[llvm] [X86] bitcnt-big-integer.ll - add test coverage for AVX512 targets with no VLX support (PR #171104)
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Mon Dec 8 02:07:52 PST 2025
https://github.com/RKSimon created https://github.com/llvm/llvm-project/pull/171104
None
>From 5d1339c88384afef80a0337d922fd78c3d150a48 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev at redking.me.uk>
Date: Mon, 8 Dec 2025 10:06:13 +0000
Subject: [PATCH] [X86] bitcnt-big-integer.ll - add test coverage for AVX512
targets with no VLX support
---
llvm/test/CodeGen/X86/bitcnt-big-integer.ll | 1976 +++++++++++++++----
1 file changed, 1563 insertions(+), 413 deletions(-)
diff --git a/llvm/test/CodeGen/X86/bitcnt-big-integer.ll b/llvm/test/CodeGen/X86/bitcnt-big-integer.ll
index 5b3b27a3d61de..0fd555991ae29 100644
--- a/llvm/test/CodeGen/X86/bitcnt-big-integer.ll
+++ b/llvm/test/CodeGen/X86/bitcnt-big-integer.ll
@@ -1,8 +1,9 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=CHECK,SSE
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=CHECK,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 -mattr=+avx512vpopcntdq | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512POPCNT
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 -mattr=+avx512vpopcntdq | FileCheck %s --check-prefixes=AVX512,AVX512POPCNT
;
; CTPOP
@@ -16,6 +17,14 @@ define i32 @test_ctpop_i128(i128 %a0) nounwind {
; CHECK-NEXT: addl %ecx, %eax
; CHECK-NEXT: # kill: def $eax killed $eax killed $rax
; CHECK-NEXT: retq
+;
+; AVX512-LABEL: test_ctpop_i128:
+; AVX512: # %bb.0:
+; AVX512-NEXT: popcntq %rsi, %rcx
+; AVX512-NEXT: popcntq %rdi, %rax
+; AVX512-NEXT: addl %ecx, %eax
+; AVX512-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512-NEXT: retq
%cnt = call i128 @llvm.ctpop.i128(i128 %a0)
%res = trunc i128 %cnt to i32
ret i32 %res
@@ -29,6 +38,14 @@ define i32 @load_ctpop_i128(ptr %p0) nounwind {
; CHECK-NEXT: addl %ecx, %eax
; CHECK-NEXT: # kill: def $eax killed $eax killed $rax
; CHECK-NEXT: retq
+;
+; AVX512-LABEL: load_ctpop_i128:
+; AVX512: # %bb.0:
+; AVX512-NEXT: popcntq 8(%rdi), %rcx
+; AVX512-NEXT: popcntq (%rdi), %rax
+; AVX512-NEXT: addl %ecx, %eax
+; AVX512-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512-NEXT: retq
%a0 = load i128, ptr %p0
%cnt = call i128 @llvm.ctpop.i128(i128 %a0)
%res = trunc i128 %cnt to i32
@@ -50,6 +67,48 @@ define i32 @test_ctpop_i256(i256 %a0) nounwind {
; CHECK-NEXT: addl %ecx, %eax
; CHECK-NEXT: # kill: def $eax killed $eax killed $rax
; CHECK-NEXT: retq
+;
+; AVX512F-LABEL: test_ctpop_i256:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: popcntq %rcx, %rax
+; AVX512F-NEXT: popcntq %rdx, %rcx
+; AVX512F-NEXT: addl %eax, %ecx
+; AVX512F-NEXT: popcntq %rsi, %rdx
+; AVX512F-NEXT: popcntq %rdi, %rax
+; AVX512F-NEXT: addl %edx, %eax
+; AVX512F-NEXT: addl %ecx, %eax
+; AVX512F-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: test_ctpop_i256:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: popcntq %rcx, %rax
+; AVX512VL-NEXT: xorl %ecx, %ecx
+; AVX512VL-NEXT: popcntq %rdx, %rcx
+; AVX512VL-NEXT: addl %eax, %ecx
+; AVX512VL-NEXT: xorl %edx, %edx
+; AVX512VL-NEXT: popcntq %rsi, %rdx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq %rdi, %rax
+; AVX512VL-NEXT: addl %edx, %eax
+; AVX512VL-NEXT: addl %ecx, %eax
+; AVX512VL-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: test_ctpop_i256:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: popcntq %rcx, %rax
+; AVX512POPCNT-NEXT: xorl %ecx, %ecx
+; AVX512POPCNT-NEXT: popcntq %rdx, %rcx
+; AVX512POPCNT-NEXT: addl %eax, %ecx
+; AVX512POPCNT-NEXT: xorl %edx, %edx
+; AVX512POPCNT-NEXT: popcntq %rsi, %rdx
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq %rdi, %rax
+; AVX512POPCNT-NEXT: addl %edx, %eax
+; AVX512POPCNT-NEXT: addl %ecx, %eax
+; AVX512POPCNT-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512POPCNT-NEXT: retq
%cnt = call i256 @llvm.ctpop.i256(i256 %a0)
%res = trunc i256 %cnt to i32
ret i32 %res
@@ -81,18 +140,43 @@ define i32 @load_ctpop_i256(ptr %p0) nounwind {
; AVX2-NEXT: # kill: def $eax killed $eax killed $rax
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_ctpop_i256:
-; AVX512: # %bb.0:
-; AVX512-NEXT: popcntq 24(%rdi), %rax
-; AVX512-NEXT: popcntq 16(%rdi), %rcx
-; AVX512-NEXT: addl %eax, %ecx
-; AVX512-NEXT: popcntq 8(%rdi), %rdx
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq (%rdi), %rax
-; AVX512-NEXT: addl %edx, %eax
-; AVX512-NEXT: addl %ecx, %eax
-; AVX512-NEXT: # kill: def $eax killed $eax killed $rax
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_ctpop_i256:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: popcntq 24(%rdi), %rax
+; AVX512F-NEXT: popcntq 16(%rdi), %rcx
+; AVX512F-NEXT: addl %eax, %ecx
+; AVX512F-NEXT: popcntq 8(%rdi), %rdx
+; AVX512F-NEXT: popcntq (%rdi), %rax
+; AVX512F-NEXT: addl %edx, %eax
+; AVX512F-NEXT: addl %ecx, %eax
+; AVX512F-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_ctpop_i256:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: popcntq 24(%rdi), %rax
+; AVX512VL-NEXT: popcntq 16(%rdi), %rcx
+; AVX512VL-NEXT: addl %eax, %ecx
+; AVX512VL-NEXT: popcntq 8(%rdi), %rdx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq (%rdi), %rax
+; AVX512VL-NEXT: addl %edx, %eax
+; AVX512VL-NEXT: addl %ecx, %eax
+; AVX512VL-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: load_ctpop_i256:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: popcntq 24(%rdi), %rax
+; AVX512POPCNT-NEXT: popcntq 16(%rdi), %rcx
+; AVX512POPCNT-NEXT: addl %eax, %ecx
+; AVX512POPCNT-NEXT: popcntq 8(%rdi), %rdx
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq (%rdi), %rax
+; AVX512POPCNT-NEXT: addl %edx, %eax
+; AVX512POPCNT-NEXT: addl %ecx, %eax
+; AVX512POPCNT-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512POPCNT-NEXT: retq
%a0 = load i256, ptr %p0
%cnt = call i256 @llvm.ctpop.i256(i256 %a0)
%res = trunc i256 %cnt to i32
@@ -124,6 +208,76 @@ define i32 @test_ctpop_i512(i512 %a0) nounwind {
; CHECK-NEXT: addl %r8d, %eax
; CHECK-NEXT: # kill: def $eax killed $eax killed $rax
; CHECK-NEXT: retq
+;
+; AVX512F-LABEL: test_ctpop_i512:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
+; AVX512F-NEXT: addl %eax, %r10d
+; AVX512F-NEXT: popcntq %r9, %rax
+; AVX512F-NEXT: popcntq %r8, %r8
+; AVX512F-NEXT: addl %eax, %r8d
+; AVX512F-NEXT: addl %r10d, %r8d
+; AVX512F-NEXT: popcntq %rcx, %rax
+; AVX512F-NEXT: popcntq %rdx, %rcx
+; AVX512F-NEXT: addl %eax, %ecx
+; AVX512F-NEXT: popcntq %rsi, %rdx
+; AVX512F-NEXT: popcntq %rdi, %rax
+; AVX512F-NEXT: addl %edx, %eax
+; AVX512F-NEXT: addl %ecx, %eax
+; AVX512F-NEXT: addl %r8d, %eax
+; AVX512F-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: test_ctpop_i512:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
+; AVX512VL-NEXT: addl %eax, %r10d
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq %r9, %rax
+; AVX512VL-NEXT: popcntq %r8, %r8
+; AVX512VL-NEXT: addl %eax, %r8d
+; AVX512VL-NEXT: addl %r10d, %r8d
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq %rcx, %rax
+; AVX512VL-NEXT: xorl %ecx, %ecx
+; AVX512VL-NEXT: popcntq %rdx, %rcx
+; AVX512VL-NEXT: addl %eax, %ecx
+; AVX512VL-NEXT: xorl %edx, %edx
+; AVX512VL-NEXT: popcntq %rsi, %rdx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq %rdi, %rax
+; AVX512VL-NEXT: addl %edx, %eax
+; AVX512VL-NEXT: addl %ecx, %eax
+; AVX512VL-NEXT: addl %r8d, %eax
+; AVX512VL-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: test_ctpop_i512:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
+; AVX512POPCNT-NEXT: addl %eax, %r10d
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq %r9, %rax
+; AVX512POPCNT-NEXT: popcntq %r8, %r8
+; AVX512POPCNT-NEXT: addl %eax, %r8d
+; AVX512POPCNT-NEXT: addl %r10d, %r8d
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq %rcx, %rax
+; AVX512POPCNT-NEXT: xorl %ecx, %ecx
+; AVX512POPCNT-NEXT: popcntq %rdx, %rcx
+; AVX512POPCNT-NEXT: addl %eax, %ecx
+; AVX512POPCNT-NEXT: xorl %edx, %edx
+; AVX512POPCNT-NEXT: popcntq %rsi, %rdx
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq %rdi, %rax
+; AVX512POPCNT-NEXT: addl %edx, %eax
+; AVX512POPCNT-NEXT: addl %ecx, %eax
+; AVX512POPCNT-NEXT: addl %r8d, %eax
+; AVX512POPCNT-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512POPCNT-NEXT: retq
%cnt = call i512 @llvm.ctpop.i512(i512 %a0)
%res = trunc i512 %cnt to i32
ret i32 %res
@@ -177,29 +331,73 @@ define i32 @load_ctpop_i512(ptr %p0) nounwind {
; AVX2-NEXT: # kill: def $eax killed $eax killed $rax
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_ctpop_i512:
-; AVX512: # %bb.0:
-; AVX512-NEXT: popcntq 56(%rdi), %rax
-; AVX512-NEXT: popcntq 48(%rdi), %rcx
-; AVX512-NEXT: addl %eax, %ecx
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq 40(%rdi), %rax
-; AVX512-NEXT: popcntq 32(%rdi), %rdx
-; AVX512-NEXT: addl %eax, %edx
-; AVX512-NEXT: addl %ecx, %edx
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq 24(%rdi), %rax
-; AVX512-NEXT: xorl %ecx, %ecx
-; AVX512-NEXT: popcntq 16(%rdi), %rcx
-; AVX512-NEXT: popcntq 8(%rdi), %rsi
-; AVX512-NEXT: addl %eax, %ecx
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq (%rdi), %rax
-; AVX512-NEXT: addl %esi, %eax
-; AVX512-NEXT: addl %ecx, %eax
-; AVX512-NEXT: addl %edx, %eax
-; AVX512-NEXT: # kill: def $eax killed $eax killed $rax
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_ctpop_i512:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: popcntq 56(%rdi), %rax
+; AVX512F-NEXT: popcntq 48(%rdi), %rcx
+; AVX512F-NEXT: addl %eax, %ecx
+; AVX512F-NEXT: popcntq 40(%rdi), %rax
+; AVX512F-NEXT: popcntq 32(%rdi), %rdx
+; AVX512F-NEXT: addl %eax, %edx
+; AVX512F-NEXT: addl %ecx, %edx
+; AVX512F-NEXT: popcntq 24(%rdi), %rcx
+; AVX512F-NEXT: popcntq 16(%rdi), %rsi
+; AVX512F-NEXT: popcntq 8(%rdi), %r8
+; AVX512F-NEXT: popcntq (%rdi), %rax
+; AVX512F-NEXT: addl %ecx, %esi
+; AVX512F-NEXT: addl %r8d, %eax
+; AVX512F-NEXT: addl %esi, %eax
+; AVX512F-NEXT: addl %edx, %eax
+; AVX512F-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_ctpop_i512:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: popcntq 56(%rdi), %rax
