[llvm] e8d67ad - [X86] Disable autoupgrade support for avx512.mask.broadcasti32x2.* and avx512.mask.broadcastf32x2.*.
Craig Topper via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 24 12:35:48 PDT 2020
Author: Craig Topper
Date: 2020-03-24T12:35:24-07:00
New Revision: e8d67ada2df35ca6c70dbbe8185b0edbb18c1150
URL: https://github.com/llvm/llvm-project/commit/e8d67ada2df35ca6c70dbbe8185b0edbb18c1150
DIFF: https://github.com/llvm/llvm-project/commit/e8d67ada2df35ca6c70dbbe8185b0edbb18c1150.diff
LOG: [X86] Disable autoupgrade support for avx512.mask.broadcasti32x2.* and avx512.mask.broadcastf32x2.*.
These intrinsics take a v4i32/v4f32 input and are supposed to
broadcast elements 0 and 1. Instead the autoupgrade code was
broadcasting elements 0, 1, 2, and 3.
I could fix the autoupgrade, but since its been broken for years
it seemed better just to steer anyone still trying to use it away
completely.
Added:
Modified:
llvm/lib/IR/AutoUpgrade.cpp
llvm/test/CodeGen/X86/avx512dq-intrinsics-upgrade.ll
llvm/test/CodeGen/X86/avx512dqvl-intrinsics-upgrade.ll
Removed:
################################################################################
diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index 408c28a9bae7..de5b94403bd7 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -374,8 +374,14 @@ static bool ShouldUpgradeX86Intrinsic(Function *F, StringRef Name) {
Name.startswith("avx2.pblendd.") || // Added in 3.7
Name.startswith("avx.vbroadcastf128") || // Added in 4.0
Name == "avx2.vbroadcasti128" || // Added in 3.7
- Name.startswith("avx512.mask.broadcastf") || // Added in 6.0
- Name.startswith("avx512.mask.broadcasti") || // Added in 6.0
+ Name.startswith("avx512.mask.broadcastf32x4.") || // Added in 6.0
+ Name.startswith("avx512.mask.broadcastf64x2.") || // Added in 6.0
+ Name.startswith("avx512.mask.broadcastf32x8.") || // Added in 6.0
+ Name.startswith("avx512.mask.broadcastf64x4.") || // Added in 6.0
+ Name.startswith("avx512.mask.broadcasti32x4.") || // Added in 6.0
+ Name.startswith("avx512.mask.broadcasti64x2.") || // Added in 6.0
+ Name.startswith("avx512.mask.broadcasti32x8.") || // Added in 6.0
+ Name.startswith("avx512.mask.broadcasti64x4.") || // Added in 6.0
Name == "xop.vpcmov" || // Added in 3.8
Name == "xop.vpcmov.256" || // Added in 5.0
Name.startswith("avx512.mask.move.s") || // Added in 4.0
diff --git a/llvm/test/CodeGen/X86/avx512dq-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avx512dq-intrinsics-upgrade.ll
index 64659d455860..ab0cf02b4db9 100644
--- a/llvm/test/CodeGen/X86/avx512dq-intrinsics-upgrade.ll
+++ b/llvm/test/CodeGen/X86/avx512dq-intrinsics-upgrade.ll
@@ -441,74 +441,6 @@ define <8 x i64>@test_int_x86_avx512_mask_broadcasti64x2_512_load(<2 x i64>* %x0
ret <8 x i64> %res
}
-declare <16 x float> @llvm.x86.avx512.mask.broadcastf32x2.512(<4 x float>, <16 x float>, i16)
-
-define <16 x float>@test_int_x86_avx512_mask_broadcastf32x2_512(<4 x float> %x0, <16 x float> %x2, i16 %x3) {
-; X86-LABEL: test_int_x86_avx512_mask_broadcastf32x2_512:
-; X86: # %bb.0:
-; X86-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
-; X86-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 # encoding: [0xc4,0xe3,0x7d,0x18,0xc0,0x01]
-; X86-NEXT: vinsertf64x4 $1, %ymm0, %zmm0, %zmm2 # encoding: [0x62,0xf3,0xfd,0x48,0x1a,0xd0,0x01]
-; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
-; X86-NEXT: vinsertf32x8 $1, %ymm0, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf3,0x7d,0x49,0x1a,0xc8,0x01]
-; X86-NEXT: vinsertf32x8 $1, %ymm0, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0xc9,0x1a,0xc0,0x01]
-; X86-NEXT: vaddps %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x74,0x48,0x58,0xc0]
-; X86-NEXT: vaddps %zmm2, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x58,0xc2]
-; X86-NEXT: retl # encoding: [0xc3]
-;
-; X64-LABEL: test_int_x86_avx512_mask_broadcastf32x2_512:
-; X64: # %bb.0:
-; X64-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
-; X64-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 # encoding: [0xc4,0xe3,0x7d,0x18,0xc0,0x01]
-; X64-NEXT: vinsertf64x4 $1, %ymm0, %zmm0, %zmm2 # encoding: [0x62,0xf3,0xfd,0x48,0x1a,0xd0,0x01]
-; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT: vinsertf32x8 $1, %ymm0, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf3,0x7d,0x49,0x1a,0xc8,0x01]
-; X64-NEXT: vinsertf32x8 $1, %ymm0, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0xc9,0x1a,0xc0,0x01]
-; X64-NEXT: vaddps %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x74,0x48,0x58,0xc0]
-; X64-NEXT: vaddps %zmm2, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x58,0xc2]
-; X64-NEXT: retq # encoding: [0xc3]
- %res = call <16 x float> @llvm.x86.avx512.mask.broadcastf32x2.512(<4 x float> %x0, <16 x float> %x2, i16 %x3)
- %res1 = call <16 x float> @llvm.x86.avx512.mask.broadcastf32x2.512(<4 x float> %x0, <16 x float> zeroinitializer, i16 %x3)
