summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/X86/avx2-schedule.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/X86/avx2-schedule.ll')
-rw-r--r--llvm/test/CodeGen/X86/avx2-schedule.ll110
1 files changed, 110 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/X86/avx2-schedule.ll b/llvm/test/CodeGen/X86/avx2-schedule.ll
index a3862d7e27c..971ebe5e711 100644
--- a/llvm/test/CodeGen/X86/avx2-schedule.ll
+++ b/llvm/test/CodeGen/X86/avx2-schedule.ll
@@ -1,9 +1,17 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=haswell | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
define <32 x i8> @test_pabsb(<32 x i8> %a0, <32 x i8> *%a1) {
+; GENERIC-LABEL: test_pabsb:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpabsb %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpabsb (%rdi), %ymm1 # sched: [7:1.00]
+; GENERIC-NEXT: vpor %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_pabsb:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpabsb %ymm0, %ymm0 # sched: [1:0.50]
@@ -26,6 +34,13 @@ define <32 x i8> @test_pabsb(<32 x i8> %a0, <32 x i8> *%a1) {
declare <32 x i8> @llvm.x86.avx2.pabs.b(<32 x i8>) nounwind readnone
define <8 x i32> @test_pabsd(<8 x i32> %a0, <8 x i32> *%a1) {
+; GENERIC-LABEL: test_pabsd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpabsd %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpabsd (%rdi), %ymm1 # sched: [7:1.00]
+; GENERIC-NEXT: vpor %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_pabsd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpabsd %ymm0, %ymm0 # sched: [1:0.50]
@@ -48,6 +63,13 @@ define <8 x i32> @test_pabsd(<8 x i32> %a0, <8 x i32> *%a1) {
declare <8 x i32> @llvm.x86.avx2.pabs.d(<8 x i32>) nounwind readnone
define <16 x i16> @test_pabsw(<16 x i16> %a0, <16 x i16> *%a1) {
+; GENERIC-LABEL: test_pabsw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpabsw %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpabsw (%rdi), %ymm1 # sched: [7:1.00]
+; GENERIC-NEXT: vpor %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_pabsw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpabsw %ymm0, %ymm0 # sched: [1:0.50]
@@ -70,6 +92,12 @@ define <16 x i16> @test_pabsw(<16 x i16> %a0, <16 x i16> *%a1) {
declare <16 x i16> @llvm.x86.avx2.pabs.w(<16 x i16>) nounwind readnone
define <32 x i8> @test_paddb(<32 x i8> %a0, <32 x i8> %a1, <32 x i8> *%a2) {
+; GENERIC-LABEL: test_paddb:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpaddb %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpaddb (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_paddb:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpaddb %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
@@ -88,6 +116,12 @@ define <32 x i8> @test_paddb(<32 x i8> %a0, <32 x i8> %a1, <32 x i8> *%a2) {
}
define <8 x i32> @test_paddd(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> *%a2) {
+; GENERIC-LABEL: test_paddd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpaddd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpaddd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_paddd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpaddd %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
@@ -106,6 +140,12 @@ define <8 x i32> @test_paddd(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> *%a2) {
}
define <4 x i64> @test_paddq(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
+; GENERIC-LABEL: test_paddq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpaddq (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_paddq:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
@@ -124,6 +164,12 @@ define <4 x i64> @test_paddq(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
}
define <16 x i16> @test_paddw(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> *%a2) {
+; GENERIC-LABEL: test_paddw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpaddw %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpaddw (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_paddw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpaddw %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
@@ -142,6 +188,13 @@ define <16 x i16> @test_paddw(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> *%a2) {
}
define <4 x i64> @test_pand(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
+; GENERIC-LABEL: test_pand:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpand %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
+; GENERIC-NEXT: vpand (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
+; GENERIC-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_pand:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpand %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
@@ -163,6 +216,13 @@ define <4 x i64> @test_pand(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
}
define <4 x i64> @test_pandn(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
+; GENERIC-LABEL: test_pandn:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpandn %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
+; GENERIC-NEXT: vpandn (%rdi), %ymm0, %ymm1 # sched: [5:1.00]
+; GENERIC-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_pandn:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpandn %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
@@ -186,6 +246,12 @@ define <4 x i64> @test_pandn(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
}
define <8 x i32> @test_pmulld(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> *%a2) {
+; GENERIC-LABEL: test_pmulld:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpmulld %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
+; GENERIC-NEXT: vpmulld (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_pmulld:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpmulld %ymm1, %ymm0, %ymm0 # sched: [10:2.00]
@@ -204,6 +270,12 @@ define <8 x i32> @test_pmulld(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> *%a2) {
}
define <16 x i16> @test_pmullw(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> *%a2) {
+; GENERIC-LABEL: test_pmullw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpmullw %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
+; GENERIC-NEXT: vpmullw (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_pmullw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpmullw %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
@@ -222,6 +294,13 @@ define <16 x i16> @test_pmullw(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> *%a2)
}
define <4 x i64> @test_por(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
+; GENERIC-LABEL: test_por:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpor %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
+; GENERIC-NEXT: vpor (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
+; GENERIC-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_por:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpor %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
@@ -243,6 +322,12 @@ define <4 x i64> @test_por(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
}
define <32 x i8> @test_psubb(<32 x i8> %a0, <32 x i8> %a1, <32 x i8> *%a2) {
+; GENERIC-LABEL: test_psubb:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpsubb %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpsubb (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_psubb:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsubb %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
@@ -261,6 +346,12 @@ define <32 x i8> @test_psubb(<32 x i8> %a0, <32 x i8> %a1, <32 x i8> *%a2) {
}
define <8 x i32> @test_psubd(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> *%a2) {
+; GENERIC-LABEL: test_psubd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpsubd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpsubd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_psubd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsubd %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
@@ -279,6 +370,12 @@ define <8 x i32> @test_psubd(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> *%a2) {
}
define <4 x i64> @test_psubq(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
+; GENERIC-LABEL: test_psubq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpsubq %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpsubq (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_psubq:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsubq %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
@@ -297,6 +394,12 @@ define <4 x i64> @test_psubq(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
}
define <16 x i16> @test_psubw(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> *%a2) {
+; GENERIC-LABEL: test_psubw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpsubw %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: vpsubw (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_psubw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsubw %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
@@ -315,6 +418,13 @@ define <16 x i16> @test_psubw(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> *%a2) {
}
define <4 x i64> @test_pxor(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
+; GENERIC-LABEL: test_pxor:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: vpxor %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
+; GENERIC-NEXT: vpxor (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
+; GENERIC-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
; HASWELL-LABEL: test_pxor:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpxor %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
OpenPOWER on IntegriCloud