+; AVX512VL-NEXT: popcntq 48(%rdi), %rcx
+; AVX512VL-NEXT: addl %eax, %ecx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq 40(%rdi), %rax
+; AVX512VL-NEXT: popcntq 32(%rdi), %rdx
+; AVX512VL-NEXT: addl %eax, %edx
+; AVX512VL-NEXT: addl %ecx, %edx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq 24(%rdi), %rax
+; AVX512VL-NEXT: xorl %ecx, %ecx
+; AVX512VL-NEXT: popcntq 16(%rdi), %rcx
+; AVX512VL-NEXT: popcntq 8(%rdi), %rsi
+; AVX512VL-NEXT: addl %eax, %ecx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq (%rdi), %rax
+; AVX512VL-NEXT: addl %esi, %eax
+; AVX512VL-NEXT: addl %ecx, %eax
+; AVX512VL-NEXT: addl %edx, %eax
+; AVX512VL-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: load_ctpop_i512:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: popcntq 56(%rdi), %rax
+; AVX512POPCNT-NEXT: popcntq 48(%rdi), %rcx
+; AVX512POPCNT-NEXT: addl %eax, %ecx
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq 40(%rdi), %rax
+; AVX512POPCNT-NEXT: popcntq 32(%rdi), %rdx
+; AVX512POPCNT-NEXT: addl %eax, %edx
+; AVX512POPCNT-NEXT: addl %ecx, %edx
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq 24(%rdi), %rax
+; AVX512POPCNT-NEXT: xorl %ecx, %ecx
+; AVX512POPCNT-NEXT: popcntq 16(%rdi), %rcx
+; AVX512POPCNT-NEXT: popcntq 8(%rdi), %rsi
+; AVX512POPCNT-NEXT: addl %eax, %ecx
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq (%rdi), %rax
+; AVX512POPCNT-NEXT: addl %esi, %eax
+; AVX512POPCNT-NEXT: addl %ecx, %eax
+; AVX512POPCNT-NEXT: addl %edx, %eax
+; AVX512POPCNT-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512POPCNT-NEXT: retq
%a0 = load i512, ptr %p0
%cnt = call i512 @llvm.ctpop.i512(i512 %a0)
%res = trunc i512 %cnt to i32
@@ -309,57 +507,149 @@ define i32 @test_ctpop_i1024(i1024 %a0) nounwind {
; AVX2-NEXT: popq %r14
; AVX2-NEXT: retq
;
-; AVX512-LABEL: test_ctpop_i1024:
-; AVX512: # %bb.0:
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
-; AVX512-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
-; AVX512-NEXT: addl %eax, %r10d
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
-; AVX512-NEXT: popcntq {{[0-9]+}}(%rsp), %r11
-; AVX512-NEXT: addl %eax, %r11d
-; AVX512-NEXT: addl %r10d, %r11d
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
-; AVX512-NEXT: xorl %ebx, %ebx
-; AVX512-NEXT: popcntq {{[0-9]+}}(%rsp), %rbx
-; AVX512-NEXT: xorl %r14d, %r14d
-; AVX512-NEXT: popcntq {{[0-9]+}}(%rsp), %r14
-; AVX512-NEXT: addl %eax, %ebx
-; AVX512-NEXT: xorl %r10d, %r10d
-; AVX512-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
-; AVX512-NEXT: addl %r14d, %r10d
-; AVX512-NEXT: addl %ebx, %r10d
-; AVX512-NEXT: addl %r11d, %r10d
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
-; AVX512-NEXT: xorl %r11d, %r11d
-; AVX512-NEXT: popcntq {{[0-9]+}}(%rsp), %r11
-; AVX512-NEXT: addl %eax, %r11d
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq %r9, %rax
-; AVX512-NEXT: popcntq %r8, %r8
-; AVX512-NEXT: addl %eax, %r8d
-; AVX512-NEXT: addl %r11d, %r8d
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq %rcx, %rax
-; AVX512-NEXT: xorl %ecx, %ecx
-; AVX512-NEXT: popcntq %rdx, %rcx
-; AVX512-NEXT: addl %eax, %ecx
-; AVX512-NEXT: xorl %edx, %edx
-; AVX512-NEXT: popcntq %rsi, %rdx
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq %rdi, %rax
-; AVX512-NEXT: addl %edx, %eax
-; AVX512-NEXT: addl %ecx, %eax
-; AVX512-NEXT: addl %r8d, %eax
-; AVX512-NEXT: addl %r10d, %eax
-; AVX512-NEXT: # kill: def $eax killed $eax killed $rax
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: retq
+; AVX512F-LABEL: test_ctpop_i1024:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
+; AVX512F-NEXT: addl %eax, %r10d
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %r11
+; AVX512F-NEXT: addl %eax, %r11d
+; AVX512F-NEXT: addl %r10d, %r11d
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %rbx
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %r14
+; AVX512F-NEXT: addl %eax, %ebx
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
+; AVX512F-NEXT: addl %r14d, %r10d
+; AVX512F-NEXT: addl %ebx, %r10d
+; AVX512F-NEXT: addl %r11d, %r10d
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512F-NEXT: popcntq {{[0-9]+}}(%rsp), %r11
+; AVX512F-NEXT: addl %eax, %r11d
+; AVX512F-NEXT: popcntq %r9, %rax
+; AVX512F-NEXT: popcntq %r8, %r8
+; AVX512F-NEXT: addl %eax, %r8d
+; AVX512F-NEXT: addl %r11d, %r8d
+; AVX512F-NEXT: popcntq %rcx, %rax
+; AVX512F-NEXT: popcntq %rdx, %rcx
+; AVX512F-NEXT: addl %eax, %ecx
+; AVX512F-NEXT: popcntq %rsi, %rdx
+; AVX512F-NEXT: popcntq %rdi, %rax
+; AVX512F-NEXT: addl %edx, %eax
+; AVX512F-NEXT: addl %ecx, %eax
+; AVX512F-NEXT: addl %r8d, %eax
+; AVX512F-NEXT: addl %r10d, %eax
+; AVX512F-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: test_ctpop_i1024:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
+; AVX512VL-NEXT: addl %eax, %r10d
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %r11
+; AVX512VL-NEXT: addl %eax, %r11d
+; AVX512VL-NEXT: addl %r10d, %r11d
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512VL-NEXT: xorl %ebx, %ebx
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %rbx
+; AVX512VL-NEXT: xorl %r14d, %r14d
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %r14
+; AVX512VL-NEXT: addl %eax, %ebx
+; AVX512VL-NEXT: xorl %r10d, %r10d
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
+; AVX512VL-NEXT: addl %r14d, %r10d
+; AVX512VL-NEXT: addl %ebx, %r10d
+; AVX512VL-NEXT: addl %r11d, %r10d
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512VL-NEXT: xorl %r11d, %r11d
+; AVX512VL-NEXT: popcntq {{[0-9]+}}(%rsp), %r11
+; AVX512VL-NEXT: addl %eax, %r11d
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq %r9, %rax
+; AVX512VL-NEXT: popcntq %r8, %r8
+; AVX512VL-NEXT: addl %eax, %r8d
+; AVX512VL-NEXT: addl %r11d, %r8d
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq %rcx, %rax
+; AVX512VL-NEXT: xorl %ecx, %ecx
+; AVX512VL-NEXT: popcntq %rdx, %rcx
+; AVX512VL-NEXT: addl %eax, %ecx
+; AVX512VL-NEXT: xorl %edx, %edx
+; AVX512VL-NEXT: popcntq %rsi, %rdx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq %rdi, %rax
+; AVX512VL-NEXT: addl %edx, %eax
+; AVX512VL-NEXT: addl %ecx, %eax
+; AVX512VL-NEXT: addl %r8d, %eax
+; AVX512VL-NEXT: addl %r10d, %eax
+; AVX512VL-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: test_ctpop_i1024:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: pushq %r14
+; AVX512POPCNT-NEXT: pushq %rbx
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
+; AVX512POPCNT-NEXT: addl %eax, %r10d
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %r11
+; AVX512POPCNT-NEXT: addl %eax, %r11d
+; AVX512POPCNT-NEXT: addl %r10d, %r11d
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512POPCNT-NEXT: xorl %ebx, %ebx
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %rbx
+; AVX512POPCNT-NEXT: xorl %r14d, %r14d
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %r14
+; AVX512POPCNT-NEXT: addl %eax, %ebx
+; AVX512POPCNT-NEXT: xorl %r10d, %r10d
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %r10
+; AVX512POPCNT-NEXT: addl %r14d, %r10d
+; AVX512POPCNT-NEXT: addl %ebx, %r10d
+; AVX512POPCNT-NEXT: addl %r11d, %r10d
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %rax
+; AVX512POPCNT-NEXT: xorl %r11d, %r11d
+; AVX512POPCNT-NEXT: popcntq {{[0-9]+}}(%rsp), %r11
+; AVX512POPCNT-NEXT: addl %eax, %r11d
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq %r9, %rax
+; AVX512POPCNT-NEXT: popcntq %r8, %r8
+; AVX512POPCNT-NEXT: addl %eax, %r8d
+; AVX512POPCNT-NEXT: addl %r11d, %r8d
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq %rcx, %rax
+; AVX512POPCNT-NEXT: xorl %ecx, %ecx
+; AVX512POPCNT-NEXT: popcntq %rdx, %rcx
+; AVX512POPCNT-NEXT: addl %eax, %ecx
+; AVX512POPCNT-NEXT: xorl %edx, %edx
+; AVX512POPCNT-NEXT: popcntq %rsi, %rdx
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq %rdi, %rax
+; AVX512POPCNT-NEXT: addl %edx, %eax
+; AVX512POPCNT-NEXT: addl %ecx, %eax
+; AVX512POPCNT-NEXT: addl %r8d, %eax
+; AVX512POPCNT-NEXT: addl %r10d, %eax
+; AVX512POPCNT-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512POPCNT-NEXT: popq %rbx
+; AVX512POPCNT-NEXT: popq %r14
+; AVX512POPCNT-NEXT: retq
%cnt = call i1024 @llvm.ctpop.i1024(i1024 %a0)
%res = trunc i1024 %cnt to i32
ret i32 %res
@@ -460,52 +750,135 @@ define i32 @load_ctpop_i1024(ptr %p0) nounwind {
; AVX2-NEXT: # kill: def $eax killed $eax killed $rax
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_ctpop_i1024:
-; AVX512: # %bb.0:
-; AVX512-NEXT: popcntq 120(%rdi), %rax
-; AVX512-NEXT: popcntq 112(%rdi), %rcx
-; AVX512-NEXT: addl %eax, %ecx
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq 104(%rdi), %rax
-; AVX512-NEXT: popcntq 96(%rdi), %rdx
-; AVX512-NEXT: addl %eax, %edx
-; AVX512-NEXT: addl %ecx, %edx
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq 88(%rdi), %rax
-; AVX512-NEXT: popcntq 80(%rdi), %rsi
-; AVX512-NEXT: popcntq 72(%rdi), %r8
-; AVX512-NEXT: addl %eax, %esi
-; AVX512-NEXT: xorl %ecx, %ecx
-; AVX512-NEXT: popcntq 64(%rdi), %rcx
-; AVX512-NEXT: addl %r8d, %ecx
-; AVX512-NEXT: addl %esi, %ecx
-; AVX512-NEXT: addl %edx, %ecx
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq 56(%rdi), %rax
-; AVX512-NEXT: xorl %edx, %edx
-; AVX512-NEXT: popcntq 48(%rdi), %rdx
-; AVX512-NEXT: xorl %esi, %esi
-; AVX512-NEXT: popcntq 40(%rdi), %rsi
-; AVX512-NEXT: addl %eax, %edx
-; AVX512-NEXT: xorl %r8d, %r8d
-; AVX512-NEXT: popcntq 32(%rdi), %r8
-; AVX512-NEXT: addl %esi, %r8d
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq 24(%rdi), %rax
-; AVX512-NEXT: addl %edx, %r8d
-; AVX512-NEXT: xorl %edx, %edx
-; AVX512-NEXT: popcntq 16(%rdi), %rdx
-; AVX512-NEXT: addl %eax, %edx
-; AVX512-NEXT: xorl %esi, %esi
-; AVX512-NEXT: popcntq 8(%rdi), %rsi
-; AVX512-NEXT: xorl %eax, %eax
-; AVX512-NEXT: popcntq (%rdi), %rax
-; AVX512-NEXT: addl %esi, %eax
-; AVX512-NEXT: addl %edx, %eax
-; AVX512-NEXT: addl %r8d, %eax
-; AVX512-NEXT: addl %ecx, %eax
-; AVX512-NEXT: # kill: def $eax killed $eax killed $rax
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_ctpop_i1024:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: popcntq 120(%rdi), %rax
+; AVX512F-NEXT: popcntq 112(%rdi), %rcx
