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $61, %xmm0
-; SSE2-NEXT: movmskpd %xmm0, %eax
-; SSE2-NEXT: testb %al, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: allzeros_v2i64_and4:
-; AVX: # %bb.0:
-; AVX-NEXT: vpsllq $61, %xmm0, %xmm0
-; AVX-NEXT: vmovmskpd %xmm0, %eax
-; AVX-NEXT: testb %al, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: allzeros_v2i64_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $61, %xmm0
+; SSE-NEXT: movmskpd %xmm0, %eax
+; SSE-NEXT: testb %al, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: allzeros_v2i64_and4:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpsllq $61, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
+; AVX1OR2-NEXT: testb %al, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: allzeros_v2i64_and4:
; KNL: # %bb.0:
@@ -3841,15 +3699,15 @@ define i1 @allzeros_v2i64_and4(<2 x i64> %arg) {
}
define i1 @allones_v4i64_and4(<4 x i64> %arg) {
-; SSE2-LABEL: allones_v4i64_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $61, %xmm1
-; SSE2-NEXT: psllq $61, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: cmpb $15, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v4i64_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $61, %xmm1
+; SSE-NEXT: psllq $61, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: cmpb $15, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v4i64_and4:
; AVX1: # %bb.0:
@@ -3899,15 +3757,15 @@ define i1 @allones_v4i64_and4(<4 x i64> %arg) {
}
define i1 @allzeros_v4i64_and4(<4 x i64> %arg) {
-; SSE2-LABEL: allzeros_v4i64_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $61, %xmm1
-; SSE2-NEXT: psllq $61, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: testb %al, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v4i64_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $61, %xmm1
+; SSE-NEXT: psllq $61, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: testb %al, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v4i64_and4:
; AVX1: # %bb.0:
@@ -3955,20 +3813,20 @@ define i1 @allzeros_v4i64_and4(<4 x i64> %arg) {
}
define i1 @allones_v8i64_and4(<8 x i64> %arg) {
-; SSE2-LABEL: allones_v8i64_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $61, %xmm3
-; SSE2-NEXT: psllq $61, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: psllq $61, %xmm1
-; SSE2-NEXT: psllq $61, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packssdw %xmm2, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: cmpb $-1, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allones_v8i64_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $61, %xmm3
+; SSE-NEXT: psllq $61, %xmm2
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: psllq $61, %xmm1
+; SSE-NEXT: psllq $61, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packssdw %xmm2, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allones_v8i64_and4:
; AVX1: # %bb.0:
@@ -4025,19 +3883,19 @@ define i1 @allones_v8i64_and4(<8 x i64> %arg) {
}
define i1 @allzeros_v8i64_and4(<8 x i64> %arg) {
-; SSE2-LABEL: allzeros_v8i64_and4:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psllq $61, %xmm3
-; SSE2-NEXT: psllq $61, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: psllq $61, %xmm1
-; SSE2-NEXT: psllq $61, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packssdw %xmm2, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: testl $43690, %eax # imm = 0xAAAA
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
+; SSE-LABEL: allzeros_v8i64_and4:
+; SSE: # %bb.0:
+; SSE-NEXT: psllq $61, %xmm3
+; SSE-NEXT: psllq $61, %xmm2
+; SSE-NEXT: packssdw %xmm3, %xmm2
+; SSE-NEXT: psllq $61, %xmm1
+; SSE-NEXT: psllq $61, %xmm0
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packssdw %xmm2, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testl $43690, %eax # imm = 0xAAAA
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
;
; AVX1-LABEL: allzeros_v8i64_and4:
; AVX1: # %bb.0:
@@ -4097,25 +3955,15 @@ define i1 @allzeros_v8i64_and4(<8 x i64> %arg) {
; MOVMSK instruction.