+; AVX512F-NEXT: addl %eax, %ecx
+; AVX512F-NEXT: popcntq 104(%rdi), %rax
+; AVX512F-NEXT: popcntq 96(%rdi), %rdx
+; AVX512F-NEXT: addl %eax, %edx
+; AVX512F-NEXT: addl %ecx, %edx
+; AVX512F-NEXT: popcntq 88(%rdi), %rax
+; AVX512F-NEXT: popcntq 80(%rdi), %rsi
+; AVX512F-NEXT: popcntq 72(%rdi), %r8
+; AVX512F-NEXT: popcntq 64(%rdi), %rcx
+; AVX512F-NEXT: addl %eax, %esi
+; AVX512F-NEXT: addl %r8d, %ecx
+; AVX512F-NEXT: addl %esi, %ecx
+; AVX512F-NEXT: addl %edx, %ecx
+; AVX512F-NEXT: popcntq 56(%rdi), %rax
+; AVX512F-NEXT: popcntq 48(%rdi), %rdx
+; AVX512F-NEXT: popcntq 40(%rdi), %rsi
+; AVX512F-NEXT: popcntq 32(%rdi), %r8
+; AVX512F-NEXT: addl %eax, %edx
+; AVX512F-NEXT: addl %esi, %r8d
+; AVX512F-NEXT: popcntq 24(%rdi), %rax
+; AVX512F-NEXT: addl %edx, %r8d
+; AVX512F-NEXT: popcntq 16(%rdi), %rdx
+; AVX512F-NEXT: addl %eax, %edx
+; AVX512F-NEXT: popcntq 8(%rdi), %rsi
+; AVX512F-NEXT: popcntq (%rdi), %rax
+; AVX512F-NEXT: addl %esi, %eax
+; AVX512F-NEXT: addl %edx, %eax
+; AVX512F-NEXT: addl %r8d, %eax
+; AVX512F-NEXT: addl %ecx, %eax
+; AVX512F-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_ctpop_i1024:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: popcntq 120(%rdi), %rax
+; AVX512VL-NEXT: popcntq 112(%rdi), %rcx
+; AVX512VL-NEXT: addl %eax, %ecx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq 104(%rdi), %rax
+; AVX512VL-NEXT: popcntq 96(%rdi), %rdx
+; AVX512VL-NEXT: addl %eax, %edx
+; AVX512VL-NEXT: addl %ecx, %edx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq 88(%rdi), %rax
+; AVX512VL-NEXT: popcntq 80(%rdi), %rsi
+; AVX512VL-NEXT: popcntq 72(%rdi), %r8
+; AVX512VL-NEXT: addl %eax, %esi
+; AVX512VL-NEXT: xorl %ecx, %ecx
+; AVX512VL-NEXT: popcntq 64(%rdi), %rcx
+; AVX512VL-NEXT: addl %r8d, %ecx
+; AVX512VL-NEXT: addl %esi, %ecx
+; AVX512VL-NEXT: addl %edx, %ecx
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq 56(%rdi), %rax
+; AVX512VL-NEXT: xorl %edx, %edx
+; AVX512VL-NEXT: popcntq 48(%rdi), %rdx
+; AVX512VL-NEXT: xorl %esi, %esi
+; AVX512VL-NEXT: popcntq 40(%rdi), %rsi
+; AVX512VL-NEXT: addl %eax, %edx
+; AVX512VL-NEXT: xorl %r8d, %r8d
+; AVX512VL-NEXT: popcntq 32(%rdi), %r8
+; AVX512VL-NEXT: addl %esi, %r8d
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq 24(%rdi), %rax
+; AVX512VL-NEXT: addl %edx, %r8d
+; AVX512VL-NEXT: xorl %edx, %edx
+; AVX512VL-NEXT: popcntq 16(%rdi), %rdx
+; AVX512VL-NEXT: addl %eax, %edx
+; AVX512VL-NEXT: xorl %esi, %esi
+; AVX512VL-NEXT: popcntq 8(%rdi), %rsi
+; AVX512VL-NEXT: xorl %eax, %eax
+; AVX512VL-NEXT: popcntq (%rdi), %rax
+; AVX512VL-NEXT: addl %esi, %eax
+; AVX512VL-NEXT: addl %edx, %eax
+; AVX512VL-NEXT: addl %r8d, %eax
+; AVX512VL-NEXT: addl %ecx, %eax
+; AVX512VL-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: load_ctpop_i1024:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: popcntq 120(%rdi), %rax
+; AVX512POPCNT-NEXT: popcntq 112(%rdi), %rcx
+; AVX512POPCNT-NEXT: addl %eax, %ecx
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq 104(%rdi), %rax
+; AVX512POPCNT-NEXT: popcntq 96(%rdi), %rdx
+; AVX512POPCNT-NEXT: addl %eax, %edx
+; AVX512POPCNT-NEXT: addl %ecx, %edx
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq 88(%rdi), %rax
+; AVX512POPCNT-NEXT: popcntq 80(%rdi), %rsi
+; AVX512POPCNT-NEXT: popcntq 72(%rdi), %r8
+; AVX512POPCNT-NEXT: addl %eax, %esi
+; AVX512POPCNT-NEXT: xorl %ecx, %ecx
+; AVX512POPCNT-NEXT: popcntq 64(%rdi), %rcx
+; AVX512POPCNT-NEXT: addl %r8d, %ecx
+; AVX512POPCNT-NEXT: addl %esi, %ecx
+; AVX512POPCNT-NEXT: addl %edx, %ecx
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq 56(%rdi), %rax
+; AVX512POPCNT-NEXT: xorl %edx, %edx
+; AVX512POPCNT-NEXT: popcntq 48(%rdi), %rdx
+; AVX512POPCNT-NEXT: xorl %esi, %esi
+; AVX512POPCNT-NEXT: popcntq 40(%rdi), %rsi
+; AVX512POPCNT-NEXT: addl %eax, %edx
+; AVX512POPCNT-NEXT: xorl %r8d, %r8d
+; AVX512POPCNT-NEXT: popcntq 32(%rdi), %r8
+; AVX512POPCNT-NEXT: addl %esi, %r8d
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq 24(%rdi), %rax
+; AVX512POPCNT-NEXT: addl %edx, %r8d
+; AVX512POPCNT-NEXT: xorl %edx, %edx
+; AVX512POPCNT-NEXT: popcntq 16(%rdi), %rdx
+; AVX512POPCNT-NEXT: addl %eax, %edx
+; AVX512POPCNT-NEXT: xorl %esi, %esi
+; AVX512POPCNT-NEXT: popcntq 8(%rdi), %rsi
+; AVX512POPCNT-NEXT: xorl %eax, %eax
+; AVX512POPCNT-NEXT: popcntq (%rdi), %rax
+; AVX512POPCNT-NEXT: addl %esi, %eax
+; AVX512POPCNT-NEXT: addl %edx, %eax
+; AVX512POPCNT-NEXT: addl %r8d, %eax
+; AVX512POPCNT-NEXT: addl %ecx, %eax
+; AVX512POPCNT-NEXT: # kill: def $eax killed $eax killed $rax
+; AVX512POPCNT-NEXT: retq
%a0 = load i1024, ptr %p0
%cnt = call i1024 @llvm.ctpop.i1024(i1024 %a0)
%res = trunc i1024 %cnt to i32
@@ -710,17 +1083,42 @@ define i32 @load_ctlz_i256(ptr %p0) nounwind {
; AVX2-NEXT: # kill: def $eax killed $eax killed $rax
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_ctlz_i256:
-; AVX512: # %bb.0:
-; AVX512-NEXT: vpermq {{.*#+}} ymm0 = mem[3,2,1,0]
-; AVX512-NEXT: vplzcntq %ymm0, %ymm1
-; AVX512-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
-; AVX512-NEXT: vptestmq %ymm0, %ymm0, %k1
-; AVX512-NEXT: vpbroadcastq {{.*#+}} ymm0 = [256,256,256,256]
-; AVX512-NEXT: vpcompressq %ymm1, %ymm0 {%k1}
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_ctlz_i256:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm0 = [256,256,256,256]
+; AVX512F-NEXT: vpermq {{.*#+}} ymm1 = mem[3,2,1,0]
+; AVX512F-NEXT: vplzcntq %zmm1, %zmm2
+; AVX512F-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT: vptestmq %zmm1, %zmm1, %k0
+; AVX512F-NEXT: kshiftlw $12, %k0, %k0
+; AVX512F-NEXT: kshiftrw $12, %k0, %k1
+; AVX512F-NEXT: vpcompressq %zmm2, %zmm0 {%k1}
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_ctlz_i256:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vpermq {{.*#+}} ymm0 = mem[3,2,1,0]
+; AVX512VL-NEXT: vplzcntq %ymm0, %ymm1
+; AVX512VL-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
+; AVX512VL-NEXT: vptestmq %ymm0, %ymm0, %k1
+; AVX512VL-NEXT: vpbroadcastq {{.*#+}} ymm0 = [256,256,256,256]
+; AVX512VL-NEXT: vpcompressq %ymm1, %ymm0 {%k1}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: load_ctlz_i256:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: vpermq {{.*#+}} ymm0 = mem[3,2,1,0]
+; AVX512POPCNT-NEXT: vplzcntq %ymm0, %ymm1
+; AVX512POPCNT-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
+; AVX512POPCNT-NEXT: vptestmq %ymm0, %ymm0, %k1
+; AVX512POPCNT-NEXT: vpbroadcastq {{.*#+}} ymm0 = [256,256,256,256]
+; AVX512POPCNT-NEXT: vpcompressq %ymm1, %ymm0 {%k1}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %eax
+; AVX512POPCNT-NEXT: vzeroupper
+; AVX512POPCNT-NEXT: retq
%a0 = load i256, ptr %p0
%cnt = call i256 @llvm.ctlz.i256(i256 %a0, i1 0)
%res = trunc i256 %cnt to i32
@@ -834,29 +1232,76 @@ define i32 @test_ctlz_i512(i512 %a0) nounwind {
; AVX2-NEXT: popq %r15
; AVX2-NEXT: retq
;
-; AVX512-LABEL: test_ctlz_i512:
-; AVX512: # %bb.0:
-; AVX512-NEXT: vmovq %rdi, %xmm0
-; AVX512-NEXT: vmovq %rsi, %xmm1
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; AVX512-NEXT: vmovq %rdx, %xmm1
-; AVX512-NEXT: vmovq %rcx, %xmm2
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
-; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
-; AVX512-NEXT: vmovq %r8, %xmm2
-; AVX512-NEXT: vmovq %r9, %xmm3
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
-; AVX512-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512-NEXT: vplzcntq %zmm0, %zmm1
-; AVX512-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
-; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
-; AVX512-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: test_ctlz_i512:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: vmovq %rdi, %xmm0
+; AVX512F-NEXT: vmovq %rsi, %xmm1
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: vmovq %rdx, %xmm1
+; AVX512F-NEXT: vmovq %rcx, %xmm2
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512F-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
+; AVX512F-NEXT: vmovq %r8, %xmm2
+; AVX512F-NEXT: vmovq %r9, %xmm3
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm0
+; AVX512F-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm1 = [512,512,512,512,512,512,512,512]
+; AVX512F-NEXT: vpcompressq %zmm0, %zmm1 {%k1}
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: test_ctlz_i512:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovq %rdi, %xmm0
+; AVX512VL-NEXT: vmovq %rsi, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovq %rdx, %xmm1
+; AVX512VL-NEXT: vmovq %rcx, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
+; AVX512VL-NEXT: vmovq %r8, %xmm2
+; AVX512VL-NEXT: vmovq %r9, %xmm3
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512VL-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: test_ctlz_i512:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: vmovq %rdi, %xmm0
+; AVX512POPCNT-NEXT: vmovq %rsi, %xmm1
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512POPCNT-NEXT: vmovq %rdx, %xmm1
+; AVX512POPCNT-NEXT: vmovq %rcx, %xmm2
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512POPCNT-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512POPCNT-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
+; AVX512POPCNT-NEXT: vmovq %r8, %xmm2
+; AVX512POPCNT-NEXT: vmovq %r9, %xmm3
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512POPCNT-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512POPCNT-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512POPCNT-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512POPCNT-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512POPCNT-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512POPCNT-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512POPCNT-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %eax
+; AVX512POPCNT-NEXT: vzeroupper
+; AVX512POPCNT-NEXT: retq
%cnt = call i512 @llvm.ctlz.i512(i512 %a0, i1 0)
%res = trunc i512 %cnt to i32
ret i32 %res
@@ -978,18 +1423,43 @@ define i32 @load_ctlz_i512(ptr %p0) nounwind {
; AVX2-NEXT: popq %r15
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_ctlz_i512:
-; AVX512: # %bb.0:
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
-; AVX512-NEXT: vpermq (%rdi), %zmm0, %zmm0
-; AVX512-NEXT: vplzcntq %zmm0, %zmm1
-; AVX512-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
-; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
-; AVX512-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_ctlz_i512:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512F-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm0
+; AVX512F-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm1 = [512,512,512,512,512,512,512,512]
+; AVX512F-NEXT: vpcompressq %zmm0, %zmm1 {%k1}
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_ctlz_i512:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512VL-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512VL-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: load_ctlz_i512:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512POPCNT-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512POPCNT-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512POPCNT-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512POPCNT-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512POPCNT-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512POPCNT-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %eax
+; AVX512POPCNT-NEXT: vzeroupper
+; AVX512POPCNT-NEXT: retq
%a0 = load i512, ptr %p0
%cnt = call i512 @llvm.ctlz.i512(i512 %a0, i1 0)
%res = trunc i512 %cnt to i32
@@ -1247,54 +1717,151 @@ define i32 @test_ctlz_i1024(i1024 %a0) nounwind {
; AVX2-NEXT: popq %rbp
; AVX2-NEXT: retq
;
-; AVX512-LABEL: test_ctlz_i1024:
-; AVX512: # %bb.0:
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: movq {{[0-9]+}}(%rsp), %rbx
-; AVX512-NEXT: movq {{[0-9]+}}(%rsp), %r10
-; AVX512-NEXT: movq {{[0-9]+}}(%rsp), %r11
-; AVX512-NEXT: movq {{[0-9]+}}(%rsp), %r14
-; AVX512-NEXT: vmovq %rdi, %xmm0
-; AVX512-NEXT: vmovq %rsi, %xmm1
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; AVX512-NEXT: vmovq %rdx, %xmm1
-; AVX512-NEXT: vmovq %rcx, %xmm2
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
-; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
-; AVX512-NEXT: vmovq %r8, %xmm2
-; AVX512-NEXT: vmovq %r9, %xmm3
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
-; AVX512-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512-NEXT: vplzcntq %zmm0, %zmm1
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,128,192,256,320,384,448]
-; AVX512-NEXT: vpaddq %zmm2, %zmm1, %zmm1
-; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
-; AVX512-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
-; AVX512-NEXT: vmovd %xmm0, %ecx
-; AVX512-NEXT: addl $512, %ecx # imm = 0x200
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
-; AVX512-NEXT: vpermq {{[0-9]+}}(%rsp), %zmm0, %zmm0
-; AVX512-NEXT: vplzcntq %zmm0, %zmm1
-; AVX512-NEXT: vpaddq %zmm2, %zmm1, %zmm1
-; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: orq {{[0-9]+}}(%rsp), %r14
-; AVX512-NEXT: orq {{[0-9]+}}(%rsp), %r11
-; AVX512-NEXT: orq {{[0-9]+}}(%rsp), %rbx
-; AVX512-NEXT: orq %r14, %r11
-; AVX512-NEXT: orq {{[0-9]+}}(%rsp), %r10
-; AVX512-NEXT: orq %rbx, %r10
-; AVX512-NEXT: orq %r11, %r10
-; AVX512-NEXT: cmovel %ecx, %eax
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: test_ctlz_i1024:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %rbx
+; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r10
+; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r11
+; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r14
+; AVX512F-NEXT: vmovq %rdi, %xmm0
+; AVX512F-NEXT: vmovq %rsi, %xmm1
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: vmovq %rdx, %xmm1
+; AVX512F-NEXT: vmovq %rcx, %xmm2
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512F-NEXT: vpshufd {{.*#+}} xmm2 = mem[2,3,0,1]
+; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512F-NEXT: vmovq %r8, %xmm1
+; AVX512F-NEXT: vmovq %r9, %xmm3
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm3[0],xmm1[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1
+; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,128,192,256,320,384,448]
+; AVX512F-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512F-NEXT: vmovd %xmm0, %ecx
+; AVX512F-NEXT: addl $512, %ecx # imm = 0x200
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512F-NEXT: vpermq {{[0-9]+}}(%rsp), %zmm0, %zmm0
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512F-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512F-NEXT: orq {{[0-9]+}}(%rsp), %r14
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: orq {{[0-9]+}}(%rsp), %r11
+; AVX512F-NEXT: orq %r14, %r11
+; AVX512F-NEXT: orq {{[0-9]+}}(%rsp), %rbx
+; AVX512F-NEXT: orq {{[0-9]+}}(%rsp), %r10
+; AVX512F-NEXT: orq %rbx, %r10
+; AVX512F-NEXT: orq %r11, %r10
+; AVX512F-NEXT: cmovel %ecx, %eax
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: test_ctlz_i1024:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rbx
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r10
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r11
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r14
+; AVX512VL-NEXT: vmovq %rdi, %xmm0
+; AVX512VL-NEXT: vmovq %rsi, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovq %rdx, %xmm1
+; AVX512VL-NEXT: vmovq %rcx, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
+; AVX512VL-NEXT: vmovq %r8, %xmm2
+; AVX512VL-NEXT: vmovq %r9, %xmm3
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,128,192,256,320,384,448]
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT: vmovd %xmm0, %ecx
+; AVX512VL-NEXT: addl $512, %ecx # imm = 0x200
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512VL-NEXT: vpermq {{[0-9]+}}(%rsp), %zmm0, %zmm0
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %r14
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %r11
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %rbx
+; AVX512VL-NEXT: orq %r14, %r11
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %r10
+; AVX512VL-NEXT: orq %rbx, %r10
+; AVX512VL-NEXT: orq %r11, %r10
+; AVX512VL-NEXT: cmovel %ecx, %eax
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: test_ctlz_i1024:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: pushq %r14
+; AVX512POPCNT-NEXT: pushq %rbx
+; AVX512POPCNT-NEXT: movq {{[0-9]+}}(%rsp), %rbx
+; AVX512POPCNT-NEXT: movq {{[0-9]+}}(%rsp), %r10
+; AVX512POPCNT-NEXT: movq {{[0-9]+}}(%rsp), %r11
+; AVX512POPCNT-NEXT: movq {{[0-9]+}}(%rsp), %r14
+; AVX512POPCNT-NEXT: vmovq %rdi, %xmm0
+; AVX512POPCNT-NEXT: vmovq %rsi, %xmm1
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512POPCNT-NEXT: vmovq %rdx, %xmm1
+; AVX512POPCNT-NEXT: vmovq %rcx, %xmm2
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512POPCNT-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512POPCNT-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
+; AVX512POPCNT-NEXT: vmovq %r8, %xmm2
+; AVX512POPCNT-NEXT: vmovq %r9, %xmm3
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512POPCNT-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512POPCNT-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512POPCNT-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512POPCNT-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,128,192,256,320,384,448]
+; AVX512POPCNT-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512POPCNT-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512POPCNT-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512POPCNT-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %ecx
+; AVX512POPCNT-NEXT: addl $512, %ecx # imm = 0x200
+; AVX512POPCNT-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512POPCNT-NEXT: vpermq {{[0-9]+}}(%rsp), %zmm0, %zmm0
+; AVX512POPCNT-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512POPCNT-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512POPCNT-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512POPCNT-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %eax
+; AVX512POPCNT-NEXT: orq {{[0-9]+}}(%rsp), %r14
+; AVX512POPCNT-NEXT: orq {{[0-9]+}}(%rsp), %r11
+; AVX512POPCNT-NEXT: orq {{[0-9]+}}(%rsp), %rbx
+; AVX512POPCNT-NEXT: orq %r14, %r11
+; AVX512POPCNT-NEXT: orq {{[0-9]+}}(%rsp), %r10
+; AVX512POPCNT-NEXT: orq %rbx, %r10
+; AVX512POPCNT-NEXT: orq %r11, %r10
+; AVX512POPCNT-NEXT: cmovel %ecx, %eax
+; AVX512POPCNT-NEXT: popq %rbx
+; AVX512POPCNT-NEXT: popq %r14
+; AVX512POPCNT-NEXT: vzeroupper
+; AVX512POPCNT-NEXT: retq
%cnt = call i1024 @llvm.ctlz.i1024(i1024 %a0, i1 0)
%res = trunc i1024 %cnt to i32
ret i32 %res
@@ -1560,38 +2127,103 @@ define i32 @load_ctlz_i1024(ptr %p0) nounwind {
; AVX2-NEXT: popq %rbp
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_ctlz_i1024:
-; AVX512: # %bb.0:
-; AVX512-NEXT: movq 80(%rdi), %rsi
-; AVX512-NEXT: movq 64(%rdi), %rcx
-; AVX512-NEXT: movq 72(%rdi), %rdx
-; AVX512-NEXT: movq 88(%rdi), %r8
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
-; AVX512-NEXT: vpermq 64(%rdi), %zmm0, %zmm1
-; AVX512-NEXT: vplzcntq %zmm1, %zmm2
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,128,192,256,320,384,448]
-; AVX512-NEXT: vpaddq %zmm3, %zmm2, %zmm2
-; AVX512-NEXT: vptestmq %zmm1, %zmm1, %k1
-; AVX512-NEXT: vpcompressq %zmm2, %zmm1 {%k1} {z}
-; AVX512-NEXT: vmovd %xmm1, %r9d
-; AVX512-NEXT: vpermq (%rdi), %zmm0, %zmm0
-; AVX512-NEXT: vplzcntq %zmm0, %zmm1
-; AVX512-NEXT: vpaddq %zmm3, %zmm1, %zmm1
-; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
-; AVX512-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: addl $512, %eax # imm = 0x200
-; AVX512-NEXT: orq 120(%rdi), %r8
-; AVX512-NEXT: orq 104(%rdi), %rdx
-; AVX512-NEXT: orq 112(%rdi), %rsi
-; AVX512-NEXT: orq %r8, %rdx
-; AVX512-NEXT: orq 96(%rdi), %rcx
-; AVX512-NEXT: orq %rsi, %rcx
-; AVX512-NEXT: orq %rdx, %rcx
-; AVX512-NEXT: cmovnel %r9d, %eax
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_ctlz_i1024:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: movq 80(%rdi), %rsi
+; AVX512F-NEXT: movq 64(%rdi), %rcx
+; AVX512F-NEXT: movq 72(%rdi), %rdx
+; AVX512F-NEXT: movq 88(%rdi), %r8
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512F-NEXT: vpermq 64(%rdi), %zmm0, %zmm1