define i32 @movmskpd(<2 x double> %x) {
-; SSE2-LABEL: movmskpd:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movmskpd %xmm0, %eax
-; SSE2-NEXT: retq
+; SSE-LABEL: movmskpd:
+; SSE: # %bb.0:
+; SSE-NEXT: movmskpd %xmm0, %eax
+; SSE-NEXT: retq
;
; AVX-LABEL: movmskpd:
; AVX: # %bb.0:
; AVX-NEXT: vmovmskpd %xmm0, %eax
; AVX-NEXT: retq
-;
-; KNL-LABEL: movmskpd:
-; KNL: # %bb.0:
-; KNL-NEXT: vmovmskpd %xmm0, %eax
-; KNL-NEXT: retq
-;
-; SKX-LABEL: movmskpd:
-; SKX: # %bb.0:
-; SKX-NEXT: vmovmskpd %xmm0, %eax
-; SKX-NEXT: retq
%a = bitcast <2 x double> %x to <2 x i64>
%b = icmp slt <2 x i64> %a, zeroinitializer
%c = bitcast <2 x i1> %b to i2
@@ -4124,25 +3972,15 @@ define i32 @movmskpd(<2 x double> %x) {
}
define i32 @movmskps(<4 x float> %x) {
-; SSE2-LABEL: movmskps:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: retq
+; SSE-LABEL: movmskps:
+; SSE: # %bb.0:
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: retq
;
; AVX-LABEL: movmskps:
; AVX: # %bb.0:
; AVX-NEXT: vmovmskps %xmm0, %eax
; AVX-NEXT: retq
-;
-; KNL-LABEL: movmskps:
-; KNL: # %bb.0:
-; KNL-NEXT: vmovmskps %xmm0, %eax
-; KNL-NEXT: retq
-;
-; SKX-LABEL: movmskps:
-; SKX: # %bb.0:
-; SKX-NEXT: vmovmskps %xmm0, %eax
-; SKX-NEXT: retq
%a = bitcast <4 x float> %x to <4 x i32>
%b = icmp slt <4 x i32> %a, zeroinitializer
%c = bitcast <4 x i1> %b to i4
@@ -4151,29 +3989,17 @@ define i32 @movmskps(<4 x float> %x) {
}
define i32 @movmskpd256(<4 x double> %x) {
-; SSE2-LABEL: movmskpd256:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: retq
+; SSE-LABEL: movmskpd256:
+; SSE: # %bb.0:
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: retq
;
; AVX-LABEL: movmskpd256:
; AVX: # %bb.0:
; AVX-NEXT: vmovmskpd %ymm0, %eax
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
-;
-; KNL-LABEL: movmskpd256:
-; KNL: # %bb.0:
-; KNL-NEXT: vmovmskpd %ymm0, %eax
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: movmskpd256:
-; SKX: # %bb.0:
-; SKX-NEXT: vmovmskpd %ymm0, %eax
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
%a = bitcast <4 x double> %x to <4 x i64>
%b = icmp slt <4 x i64> %a, zeroinitializer
%c = bitcast <4 x i1> %b to i4
@@ -4182,31 +4008,19 @@ define i32 @movmskpd256(<4 x double> %x) {
}
define i32 @movmskps256(<8 x float> %x) {
-; SSE2-LABEL: movmskps256:
-; SSE2: # %bb.0:
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: movzbl %al, %eax
-; SSE2-NEXT: retq
+; SSE-LABEL: movmskps256:
+; SSE: # %bb.0:
+; SSE-NEXT: packssdw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: movzbl %al, %eax
+; SSE-NEXT: retq
;
; AVX-LABEL: movmskps256:
; AVX: # %bb.0:
; AVX-NEXT: vmovmskps %ymm0, %eax
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
-;
-; KNL-LABEL: movmskps256:
-; KNL: # %bb.0:
-; KNL-NEXT: vmovmskps %ymm0, %eax
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: movmskps256:
-; SKX: # %bb.0:
-; SKX-NEXT: vmovmskps %ymm0, %eax
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
%a = bitcast <8 x float> %x to <8 x i32>
%b = icmp slt <8 x i32> %a, zeroinitializer
%c = bitcast <8 x i1> %b to i8
@@ -4215,25 +4029,15 @@ define i32 @movmskps256(<8 x float> %x) {
}
define i32 @movmskb(<16 x i8> %x) {