+; AVX512F-NEXT: vplzcntq %zmm1, %zmm2
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,128,192,256,320,384,448]
+; AVX512F-NEXT: vpaddq %zmm3, %zmm2, %zmm2
+; AVX512F-NEXT: vptestmq %zmm1, %zmm1, %k1
+; AVX512F-NEXT: vpcompressq %zmm2, %zmm1 {%k1} {z}
+; AVX512F-NEXT: vmovd %xmm1, %r9d
+; AVX512F-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512F-NEXT: vpaddq %zmm3, %zmm1, %zmm1
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: orq 120(%rdi), %r8
+; AVX512F-NEXT: addl $512, %eax # imm = 0x200
+; AVX512F-NEXT: orq 104(%rdi), %rdx
+; AVX512F-NEXT: orq %r8, %rdx
+; AVX512F-NEXT: orq 112(%rdi), %rsi
+; AVX512F-NEXT: orq 96(%rdi), %rcx
+; AVX512F-NEXT: orq %rsi, %rcx
+; AVX512F-NEXT: orq %rdx, %rcx
+; AVX512F-NEXT: cmovnel %r9d, %eax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_ctlz_i1024:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: movq 80(%rdi), %rsi
+; AVX512VL-NEXT: movq 64(%rdi), %rcx
+; AVX512VL-NEXT: movq 72(%rdi), %rdx
+; AVX512VL-NEXT: movq 88(%rdi), %r8
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512VL-NEXT: vpermq 64(%rdi), %zmm0, %zmm1
+; AVX512VL-NEXT: vplzcntq %zmm1, %zmm2
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,128,192,256,320,384,448]
+; AVX512VL-NEXT: vpaddq %zmm3, %zmm2, %zmm2
+; AVX512VL-NEXT: vptestmq %zmm1, %zmm1, %k1
+; AVX512VL-NEXT: vpcompressq %zmm2, %zmm1 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm1, %r9d
+; AVX512VL-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512VL-NEXT: vpaddq %zmm3, %zmm1, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: addl $512, %eax # imm = 0x200
+; AVX512VL-NEXT: orq 120(%rdi), %r8
+; AVX512VL-NEXT: orq 104(%rdi), %rdx
+; AVX512VL-NEXT: orq 112(%rdi), %rsi
+; AVX512VL-NEXT: orq %r8, %rdx
+; AVX512VL-NEXT: orq 96(%rdi), %rcx
+; AVX512VL-NEXT: orq %rsi, %rcx
+; AVX512VL-NEXT: orq %rdx, %rcx
+; AVX512VL-NEXT: cmovnel %r9d, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: load_ctlz_i1024:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: movq 80(%rdi), %rsi
+; AVX512POPCNT-NEXT: movq 64(%rdi), %rcx
+; AVX512POPCNT-NEXT: movq 72(%rdi), %rdx
+; AVX512POPCNT-NEXT: movq 88(%rdi), %r8
+; AVX512POPCNT-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512POPCNT-NEXT: vpermq 64(%rdi), %zmm0, %zmm1
+; AVX512POPCNT-NEXT: vplzcntq %zmm1, %zmm2
+; AVX512POPCNT-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,128,192,256,320,384,448]
+; AVX512POPCNT-NEXT: vpaddq %zmm3, %zmm2, %zmm2
+; AVX512POPCNT-NEXT: vptestmq %zmm1, %zmm1, %k1
+; AVX512POPCNT-NEXT: vpcompressq %zmm2, %zmm1 {%k1} {z}
+; AVX512POPCNT-NEXT: vmovd %xmm1, %r9d
+; AVX512POPCNT-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512POPCNT-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512POPCNT-NEXT: vpaddq %zmm3, %zmm1, %zmm1
+; AVX512POPCNT-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512POPCNT-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512POPCNT-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %eax
+; AVX512POPCNT-NEXT: addl $512, %eax # imm = 0x200
+; AVX512POPCNT-NEXT: orq 120(%rdi), %r8
+; AVX512POPCNT-NEXT: orq 104(%rdi), %rdx
+; AVX512POPCNT-NEXT: orq 112(%rdi), %rsi
+; AVX512POPCNT-NEXT: orq %r8, %rdx
+; AVX512POPCNT-NEXT: orq 96(%rdi), %rcx
+; AVX512POPCNT-NEXT: orq %rsi, %rcx
+; AVX512POPCNT-NEXT: orq %rdx, %rcx
+; AVX512POPCNT-NEXT: cmovnel %r9d, %eax
+; AVX512POPCNT-NEXT: vzeroupper
+; AVX512POPCNT-NEXT: retq
%a0 = load i1024, ptr %p0
%cnt = call i1024 @llvm.ctlz.i1024(i1024 %a0, i1 0)
%res = trunc i1024 %cnt to i32
@@ -1792,16 +2424,39 @@ define i32 @load_ctlz_undef_i256(ptr %p0) nounwind {
; AVX2-NEXT: # kill: def $eax killed $eax killed $rax
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_ctlz_undef_i256:
-; AVX512: # %bb.0:
-; AVX512-NEXT: vpermq {{.*#+}} ymm0 = mem[3,2,1,0]
-; AVX512-NEXT: vptestmq %ymm0, %ymm0, %k1
-; AVX512-NEXT: vplzcntq %ymm0, %ymm0
-; AVX512-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX512-NEXT: vpcompressq %ymm0, %ymm0 {%k1} {z}
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_ctlz_undef_i256:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: vpermq {{.*#+}} ymm0 = mem[3,2,1,0]
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512F-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0
+; AVX512F-NEXT: kshiftlw $12, %k0, %k0
+; AVX512F-NEXT: kshiftrw $12, %k0, %k1
+; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_ctlz_undef_i256:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vpermq {{.*#+}} ymm0 = mem[3,2,1,0]
+; AVX512VL-NEXT: vptestmq %ymm0, %ymm0, %k1
+; AVX512VL-NEXT: vplzcntq %ymm0, %ymm0
+; AVX512VL-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX512VL-NEXT: vpcompressq %ymm0, %ymm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: load_ctlz_undef_i256:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: vpermq {{.*#+}} ymm0 = mem[3,2,1,0]
+; AVX512POPCNT-NEXT: vptestmq %ymm0, %ymm0, %k1
+; AVX512POPCNT-NEXT: vplzcntq %ymm0, %ymm0
+; AVX512POPCNT-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX512POPCNT-NEXT: vpcompressq %ymm0, %ymm0 {%k1} {z}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %eax
+; AVX512POPCNT-NEXT: vzeroupper
+; AVX512POPCNT-NEXT: retq
%a0 = load i256, ptr %p0
%cnt = call i256 @llvm.ctlz.i256(i256 %a0, i1 -1)
%res = trunc i256 %cnt to i32
@@ -1914,28 +2569,73 @@ define i32 @test_ctlz_undef_i512(i512 %a0) nounwind {
; AVX2-NEXT: popq %r15
; AVX2-NEXT: retq
;
-; AVX512-LABEL: test_ctlz_undef_i512:
-; AVX512: # %bb.0:
-; AVX512-NEXT: vmovq %rdi, %xmm0
-; AVX512-NEXT: vmovq %rsi, %xmm1
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; AVX512-NEXT: vmovq %rdx, %xmm1
-; AVX512-NEXT: vmovq %rcx, %xmm2
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
-; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
-; AVX512-NEXT: vmovq %r8, %xmm2
-; AVX512-NEXT: vmovq %r9, %xmm3
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
-; AVX512-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512-NEXT: vplzcntq %zmm0, %zmm0
-; AVX512-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; AVX512-NEXT: vpcompressq %zmm0, %zmm0 {%k1} {z}
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: test_ctlz_undef_i512:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: vmovq %rdi, %xmm0
+; AVX512F-NEXT: vmovq %rsi, %xmm1
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: vmovq %rdx, %xmm1
+; AVX512F-NEXT: vmovq %rcx, %xmm2
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512F-NEXT: vpshufd {{.*#+}} xmm2 = mem[2,3,0,1]
+; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512F-NEXT: vmovq %r8, %xmm1
+; AVX512F-NEXT: vmovq %r9, %xmm3
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm3[0],xmm1[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1
+; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm0
+; AVX512F-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512F-NEXT: vpcompressq %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: test_ctlz_undef_i512:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovq %rdi, %xmm0
+; AVX512VL-NEXT: vmovq %rsi, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovq %rdx, %xmm1
+; AVX512VL-NEXT: vmovq %rcx, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
+; AVX512VL-NEXT: vmovq %r8, %xmm2
+; AVX512VL-NEXT: vmovq %r9, %xmm3
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm0
+; AVX512VL-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512VL-NEXT: vpcompressq %zmm0, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: test_ctlz_undef_i512:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: vmovq %rdi, %xmm0
+; AVX512POPCNT-NEXT: vmovq %rsi, %xmm1
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512POPCNT-NEXT: vmovq %rdx, %xmm1
+; AVX512POPCNT-NEXT: vmovq %rcx, %xmm2
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512POPCNT-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512POPCNT-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
+; AVX512POPCNT-NEXT: vmovq %r8, %xmm2
+; AVX512POPCNT-NEXT: vmovq %r9, %xmm3
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512POPCNT-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512POPCNT-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512POPCNT-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512POPCNT-NEXT: vplzcntq %zmm0, %zmm0
+; AVX512POPCNT-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512POPCNT-NEXT: vpcompressq %zmm0, %zmm0 {%k1} {z}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %eax
+; AVX512POPCNT-NEXT: vzeroupper
+; AVX512POPCNT-NEXT: retq
%cnt = call i512 @llvm.ctlz.i512(i512 %a0, i1 -1)
%res = trunc i512 %cnt to i32
ret i32 %res
@@ -2056,17 +2756,40 @@ define i32 @load_ctlz_undef_i512(ptr %p0) nounwind {
; AVX2-NEXT: popq %r15
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_ctlz_undef_i512:
-; AVX512: # %bb.0:
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
-; AVX512-NEXT: vpermq (%rdi), %zmm0, %zmm0
-; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512-NEXT: vplzcntq %zmm0, %zmm0
-; AVX512-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; AVX512-NEXT: vpcompressq %zmm0, %zmm0 {%k1} {z}
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_ctlz_undef_i512:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512F-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm0
+; AVX512F-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512F-NEXT: vpcompressq %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_ctlz_undef_i512:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512VL-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm0
+; AVX512VL-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512VL-NEXT: vpcompressq %zmm0, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: load_ctlz_undef_i512:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512POPCNT-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512POPCNT-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512POPCNT-NEXT: vplzcntq %zmm0, %zmm0
+; AVX512POPCNT-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512POPCNT-NEXT: vpcompressq %zmm0, %zmm0 {%k1} {z}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %eax
+; AVX512POPCNT-NEXT: vzeroupper
+; AVX512POPCNT-NEXT: retq
%a0 = load i512, ptr %p0
%cnt = call i512 @llvm.ctlz.i512(i512 %a0, i1 -1)
%res = trunc i512 %cnt to i32
@@ -2326,53 +3049,148 @@ define i32 @test_ctlz_undef_i1024(i1024 %a0) nounwind {
; AVX2-NEXT: popq %rbp
; AVX2-NEXT: retq
;
-; AVX512-LABEL: test_ctlz_undef_i1024:
-; AVX512: # %bb.0:
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: movq {{[0-9]+}}(%rsp), %rbx
-; AVX512-NEXT: movq {{[0-9]+}}(%rsp), %r10
-; AVX512-NEXT: movq {{[0-9]+}}(%rsp), %r11
-; AVX512-NEXT: movq {{[0-9]+}}(%rsp), %r14
-; AVX512-NEXT: vmovq %rdi, %xmm0
-; AVX512-NEXT: vmovq %rsi, %xmm1
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; AVX512-NEXT: vmovq %rdx, %xmm1
-; AVX512-NEXT: vmovq %rcx, %xmm2
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
-; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
-; AVX512-NEXT: vmovq %r8, %xmm2
-; AVX512-NEXT: vmovq %r9, %xmm3
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
-; AVX512-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512-NEXT: vplzcntq %zmm0, %zmm1
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,128,192,256,320,384,448]
-; AVX512-NEXT: vpaddq %zmm2, %zmm1, %zmm1
-; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
-; AVX512-NEXT: vmovd %xmm0, %ecx
-; AVX512-NEXT: addl $512, %ecx # imm = 0x200
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
-; AVX512-NEXT: vpermq {{[0-9]+}}(%rsp), %zmm0, %zmm0
-; AVX512-NEXT: vplzcntq %zmm0, %zmm1
-; AVX512-NEXT: vpaddq %zmm2, %zmm1, %zmm1
-; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: orq {{[0-9]+}}(%rsp), %r14
-; AVX512-NEXT: orq {{[0-9]+}}(%rsp), %r11
-; AVX512-NEXT: orq {{[0-9]+}}(%rsp), %rbx
-; AVX512-NEXT: orq %r14, %r11
-; AVX512-NEXT: orq {{[0-9]+}}(%rsp), %r10
-; AVX512-NEXT: orq %rbx, %r10
-; AVX512-NEXT: orq %r11, %r10
-; AVX512-NEXT: cmovel %ecx, %eax
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: test_ctlz_undef_i1024:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %rbx
+; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r10
+; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r11
+; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r14
+; AVX512F-NEXT: vmovq %rdi, %xmm0
+; AVX512F-NEXT: vmovq %rsi, %xmm1
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: vmovq %rdx, %xmm1
+; AVX512F-NEXT: vmovq %rcx, %xmm2
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512F-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
+; AVX512F-NEXT: vmovq %r8, %xmm2
+; AVX512F-NEXT: vmovq %r9, %xmm3
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,128,192,256,320,384,448]
+; AVX512F-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512F-NEXT: vmovd %xmm0, %ecx
+; AVX512F-NEXT: addl $512, %ecx # imm = 0x200
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512F-NEXT: vpermq {{[0-9]+}}(%rsp), %zmm0, %zmm0
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512F-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512F-NEXT: orq {{[0-9]+}}(%rsp), %r14
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: orq {{[0-9]+}}(%rsp), %r11
+; AVX512F-NEXT: orq %r14, %r11
+; AVX512F-NEXT: orq {{[0-9]+}}(%rsp), %rbx
+; AVX512F-NEXT: orq {{[0-9]+}}(%rsp), %r10
+; AVX512F-NEXT: orq %rbx, %r10
+; AVX512F-NEXT: orq %r11, %r10
+; AVX512F-NEXT: cmovel %ecx, %eax
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: test_ctlz_undef_i1024:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rbx
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r10
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r11
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r14
+; AVX512VL-NEXT: vmovq %rdi, %xmm0
+; AVX512VL-NEXT: vmovq %rsi, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovq %rdx, %xmm1
+; AVX512VL-NEXT: vmovq %rcx, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
+; AVX512VL-NEXT: vmovq %r8, %xmm2
+; AVX512VL-NEXT: vmovq %r9, %xmm3
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,128,192,256,320,384,448]
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %ecx
+; AVX512VL-NEXT: addl $512, %ecx # imm = 0x200
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512VL-NEXT: vpermq {{[0-9]+}}(%rsp), %zmm0, %zmm0
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %r14
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %r11
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %rbx
+; AVX512VL-NEXT: orq %r14, %r11
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %r10
+; AVX512VL-NEXT: orq %rbx, %r10
+; AVX512VL-NEXT: orq %r11, %r10
+; AVX512VL-NEXT: cmovel %ecx, %eax
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: test_ctlz_undef_i1024:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: pushq %r14
+; AVX512POPCNT-NEXT: pushq %rbx
+; AVX512POPCNT-NEXT: movq {{[0-9]+}}(%rsp), %rbx
+; AVX512POPCNT-NEXT: movq {{[0-9]+}}(%rsp), %r10
+; AVX512POPCNT-NEXT: movq {{[0-9]+}}(%rsp), %r11
+; AVX512POPCNT-NEXT: movq {{[0-9]+}}(%rsp), %r14
+; AVX512POPCNT-NEXT: vmovq %rdi, %xmm0
+; AVX512POPCNT-NEXT: vmovq %rsi, %xmm1
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512POPCNT-NEXT: vmovq %rdx, %xmm1
+; AVX512POPCNT-NEXT: vmovq %rcx, %xmm2
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512POPCNT-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512POPCNT-NEXT: vpshufd {{.*#+}} xmm1 = mem[2,3,0,1]
+; AVX512POPCNT-NEXT: vmovq %r8, %xmm2
+; AVX512POPCNT-NEXT: vmovq %r9, %xmm3
+; AVX512POPCNT-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512POPCNT-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512POPCNT-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512POPCNT-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512POPCNT-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,128,192,256,320,384,448]
+; AVX512POPCNT-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512POPCNT-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512POPCNT-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %ecx
+; AVX512POPCNT-NEXT: addl $512, %ecx # imm = 0x200
+; AVX512POPCNT-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512POPCNT-NEXT: vpermq {{[0-9]+}}(%rsp), %zmm0, %zmm0
+; AVX512POPCNT-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512POPCNT-NEXT: vpaddq %zmm2, %zmm1, %zmm1
+; AVX512POPCNT-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512POPCNT-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %eax
+; AVX512POPCNT-NEXT: orq {{[0-9]+}}(%rsp), %r14
+; AVX512POPCNT-NEXT: orq {{[0-9]+}}(%rsp), %r11
+; AVX512POPCNT-NEXT: orq {{[0-9]+}}(%rsp), %rbx
+; AVX512POPCNT-NEXT: orq %r14, %r11
+; AVX512POPCNT-NEXT: orq {{[0-9]+}}(%rsp), %r10
+; AVX512POPCNT-NEXT: orq %rbx, %r10
+; AVX512POPCNT-NEXT: orq %r11, %r10
+; AVX512POPCNT-NEXT: cmovel %ecx, %eax
+; AVX512POPCNT-NEXT: popq %rbx
+; AVX512POPCNT-NEXT: popq %r14
+; AVX512POPCNT-NEXT: vzeroupper
+; AVX512POPCNT-NEXT: retq
%cnt = call i1024 @llvm.ctlz.i1024(i1024 %a0, i1 -1)
%res = trunc i1024 %cnt to i32
ret i32 %res
@@ -2638,37 +3456,100 @@ define i32 @load_ctlz_undef_i1024(ptr %p0) nounwind {
; AVX2-NEXT: popq %rbp
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_ctlz_undef_i1024:
-; AVX512: # %bb.0:
-; AVX512-NEXT: movq 80(%rdi), %rsi
-; AVX512-NEXT: movq 64(%rdi), %rcx
-; AVX512-NEXT: movq 72(%rdi), %rdx
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
-; AVX512-NEXT: vpermq 64(%rdi), %zmm0, %zmm1
-; AVX512-NEXT: movq 88(%rdi), %r8
-; AVX512-NEXT: vplzcntq %zmm1, %zmm2
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,128,192,256,320,384,448]
-; AVX512-NEXT: vpaddq %zmm3, %zmm2, %zmm2
-; AVX512-NEXT: vptestmq %zmm1, %zmm1, %k1
-; AVX512-NEXT: vpcompressq %zmm2, %zmm1 {%k1} {z}
-; AVX512-NEXT: vmovd %xmm1, %r9d
-; AVX512-NEXT: vpermq (%rdi), %zmm0, %zmm0
-; AVX512-NEXT: vplzcntq %zmm0, %zmm1
-; AVX512-NEXT: vpaddq %zmm3, %zmm1, %zmm1
-; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: addl $512, %eax # imm = 0x200
-; AVX512-NEXT: orq 120(%rdi), %r8
-; AVX512-NEXT: orq 104(%rdi), %rdx
-; AVX512-NEXT: orq 112(%rdi), %rsi
-; AVX512-NEXT: orq %r8, %rdx
-; AVX512-NEXT: orq 96(%rdi), %rcx
-; AVX512-NEXT: orq %rsi, %rcx
-; AVX512-NEXT: orq %rdx, %rcx
-; AVX512-NEXT: cmovnel %r9d, %eax
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_ctlz_undef_i1024:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: movq 80(%rdi), %rsi
+; AVX512F-NEXT: movq 64(%rdi), %rcx
+; AVX512F-NEXT: movq 72(%rdi), %rdx
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512F-NEXT: vpermq 64(%rdi), %zmm0, %zmm1
+; AVX512F-NEXT: movq 88(%rdi), %r8
+; AVX512F-NEXT: vplzcntq %zmm1, %zmm2
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,128,192,256,320,384,448]
+; AVX512F-NEXT: vpaddq %zmm3, %zmm2, %zmm2
+; AVX512F-NEXT: vptestmq %zmm1, %zmm1, %k1
+; AVX512F-NEXT: vpcompressq %zmm2, %zmm1 {%k1} {z}
+; AVX512F-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512F-NEXT: vmovd %xmm1, %r9d
+; AVX512F-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512F-NEXT: vpaddq %zmm3, %zmm1, %zmm1
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: orq 120(%rdi), %r8
+; AVX512F-NEXT: addl $512, %eax # imm = 0x200
+; AVX512F-NEXT: orq 104(%rdi), %rdx
+; AVX512F-NEXT: orq %r8, %rdx
+; AVX512F-NEXT: orq 112(%rdi), %rsi
+; AVX512F-NEXT: orq 96(%rdi), %rcx
+; AVX512F-NEXT: orq %rsi, %rcx
+; AVX512F-NEXT: orq %rdx, %rcx
+; AVX512F-NEXT: cmovnel %r9d, %eax
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_ctlz_undef_i1024:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: movq 80(%rdi), %rsi
+; AVX512VL-NEXT: movq 64(%rdi), %rcx
+; AVX512VL-NEXT: movq 72(%rdi), %rdx
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512VL-NEXT: vpermq 64(%rdi), %zmm0, %zmm1
+; AVX512VL-NEXT: movq 88(%rdi), %r8
+; AVX512VL-NEXT: vplzcntq %zmm1, %zmm2
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,128,192,256,320,384,448]
+; AVX512VL-NEXT: vpaddq %zmm3, %zmm2, %zmm2
+; AVX512VL-NEXT: vptestmq %zmm1, %zmm1, %k1
+; AVX512VL-NEXT: vpcompressq %zmm2, %zmm1 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm1, %r9d