-; SSE2-LABEL: movmskb:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: retq
+; SSE-LABEL: movmskb:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: retq
;
; AVX-LABEL: movmskb:
; AVX: # %bb.0:
; AVX-NEXT: vpmovmskb %xmm0, %eax
; AVX-NEXT: retq
-;
-; KNL-LABEL: movmskb:
-; KNL: # %bb.0:
-; KNL-NEXT: vpmovmskb %xmm0, %eax
-; KNL-NEXT: retq
-;
-; SKX-LABEL: movmskb:
-; SKX: # %bb.0:
-; SKX-NEXT: vpmovmskb %xmm0, %eax
-; SKX-NEXT: retq
%a = icmp slt <16 x i8> %x, zeroinitializer
%b = bitcast <16 x i1> %a to i16
%c = zext i16 %b to i32
@@ -4241,13 +4045,13 @@ define i32 @movmskb(<16 x i8> %x) {
}
define i32 @movmskb256(<32 x i8> %x) {
-; SSE2-LABEL: movmskb256:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pmovmskb %xmm0, %ecx
-; SSE2-NEXT: pmovmskb %xmm1, %eax
-; SSE2-NEXT: shll $16, %eax
-; SSE2-NEXT: orl %ecx, %eax
-; SSE2-NEXT: retq
+; SSE-LABEL: movmskb256:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovmskb %xmm0, %ecx
+; SSE-NEXT: pmovmskb %xmm1, %eax
+; SSE-NEXT: shll $16, %eax
+; SSE-NEXT: orl %ecx, %eax
+; SSE-NEXT: retq
;
; AVX1-LABEL: movmskb256:
; AVX1: # %bb.0:
@@ -4265,17 +4069,11 @@ define i32 @movmskb256(<32 x i8> %x) {
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; KNL-LABEL: movmskb256:
-; KNL: # %bb.0:
-; KNL-NEXT: vpmovmskb %ymm0, %eax
-; KNL-NEXT: vzeroupper
-; KNL-NEXT: retq
-;
-; SKX-LABEL: movmskb256:
-; SKX: # %bb.0:
-; SKX-NEXT: vpmovmskb %ymm0, %eax
-; SKX-NEXT: vzeroupper
-; SKX-NEXT: retq
+; AVX512-LABEL: movmskb256:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpmovmskb %ymm0, %eax
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%a = icmp slt <32 x i8> %x, zeroinitializer
%b = bitcast <32 x i1> %a to i32
ret i32 %b
@@ -4284,37 +4082,37 @@ define i32 @movmskb256(<32 x i8> %x) {
; Multiple extract elements from a vector compare.
define i1 @movmsk_v16i8(<16 x i8> %x, <16 x i8> %y) {
-; SSE2-LABEL: movmsk_v16i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pcmpeqb %xmm1, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: movl %eax, %ecx
-; SSE2-NEXT: shrl $15, %ecx
-; SSE2-NEXT: movl %eax, %edx
-; SSE2-NEXT: shrl $8, %edx
-; SSE2-NEXT: andl $1, %edx
-; SSE2-NEXT: andl $8, %eax
-; SSE2-NEXT: shrl $3, %eax
-; SSE2-NEXT: xorl %edx, %eax
-; SSE2-NEXT: andl %ecx, %eax
-; SSE2-NEXT: # kill: def $al killed $al killed $eax
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: movmsk_v16i8:
-; AVX: # %bb.0:
-; AVX-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: movl %eax, %ecx
-; AVX-NEXT: shrl $15, %ecx
-; AVX-NEXT: movl %eax, %edx
-; AVX-NEXT: shrl $8, %edx
-; AVX-NEXT: andl $1, %edx
-; AVX-NEXT: andl $8, %eax
-; AVX-NEXT: shrl $3, %eax
-; AVX-NEXT: xorl %edx, %eax
-; AVX-NEXT: andl %ecx, %eax
-; AVX-NEXT: # kill: def $al killed $al killed $eax
-; AVX-NEXT: retq
+; SSE-LABEL: movmsk_v16i8:
+; SSE: # %bb.0:
+; SSE-NEXT: pcmpeqb %xmm1, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: movl %eax, %ecx
+; SSE-NEXT: shrl $15, %ecx
+; SSE-NEXT: movl %eax, %edx
+; SSE-NEXT: shrl $8, %edx
+; SSE-NEXT: andl $1, %edx
+; SSE-NEXT: andl $8, %eax
+; SSE-NEXT: shrl $3, %eax
+; SSE-NEXT: xorl %edx, %eax
+; SSE-NEXT: andl %ecx, %eax
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: movmsk_v16i8:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: movl %eax, %ecx