+; AVX512VL-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512VL-NEXT: vpaddq %zmm3, %zmm1, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: addl $512, %eax # imm = 0x200
+; AVX512VL-NEXT: orq 120(%rdi), %r8
+; AVX512VL-NEXT: orq 104(%rdi), %rdx
+; AVX512VL-NEXT: orq 112(%rdi), %rsi
+; AVX512VL-NEXT: orq %r8, %rdx
+; AVX512VL-NEXT: orq 96(%rdi), %rcx
+; AVX512VL-NEXT: orq %rsi, %rcx
+; AVX512VL-NEXT: orq %rdx, %rcx
+; AVX512VL-NEXT: cmovnel %r9d, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512POPCNT-LABEL: load_ctlz_undef_i1024:
+; AVX512POPCNT: # %bb.0:
+; AVX512POPCNT-NEXT: movq 80(%rdi), %rsi
+; AVX512POPCNT-NEXT: movq 64(%rdi), %rcx
+; AVX512POPCNT-NEXT: movq 72(%rdi), %rdx
+; AVX512POPCNT-NEXT: vmovdqa64 {{.*#+}} zmm0 = [7,6,5,4,3,2,1,0]
+; AVX512POPCNT-NEXT: vpermq 64(%rdi), %zmm0, %zmm1
+; AVX512POPCNT-NEXT: movq 88(%rdi), %r8
+; AVX512POPCNT-NEXT: vplzcntq %zmm1, %zmm2
+; AVX512POPCNT-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,128,192,256,320,384,448]
+; AVX512POPCNT-NEXT: vpaddq %zmm3, %zmm2, %zmm2
+; AVX512POPCNT-NEXT: vptestmq %zmm1, %zmm1, %k1
+; AVX512POPCNT-NEXT: vpcompressq %zmm2, %zmm1 {%k1} {z}
+; AVX512POPCNT-NEXT: vmovd %xmm1, %r9d
+; AVX512POPCNT-NEXT: vpermq (%rdi), %zmm0, %zmm0
+; AVX512POPCNT-NEXT: vplzcntq %zmm0, %zmm1
+; AVX512POPCNT-NEXT: vpaddq %zmm3, %zmm1, %zmm1
+; AVX512POPCNT-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512POPCNT-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512POPCNT-NEXT: vmovd %xmm0, %eax
+; AVX512POPCNT-NEXT: addl $512, %eax # imm = 0x200
+; AVX512POPCNT-NEXT: orq 120(%rdi), %r8
+; AVX512POPCNT-NEXT: orq 104(%rdi), %rdx
+; AVX512POPCNT-NEXT: orq 112(%rdi), %rsi
+; AVX512POPCNT-NEXT: orq %r8, %rdx
+; AVX512POPCNT-NEXT: orq 96(%rdi), %rcx
+; AVX512POPCNT-NEXT: orq %rsi, %rcx
+; AVX512POPCNT-NEXT: orq %rdx, %rcx
+; AVX512POPCNT-NEXT: cmovnel %r9d, %eax
+; AVX512POPCNT-NEXT: vzeroupper
+; AVX512POPCNT-NEXT: retq
%a0 = load i1024, ptr %p0
%cnt = call i1024 @llvm.ctlz.i1024(i1024 %a0, i1 -1)
%res = trunc i1024 %cnt to i32
@@ -2864,19 +3745,36 @@ define i32 @load_cttz_i256(ptr %p0) nounwind {
; AVX512F-LABEL: load_cttz_i256:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovdqu (%rdi), %ymm0
-; AVX512F-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512F-NEXT: vpaddq %ymm1, %ymm0, %ymm1
-; AVX512F-NEXT: vpandn %ymm1, %ymm0, %ymm1
-; AVX512F-NEXT: vplzcntq %ymm1, %ymm1
-; AVX512F-NEXT: vmovdqa {{.*#+}} ymm2 = [64,128,192,256]
-; AVX512F-NEXT: vpsubq %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT: vptestmq %ymm0, %ymm0, %k1
-; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm0 = [256,256,256,256]
-; AVX512F-NEXT: vpcompressq %ymm1, %ymm0 {%k1}
-; AVX512F-NEXT: vmovd %xmm0, %eax
-; AVX512F-NEXT: vzeroupper
+; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [256,256,256,256]
+; AVX512F-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512F-NEXT: vpaddq %ymm2, %ymm0, %ymm2
+; AVX512F-NEXT: vpandn %ymm2, %ymm0, %ymm2
+; AVX512F-NEXT: vplzcntq %zmm2, %zmm2
+; AVX512F-NEXT: vmovdqa {{.*#+}} ymm3 = [64,128,192,256]
+; AVX512F-NEXT: vpsubq %ymm2, %ymm3, %ymm2
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0
+; AVX512F-NEXT: kshiftlw $12, %k0, %k0
+; AVX512F-NEXT: kshiftrw $12, %k0, %k1
+; AVX512F-NEXT: vpcompressq %zmm2, %zmm1 {%k1}
+; AVX512F-NEXT: vmovd %xmm1, %eax
; AVX512F-NEXT: retq
;
+; AVX512VL-LABEL: load_cttz_i256:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovdqu (%rdi), %ymm0
+; AVX512VL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; AVX512VL-NEXT: vpandn %ymm1, %ymm0, %ymm1
+; AVX512VL-NEXT: vplzcntq %ymm1, %ymm1
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm2 = [64,128,192,256]
+; AVX512VL-NEXT: vpsubq %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vptestmq %ymm0, %ymm0, %k1
+; AVX512VL-NEXT: vpbroadcastq {{.*#+}} ymm0 = [256,256,256,256]
+; AVX512VL-NEXT: vpcompressq %ymm1, %ymm0 {%k1}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
; AVX512POPCNT-LABEL: load_cttz_i256:
; AVX512POPCNT: # %bb.0:
; AVX512POPCNT-NEXT: vmovdqu (%rdi), %ymm0
@@ -3012,9 +3910,35 @@ define i32 @test_cttz_i512(i512 %a0) nounwind {
; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
; AVX512F-NEXT: vmovd %xmm0, %eax
-; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
+; AVX512VL-LABEL: test_cttz_i512:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovq %rcx, %xmm0
+; AVX512VL-NEXT: vmovq %rdx, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovq %rsi, %xmm1
+; AVX512VL-NEXT: vmovq %rdi, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vmovq %r9, %xmm1
+; AVX512VL-NEXT: vmovq %r8, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinserti128 $1, {{[0-9]+}}(%rsp), %ymm1, %ymm1
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm1 = -1
+; AVX512VL-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT: vpandnq %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT: vplzcntq %zmm1, %zmm1
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm2 = [64,128,192,256,320,384,448,512]
+; AVX512VL-NEXT: vpsubq %zmm1, %zmm2, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
; AVX512POPCNT-LABEL: test_cttz_i512:
; AVX512POPCNT: # %bb.0:
; AVX512POPCNT-NEXT: vmovq %rcx, %xmm0
@@ -3166,9 +4090,24 @@ define i32 @load_cttz_i512(ptr %p0) nounwind {
; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
; AVX512F-NEXT: vmovd %xmm0, %eax
-; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
+; AVX512VL-LABEL: load_cttz_i512:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovdqu64 (%rdi), %zmm0
+; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm1 = -1
+; AVX512VL-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT: vpandnq %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT: vplzcntq %zmm1, %zmm1
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm2 = [64,128,192,256,320,384,448,512]
+; AVX512VL-NEXT: vpsubq %zmm1, %zmm2, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
; AVX512POPCNT-LABEL: load_cttz_i512:
; AVX512POPCNT: # %bb.0:
; AVX512POPCNT-NEXT: vmovdqu64 (%rdi), %zmm0
@@ -3428,12 +4367,12 @@ define i32 @test_cttz_i1024(i1024 %a0) nounwind {
; AVX512F-NEXT: vmovq %rsi, %xmm1
; AVX512F-NEXT: vmovq %rdi, %xmm2
; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
-; AVX512F-NEXT: vmovq %r9, %xmm2
-; AVX512F-NEXT: vmovq %r8, %xmm3
-; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
-; AVX512F-NEXT: vinserti128 $1, {{[0-9]+}}(%rsp), %ymm2, %ymm2
; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512F-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512F-NEXT: vmovq %r9, %xmm1
+; AVX512F-NEXT: vmovq %r8, %xmm2
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512F-NEXT: vinserti128 $1, {{[0-9]+}}(%rsp), %ymm1, %ymm1
+; AVX512F-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; AVX512F-NEXT: vmovdqu64 {{[0-9]+}}(%rsp), %zmm1
; AVX512F-NEXT: vpternlogd {{.*#+}} zmm2 = -1
; AVX512F-NEXT: vpaddq %zmm2, %zmm0, %zmm3
@@ -3461,9 +4400,52 @@ define i32 @test_cttz_i1024(i1024 %a0) nounwind {
; AVX512F-NEXT: orq %rdi, %rdx
; AVX512F-NEXT: orq %rcx, %rdx
; AVX512F-NEXT: cmovnel %r10d, %eax
-; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
+; AVX512VL-LABEL: test_cttz_i1024:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovq %rcx, %xmm0
+; AVX512VL-NEXT: vmovq %rdx, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovq %rsi, %xmm1
+; AVX512VL-NEXT: vmovq %rdi, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vmovq %r9, %xmm2
+; AVX512VL-NEXT: vmovq %r8, %xmm3
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512VL-NEXT: vinserti128 $1, {{[0-9]+}}(%rsp), %ymm2, %ymm2
+; AVX512VL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512VL-NEXT: vmovdqu64 {{[0-9]+}}(%rsp), %zmm1
+; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm2 = -1
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm0, %zmm3
+; AVX512VL-NEXT: vpandnq %zmm3, %zmm0, %zmm3
+; AVX512VL-NEXT: vplzcntq %zmm3, %zmm3
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm4 = [64,128,192,256,320,384,448,512]
+; AVX512VL-NEXT: vpsubq %zmm3, %zmm4, %zmm3
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpcompressq %zmm3, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %r10d
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm1, %zmm0
+; AVX512VL-NEXT: vpandnq %zmm0, %zmm1, %zmm0
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm0
+; AVX512VL-NEXT: vpsubq %zmm0, %zmm4, %zmm0
+; AVX512VL-NEXT: vptestmq %zmm1, %zmm1, %k1
+; AVX512VL-NEXT: vpbroadcastq {{.*#+}} zmm1 = [512,512,512,512,512,512,512,512]
+; AVX512VL-NEXT: vpcompressq %zmm0, %zmm1 {%k1}
+; AVX512VL-NEXT: vmovd %xmm1, %eax
+; AVX512VL-NEXT: addl $512, %eax # imm = 0x200
+; AVX512VL-NEXT: orq %r9, %rsi
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %rcx
+; AVX512VL-NEXT: orq %rsi, %rcx
+; AVX512VL-NEXT: orq %r8, %rdi
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %rdx
+; AVX512VL-NEXT: orq %rdi, %rdx
+; AVX512VL-NEXT: orq %rcx, %rdx
+; AVX512VL-NEXT: cmovnel %r10d, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
; AVX512POPCNT-LABEL: test_cttz_i1024:
; AVX512POPCNT: # %bb.0:
; AVX512POPCNT-NEXT: vmovq %rcx, %xmm0
@@ -3765,9 +4747,9 @@ define i32 @load_cttz_i1024(ptr %p0) nounwind {
; AVX512F-NEXT: movq 24(%rdi), %rsi
; AVX512F-NEXT: orq 56(%rdi), %rsi
; AVX512F-NEXT: orq 40(%rdi), %rdx
-; AVX512F-NEXT: orq %rsi, %rdx
; AVX512F-NEXT: orq 48(%rdi), %rax
; AVX512F-NEXT: orq 32(%rdi), %rcx
+; AVX512F-NEXT: orq %rsi, %rdx
; AVX512F-NEXT: orq %rax, %rcx
; AVX512F-NEXT: vpternlogd {{.*#+}} zmm2 = -1
; AVX512F-NEXT: vpaddq %zmm2, %zmm1, %zmm3
@@ -3789,9 +4771,45 @@ define i32 @load_cttz_i1024(ptr %p0) nounwind {
; AVX512F-NEXT: addl $512, %eax # imm = 0x200
; AVX512F-NEXT: orq %rdx, %rcx
; AVX512F-NEXT: cmovnel %esi, %eax
-; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
+; AVX512VL-LABEL: load_cttz_i1024:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovdqu64 64(%rdi), %zmm0
+; AVX512VL-NEXT: vmovdqu64 (%rdi), %zmm1
+; AVX512VL-NEXT: movq 16(%rdi), %rax
+; AVX512VL-NEXT: movq (%rdi), %rcx
+; AVX512VL-NEXT: movq 8(%rdi), %rdx
+; AVX512VL-NEXT: movq 24(%rdi), %rsi
+; AVX512VL-NEXT: orq 56(%rdi), %rsi
+; AVX512VL-NEXT: orq 40(%rdi), %rdx
+; AVX512VL-NEXT: orq %rsi, %rdx
+; AVX512VL-NEXT: orq 48(%rdi), %rax
+; AVX512VL-NEXT: orq 32(%rdi), %rcx
+; AVX512VL-NEXT: orq %rax, %rcx
+; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm2 = -1
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm1, %zmm3
+; AVX512VL-NEXT: vpandnq %zmm3, %zmm1, %zmm3
+; AVX512VL-NEXT: vplzcntq %zmm3, %zmm3
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm4 = [64,128,192,256,320,384,448,512]
+; AVX512VL-NEXT: vpsubq %zmm3, %zmm4, %zmm3
+; AVX512VL-NEXT: vptestmq %zmm1, %zmm1, %k1
+; AVX512VL-NEXT: vpcompressq %zmm3, %zmm1 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm1, %esi
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm0, %zmm1
+; AVX512VL-NEXT: vpandnq %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT: vplzcntq %zmm1, %zmm1
+; AVX512VL-NEXT: vpsubq %zmm1, %zmm4, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpbroadcastq {{.*#+}} zmm0 = [512,512,512,512,512,512,512,512]
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: addl $512, %eax # imm = 0x200
+; AVX512VL-NEXT: orq %rdx, %rcx
+; AVX512VL-NEXT: cmovnel %esi, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
; AVX512POPCNT-LABEL: load_cttz_i1024:
; AVX512POPCNT: # %bb.0:
; AVX512POPCNT-NEXT: vmovdqu64 64(%rdi), %zmm0
@@ -4022,15 +5040,31 @@ define i32 @load_cttz_undef_i256(ptr %p0) nounwind {
; AVX512F-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
; AVX512F-NEXT: vpaddq %ymm1, %ymm0, %ymm1
; AVX512F-NEXT: vpandn %ymm1, %ymm0, %ymm1
+; AVX512F-NEXT: vplzcntq %zmm1, %zmm1
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm2 = [64,128,192,256]
-; AVX512F-NEXT: vplzcntq %ymm1, %ymm1
; AVX512F-NEXT: vpsubq %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT: vptestmq %ymm0, %ymm0, %k1
-; AVX512F-NEXT: vpcompressq %ymm1, %ymm0 {%k1} {z}
+; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0
+; AVX512F-NEXT: kshiftlw $12, %k0, %k0
+; AVX512F-NEXT: kshiftrw $12, %k0, %k1
+; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
; AVX512F-NEXT: vmovd %xmm0, %eax
-; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
+; AVX512VL-LABEL: load_cttz_undef_i256:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovdqu (%rdi), %ymm0
+; AVX512VL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; AVX512VL-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; AVX512VL-NEXT: vpandn %ymm1, %ymm0, %ymm1
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm2 = [64,128,192,256]
+; AVX512VL-NEXT: vplzcntq %ymm1, %ymm1
+; AVX512VL-NEXT: vpsubq %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vptestmq %ymm0, %ymm0, %k1
+; AVX512VL-NEXT: vpcompressq %ymm1, %ymm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
; AVX512POPCNT-LABEL: load_cttz_undef_i256:
; AVX512POPCNT: # %bb.0:
; AVX512POPCNT-NEXT: vmovdqu (%rdi), %ymm0
@@ -4157,15 +5191,40 @@ define i32 @test_cttz_undef_i512(i512 %a0) nounwind {
; AVX512F-NEXT: vpternlogd {{.*#+}} zmm1 = -1
; AVX512F-NEXT: vpaddq %zmm1, %zmm0, %zmm1
; AVX512F-NEXT: vpandnq %zmm1, %zmm0, %zmm1
-; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = [64,128,192,256,320,384,448,512]
; AVX512F-NEXT: vplzcntq %zmm1, %zmm1
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = [64,128,192,256,320,384,448,512]
; AVX512F-NEXT: vpsubq %zmm1, %zmm2, %zmm1
; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
; AVX512F-NEXT: vmovd %xmm0, %eax
-; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
+; AVX512VL-LABEL: test_cttz_undef_i512:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovq %rcx, %xmm0
+; AVX512VL-NEXT: vmovq %rdx, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovq %rsi, %xmm1
+; AVX512VL-NEXT: vmovq %rdi, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vmovq %r9, %xmm1
+; AVX512VL-NEXT: vmovq %r8, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinserti128 $1, {{[0-9]+}}(%rsp), %ymm1, %ymm1
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm1 = -1
+; AVX512VL-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT: vpandnq %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm2 = [64,128,192,256,320,384,448,512]
+; AVX512VL-NEXT: vplzcntq %zmm1, %zmm1
+; AVX512VL-NEXT: vpsubq %zmm1, %zmm2, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
; AVX512POPCNT-LABEL: test_cttz_undef_i512:
; AVX512POPCNT: # %bb.0:
; AVX512POPCNT-NEXT: vmovq %rcx, %xmm0
@@ -4308,15 +5367,29 @@ define i32 @load_cttz_undef_i512(ptr %p0) nounwind {
; AVX512F-NEXT: vpternlogd {{.*#+}} zmm1 = -1
; AVX512F-NEXT: vpaddq %zmm1, %zmm0, %zmm1
; AVX512F-NEXT: vpandnq %zmm1, %zmm0, %zmm1
-; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = [64,128,192,256,320,384,448,512]
; AVX512F-NEXT: vplzcntq %zmm1, %zmm1
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = [64,128,192,256,320,384,448,512]
; AVX512F-NEXT: vpsubq %zmm1, %zmm2, %zmm1
; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1
; AVX512F-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
; AVX512F-NEXT: vmovd %xmm0, %eax
-; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
+; AVX512VL-LABEL: load_cttz_undef_i512:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovdqu64 (%rdi), %zmm0
+; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm1 = -1
+; AVX512VL-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT: vpandnq %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm2 = [64,128,192,256,320,384,448,512]
+; AVX512VL-NEXT: vplzcntq %zmm1, %zmm1
+; AVX512VL-NEXT: vpsubq %zmm1, %zmm2, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
; AVX512POPCNT-LABEL: load_cttz_undef_i512:
; AVX512POPCNT: # %bb.0:
; AVX512POPCNT-NEXT: vmovdqu64 (%rdi), %zmm0
@@ -4605,9 +5678,51 @@ define i32 @test_cttz_undef_i1024(i1024 %a0) nounwind {
; AVX512F-NEXT: orq %rdi, %rdx
; AVX512F-NEXT: orq %rcx, %rdx
; AVX512F-NEXT: cmovnel %r10d, %eax
-; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
+; AVX512VL-LABEL: test_cttz_undef_i1024:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovq %rcx, %xmm0
+; AVX512VL-NEXT: vmovq %rdx, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovq %rsi, %xmm1
+; AVX512VL-NEXT: vmovq %rdi, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vmovq %r9, %xmm1
+; AVX512VL-NEXT: vmovq %r8, %xmm2
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinserti128 $1, {{[0-9]+}}(%rsp), %ymm1, %ymm1
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512VL-NEXT: vmovdqu64 {{[0-9]+}}(%rsp), %zmm1
+; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm2 = -1
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm0, %zmm3
+; AVX512VL-NEXT: vpandnq %zmm3, %zmm0, %zmm3
+; AVX512VL-NEXT: vplzcntq %zmm3, %zmm3
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm4 = [64,128,192,256,320,384,448,512]
+; AVX512VL-NEXT: vpsubq %zmm3, %zmm4, %zmm3
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpcompressq %zmm3, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %r10d
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm1, %zmm0
+; AVX512VL-NEXT: vpandnq %zmm0, %zmm1, %zmm0
+; AVX512VL-NEXT: vplzcntq %zmm0, %zmm0
+; AVX512VL-NEXT: vpsubq %zmm0, %zmm4, %zmm0
+; AVX512VL-NEXT: vptestmq %zmm1, %zmm1, %k1
+; AVX512VL-NEXT: vpcompressq %zmm0, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: addl $512, %eax # imm = 0x200
+; AVX512VL-NEXT: orq %r9, %rsi
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %rcx
+; AVX512VL-NEXT: orq %rsi, %rcx
+; AVX512VL-NEXT: orq %r8, %rdi
+; AVX512VL-NEXT: orq {{[0-9]+}}(%rsp), %rdx
+; AVX512VL-NEXT: orq %rdi, %rdx
+; AVX512VL-NEXT: orq %rcx, %rdx
+; AVX512VL-NEXT: cmovnel %r10d, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
; AVX512POPCNT-LABEL: test_cttz_undef_i1024:
; AVX512POPCNT: # %bb.0:
; AVX512POPCNT-NEXT: vmovq %rcx, %xmm0
@@ -4908,8 +6023,8 @@ define i32 @load_cttz_undef_i1024(ptr %p0) nounwind {
; AVX512F-NEXT: orq 56(%rdi), %rsi
; AVX512F-NEXT: orq 40(%rdi), %rdx
; AVX512F-NEXT: orq 48(%rdi), %rax
-; AVX512F-NEXT: orq 32(%rdi), %rcx
; AVX512F-NEXT: orq %rsi, %rdx
+; AVX512F-NEXT: orq 32(%rdi), %rcx
; AVX512F-NEXT: orq %rax, %rcx
; AVX512F-NEXT: vpternlogd {{.*#+}} zmm2 = -1
; AVX512F-NEXT: vpaddq %zmm2, %zmm1, %zmm3
@@ -4930,9 +6045,44 @@ define i32 @load_cttz_undef_i1024(ptr %p0) nounwind {
; AVX512F-NEXT: addl $512, %eax # imm = 0x200
; AVX512F-NEXT: orq %rdx, %rcx
; AVX512F-NEXT: cmovnel %esi, %eax
-; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
+; AVX512VL-LABEL: load_cttz_undef_i1024:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vmovdqu64 64(%rdi), %zmm0
+; AVX512VL-NEXT: vmovdqu64 (%rdi), %zmm1
+; AVX512VL-NEXT: movq 16(%rdi), %rax
+; AVX512VL-NEXT: movq (%rdi), %rcx
+; AVX512VL-NEXT: movq 8(%rdi), %rdx
+; AVX512VL-NEXT: movq 24(%rdi), %rsi
+; AVX512VL-NEXT: orq 56(%rdi), %rsi
+; AVX512VL-NEXT: orq 40(%rdi), %rdx
+; AVX512VL-NEXT: orq 48(%rdi), %rax
+; AVX512VL-NEXT: orq 32(%rdi), %rcx
+; AVX512VL-NEXT: orq %rsi, %rdx
+; AVX512VL-NEXT: orq %rax, %rcx
+; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm2 = -1
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm1, %zmm3
+; AVX512VL-NEXT: vpandnq %zmm3, %zmm1, %zmm3
+; AVX512VL-NEXT: vplzcntq %zmm3, %zmm3
+; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm4 = [64,128,192,256,320,384,448,512]
+; AVX512VL-NEXT: vpsubq %zmm3, %zmm4, %zmm3
+; AVX512VL-NEXT: vptestmq %zmm1, %zmm1, %k1
+; AVX512VL-NEXT: vpcompressq %zmm3, %zmm1 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm1, %esi
+; AVX512VL-NEXT: vpaddq %zmm2, %zmm0, %zmm1
+; AVX512VL-NEXT: vpandnq %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT: vplzcntq %zmm1, %zmm1
+; AVX512VL-NEXT: vpsubq %zmm1, %zmm4, %zmm1
+; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
+; AVX512VL-NEXT: vpcompressq %zmm1, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vmovd %xmm0, %eax
+; AVX512VL-NEXT: addl $512, %eax # imm = 0x200
+; AVX512VL-NEXT: orq %rdx, %rcx
+; AVX512VL-NEXT: cmovnel %esi, %eax
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
; AVX512POPCNT-LABEL: load_cttz_undef_i1024:
; AVX512POPCNT: # %bb.0:
; AVX512POPCNT-NEXT: vmovdqu64 64(%rdi), %zmm0
More information about the llvm-commits
mailing list