+; AVX1OR2-NEXT: shrl $15, %ecx
+; AVX1OR2-NEXT: movl %eax, %edx
+; AVX1OR2-NEXT: shrl $8, %edx
+; AVX1OR2-NEXT: andl $1, %edx
+; AVX1OR2-NEXT: andl $8, %eax
+; AVX1OR2-NEXT: shrl $3, %eax
+; AVX1OR2-NEXT: xorl %edx, %eax
+; AVX1OR2-NEXT: andl %ecx, %eax
+; AVX1OR2-NEXT: # kill: def $al killed $al killed $eax
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: movmsk_v16i8:
; KNL: # %bb.0:
@@ -4356,25 +4154,25 @@ define i1 @movmsk_v16i8(<16 x i8> %x, <16 x i8> %y) {
}
define i1 @movmsk_v8i16(<8 x i16> %x, <8 x i16> %y) {
-; SSE2-LABEL: movmsk_v8i16:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pcmpgtw %xmm1, %xmm0
-; SSE2-NEXT: packsswb %xmm0, %xmm0
-; SSE2-NEXT: pmovmskb %xmm0, %eax
-; SSE2-NEXT: andb $-109, %al
-; SSE2-NEXT: cmpb $-109, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: movmsk_v8i16:
-; AVX: # %bb.0:
-; AVX-NEXT: vpcmpgtw %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
-; AVX-NEXT: vpmovmskb %xmm0, %eax
-; AVX-NEXT: andb $-109, %al
-; AVX-NEXT: cmpb $-109, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: movmsk_v8i16:
+; SSE: # %bb.0:
+; SSE-NEXT: pcmpgtw %xmm1, %xmm0
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: andb $-109, %al
+; SSE-NEXT: cmpb $-109, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: movmsk_v8i16:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpcmpgtw %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
+; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT: andb $-109, %al
+; AVX1OR2-NEXT: cmpb $-109, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: movmsk_v8i16:
; KNL: # %bb.0:
@@ -4409,29 +4207,29 @@ define i1 @movmsk_v8i16(<8 x i16> %x, <8 x i16> %y) {
; TODO: Replace shift+mask chain with AND+CMP.
define i1 @movmsk_v4i32(<4 x i32> %x, <4 x i32> %y) {
-; SSE2-LABEL: movmsk_v4i32:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm1
-; SSE2-NEXT: movmskps %xmm1, %eax
-; SSE2-NEXT: movl %eax, %ecx
-; SSE2-NEXT: shrb $3, %cl
-; SSE2-NEXT: andb $4, %al
-; SSE2-NEXT: shrb $2, %al
-; SSE2-NEXT: xorb %cl, %al
-; SSE2-NEXT: # kill: def $al killed $al killed $eax
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: movmsk_v4i32:
-; AVX: # %bb.0:
-; AVX-NEXT: vpcmpgtd %xmm0, %xmm1, %xmm0
-; AVX-NEXT: vmovmskps %xmm0, %eax
-; AVX-NEXT: movl %eax, %ecx
-; AVX-NEXT: shrb $3, %cl
-; AVX-NEXT: andb $4, %al
-; AVX-NEXT: shrb $2, %al
-; AVX-NEXT: xorb %cl, %al
-; AVX-NEXT: # kill: def $al killed $al killed $eax
-; AVX-NEXT: retq
+; SSE-LABEL: movmsk_v4i32:
+; SSE: # %bb.0:
+; SSE-NEXT: pcmpgtd %xmm0, %xmm1
+; SSE-NEXT: movmskps %xmm1, %eax
+; SSE-NEXT: movl %eax, %ecx
+; SSE-NEXT: shrb $3, %cl
+; SSE-NEXT: andb $4, %al
+; SSE-NEXT: shrb $2, %al
+; SSE-NEXT: xorb %cl, %al
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: movmsk_v4i32:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpcmpgtd %xmm0, %xmm1, %xmm0
+; AVX1OR2-NEXT: vmovmskps %xmm0, %eax
+; AVX1OR2-NEXT: movl %eax, %ecx
+; AVX1OR2-NEXT: shrb $3, %cl
+; AVX1OR2-NEXT: andb $4, %al
+; AVX1OR2-NEXT: shrb $2, %al
+; AVX1OR2-NEXT: xorb %cl, %al
+; AVX1OR2-NEXT: # kill: def $al killed $al killed $eax
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: movmsk_v4i32:
; KNL: # %bb.0:
@@ -4476,14 +4274,23 @@ define i1 @movmsk_and_v2i64(<2 x i64> %x, <2 x i64> %y) {
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
-; AVX-LABEL: movmsk_and_v2i64:
-; AVX: # %bb.0:
-; AVX-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vmovmskpd %xmm0, %eax
-; AVX-NEXT: xorl $3, %eax
-; AVX-NEXT: cmpb $3, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE41-LABEL: movmsk_and_v2i64:
+; SSE41: # %bb.0:
+; SSE41-NEXT: pcmpeqq %xmm1, %xmm0
+; SSE41-NEXT: movmskpd %xmm0, %eax
+; SSE41-NEXT: xorl $3, %eax
+; SSE41-NEXT: cmpb $3, %al
+; SSE41-NEXT: sete %al
+; SSE41-NEXT: retq
+;
+; AVX1OR2-LABEL: movmsk_and_v2i64:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
+; AVX1OR2-NEXT: xorl $3, %eax
+; AVX1OR2-NEXT: cmpb $3, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: movmsk_and_v2i64:
; KNL: # %bb.0:
@@ -4522,13 +4329,21 @@ define i1 @movmsk_or_v2i64(<2 x i64> %x, <2 x i64> %y) {
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
-; AVX-LABEL: movmsk_or_v2i64:
-; AVX: # %bb.0:
-; AVX-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vmovmskpd %xmm0, %eax
-; AVX-NEXT: xorb $3, %al
-; AVX-NEXT: setne %al
-; AVX-NEXT: retq
+; SSE41-LABEL: movmsk_or_v2i64:
+; SSE41: # %bb.0:
+; SSE41-NEXT: pcmpeqq %xmm1, %xmm0
+; SSE41-NEXT: movmskpd %xmm0, %eax
+; SSE41-NEXT: xorb $3, %al
+; SSE41-NEXT: setne %al
+; SSE41-NEXT: retq
+;
+; AVX1OR2-LABEL: movmsk_or_v2i64:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
+; AVX1OR2-NEXT: xorb $3, %al
+; AVX1OR2-NEXT: setne %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: movmsk_or_v2i64:
; KNL: # %bb.0:
@@ -4555,24 +4370,24 @@ define i1 @movmsk_or_v2i64(<2 x i64> %x, <2 x i64> %y) {
}
define i1 @movmsk_v4f32(<4 x float> %x, <4 x float> %y) {
-; SSE2-LABEL: movmsk_v4f32:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movaps %xmm0, %xmm2
-; SSE2-NEXT: cmpeqps %xmm1, %xmm2
-; SSE2-NEXT: cmpunordps %xmm1, %xmm0
-; SSE2-NEXT: orps %xmm2, %xmm0
-; SSE2-NEXT: movmskps %xmm0, %eax
-; SSE2-NEXT: testb $14, %al
-; SSE2-NEXT: setne %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: movmsk_v4f32:
-; AVX: # %bb.0:
-; AVX-NEXT: vcmpeq_uqps %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vmovmskps %xmm0, %eax
-; AVX-NEXT: testb $14, %al
-; AVX-NEXT: setne %al
-; AVX-NEXT: retq
+; SSE-LABEL: movmsk_v4f32:
+; SSE: # %bb.0:
+; SSE-NEXT: movaps %xmm0, %xmm2
+; SSE-NEXT: cmpeqps %xmm1, %xmm2
+; SSE-NEXT: cmpunordps %xmm1, %xmm0
+; SSE-NEXT: orps %xmm2, %xmm0
+; SSE-NEXT: movmskps %xmm0, %eax
+; SSE-NEXT: testb $14, %al
+; SSE-NEXT: setne %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: movmsk_v4f32:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vcmpeq_uqps %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT: vmovmskps %xmm0, %eax
+; AVX1OR2-NEXT: testb $14, %al
+; AVX1OR2-NEXT: setne %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: movmsk_v4f32:
; KNL: # %bb.0:
@@ -4602,21 +4417,21 @@ define i1 @movmsk_v4f32(<4 x float> %x, <4 x float> %y) {
}
define i1 @movmsk_and_v2f64(<2 x double> %x, <2 x double> %y) {
-; SSE2-LABEL: movmsk_and_v2f64:
-; SSE2: # %bb.0:
-; SSE2-NEXT: cmplepd %xmm0, %xmm1
-; SSE2-NEXT: movmskpd %xmm1, %eax
-; SSE2-NEXT: cmpb $3, %al
-; SSE2-NEXT: sete %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: movmsk_and_v2f64:
-; AVX: # %bb.0:
-; AVX-NEXT: vcmplepd %xmm0, %xmm1, %xmm0
-; AVX-NEXT: vmovmskpd %xmm0, %eax
-; AVX-NEXT: cmpb $3, %al
-; AVX-NEXT: sete %al
-; AVX-NEXT: retq
+; SSE-LABEL: movmsk_and_v2f64:
+; SSE: # %bb.0:
+; SSE-NEXT: cmplepd %xmm0, %xmm1
+; SSE-NEXT: movmskpd %xmm1, %eax
+; SSE-NEXT: cmpb $3, %al
+; SSE-NEXT: sete %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: movmsk_and_v2f64:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vcmplepd %xmm0, %xmm1, %xmm0
+; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
+; AVX1OR2-NEXT: cmpb $3, %al
+; AVX1OR2-NEXT: sete %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: movmsk_and_v2f64:
; KNL: # %bb.0:
@@ -4645,21 +4460,21 @@ define i1 @movmsk_and_v2f64(<2 x double> %x, <2 x double> %y) {
}
define i1 @movmsk_or_v2f64(<2 x double> %x, <2 x double> %y) {
-; SSE2-LABEL: movmsk_or_v2f64:
-; SSE2: # %bb.0:
-; SSE2-NEXT: cmplepd %xmm0, %xmm1
-; SSE2-NEXT: movmskpd %xmm1, %eax
-; SSE2-NEXT: testb %al, %al
-; SSE2-NEXT: setne %al
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: movmsk_or_v2f64:
-; AVX: # %bb.0:
-; AVX-NEXT: vcmplepd %xmm0, %xmm1, %xmm0
-; AVX-NEXT: vmovmskpd %xmm0, %eax
-; AVX-NEXT: testb %al, %al
-; AVX-NEXT: setne %al
-; AVX-NEXT: retq
+; SSE-LABEL: movmsk_or_v2f64:
+; SSE: # %bb.0:
+; SSE-NEXT: cmplepd %xmm0, %xmm1
+; SSE-NEXT: movmskpd %xmm1, %eax
+; SSE-NEXT: testb %al, %al
+; SSE-NEXT: setne %al
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: movmsk_or_v2f64:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vcmplepd %xmm0, %xmm1, %xmm0
+; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
+; AVX1OR2-NEXT: testb %al, %al
+; AVX1OR2-NEXT: setne %al
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: movmsk_or_v2f64:
; KNL: # %bb.0:
@@ -4686,25 +4501,25 @@ define i1 @movmsk_or_v2f64(<2 x double> %x, <2 x double> %y) {
}
define i32 @PR39665_c_ray(<2 x double> %x, <2 x double> %y) {
-; SSE2-LABEL: PR39665_c_ray:
-; SSE2: # %bb.0:
-; SSE2-NEXT: cmpltpd %xmm0, %xmm1
-; SSE2-NEXT: movmskpd %xmm1, %eax
-; SSE2-NEXT: cmpb $3, %al
-; SSE2-NEXT: movl $42, %ecx
-; SSE2-NEXT: movl $99, %eax
-; SSE2-NEXT: cmovel %ecx, %eax
-; SSE2-NEXT: retq
-;
-; AVX-LABEL: PR39665_c_ray:
-; AVX: # %bb.0:
-; AVX-NEXT: vcmpltpd %xmm0, %xmm1, %xmm0
-; AVX-NEXT: vmovmskpd %xmm0, %eax
-; AVX-NEXT: cmpb $3, %al
-; AVX-NEXT: movl $42, %ecx
-; AVX-NEXT: movl $99, %eax
-; AVX-NEXT: cmovel %ecx, %eax
-; AVX-NEXT: retq
+; SSE-LABEL: PR39665_c_ray:
+; SSE: # %bb.0:
+; SSE-NEXT: cmpltpd %xmm0, %xmm1
+; SSE-NEXT: movmskpd %xmm1, %eax
+; SSE-NEXT: cmpb $3, %al
+; SSE-NEXT: movl $42, %ecx
+; SSE-NEXT: movl $99, %eax
+; SSE-NEXT: cmovel %ecx, %eax
+; SSE-NEXT: retq
+;
+; AVX1OR2-LABEL: PR39665_c_ray:
+; AVX1OR2: # %bb.0:
+; AVX1OR2-NEXT: vcmpltpd %xmm0, %xmm1, %xmm0
+; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
+; AVX1OR2-NEXT: cmpb $3, %al
+; AVX1OR2-NEXT: movl $42, %ecx
+; AVX1OR2-NEXT: movl $99, %eax
+; AVX1OR2-NEXT: cmovel %ecx, %eax
+; AVX1OR2-NEXT: retq
;
; KNL-LABEL: PR39665_c_ray:
; KNL: # %bb.0:
More information about the llvm-commits
mailing list