- %res2 = call <16 x float> @llvm.x86.avx512.mask.broadcastf32x2.512(<4 x float> %x0, <16 x float> %x2, i16 -1)
- %res3 = fadd <16 x float> %res, %res1
- %res4 = fadd <16 x float> %res3, %res2
- ret <16 x float> %res4
-}
-
-declare <16 x i32> @llvm.x86.avx512.mask.broadcasti32x2.512(<4 x i32>, <16 x i32>, i16)
-
-define <16 x i32>@test_int_x86_avx512_mask_broadcasti32x2_512(<4 x i32> %x0, <16 x i32> %x2, i16 %x3) {
-; X86-LABEL: test_int_x86_avx512_mask_broadcasti32x2_512:
-; X86: # %bb.0:
-; X86-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
-; X86-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0 # encoding: [0xc4,0xe3,0x7d,0x38,0xc0,0x01]
-; X86-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm2 # encoding: [0x62,0xf3,0xfd,0x48,0x3a,0xd0,0x01]
-; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
-; X86-NEXT: vinserti32x8 $1, %ymm0, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf3,0x7d,0x49,0x3a,0xc8,0x01]
-; X86-NEXT: vinserti32x8 $1, %ymm0, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0xc9,0x3a,0xc0,0x01]
-; X86-NEXT: vpaddd %zmm2, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfe,0xc2]
-; X86-NEXT: vpaddd %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfe,0xc0]
-; X86-NEXT: retl # encoding: [0xc3]
-;
-; X64-LABEL: test_int_x86_avx512_mask_broadcasti32x2_512:
-; X64: # %bb.0:
-; X64-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
-; X64-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0 # encoding: [0xc4,0xe3,0x7d,0x38,0xc0,0x01]
-; X64-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm2 # encoding: [0x62,0xf3,0xfd,0x48,0x3a,0xd0,0x01]
-; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT: vinserti32x8 $1, %ymm0, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf3,0x7d,0x49,0x3a,0xc8,0x01]
-; X64-NEXT: vinserti32x8 $1, %ymm0, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0xc9,0x3a,0xc0,0x01]
-; X64-NEXT: vpaddd %zmm2, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfe,0xc2]
-; X64-NEXT: vpaddd %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfe,0xc0]
-; X64-NEXT: retq # encoding: [0xc3]
- %res = call <16 x i32> @llvm.x86.avx512.mask.broadcasti32x2.512(<4 x i32> %x0, <16 x i32> %x2, i16 %x3)
- %res1 = call <16 x i32> @llvm.x86.avx512.mask.broadcasti32x2.512(<4 x i32> %x0, <16 x i32> zeroinitializer, i16 %x3)
- %res2 = call <16 x i32> @llvm.x86.avx512.mask.broadcasti32x2.512(<4 x i32> %x0, <16 x i32> %x2, i16 -1)
- %res3 = add <16 x i32> %res, %res1
- %res4 = add <16 x i32> %res3, %res2
- ret <16 x i32> %res4
-}
-
declare i16 @llvm.x86.avx512.cvtd2mask.512(<16 x i32>)
define i16 at test_int_x86_avx512_cvtd2mask_512(<16 x i32> %x0) {
diff --git a/llvm/test/CodeGen/X86/avx512dqvl-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avx512dqvl-intrinsics-upgrade.ll
index 879a1643325d..c2afbe09e145 100644
--- a/llvm/test/CodeGen/X86/avx512dqvl-intrinsics-upgrade.ll
+++ b/llvm/test/CodeGen/X86/avx512dqvl-intrinsics-upgrade.ll
@@ -2658,104 +2658,6 @@ define <4 x i64>@test_int_x86_avx512_mask_broadcasti64x2_256_load(<2 x i64>* %x0
ret <4 x i64> %res
}
-declare <8 x float> @llvm.x86.avx512.mask.broadcastf32x2.256(<4 x float>, <8 x float>, i8)
-
-define <8 x float>@test_int_x86_avx512_mask_broadcastf32x2_256(<4 x float> %x0, <8 x float> %x2, i8 %x3) {
-; X86-LABEL: test_int_x86_avx512_mask_broadcastf32x2_256:
-; X86: # %bb.0:
-; X86-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
-; X86-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm2 # EVEX TO VEX Compression encoding: [0xc4,0xe3,0x7d,0x18,0xd0,0x01]
-; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT: vinsertf32x4 $1, %xmm0, %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf3,0x7d,0x29,0x18,0xc8,0x01]
-; X86-NEXT: vinsertf32x4 $1, %xmm0, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0xa9,0x18,0xc0,0x01]
-; X86-NEXT: vaddps %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf4,0x58,0xc0]
-; X86-NEXT: vaddps %ymm2, %ymm0, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x58,0xc2]
-; X86-NEXT: retl # encoding: [0xc3]
-;
-; X64-LABEL: test_int_x86_avx512_mask_broadcastf32x2_256:
-; X64: # %bb.0:
-; X64-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
-; X64-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm2 # EVEX TO VEX Compression encoding: [0xc4,0xe3,0x7d,0x18,0xd0,0x01]
-; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT: vinsertf32x4 $1, %xmm0, %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf3,0x7d,0x29,0x18,0xc8,0x01]
-; X64-NEXT: vinsertf32x4 $1, %xmm0, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0xa9,0x18,0xc0,0x01]
-; X64-NEXT: vaddps %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf4,0x58,0xc0]
-; X64-NEXT: vaddps %ymm2, %ymm0, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x58,0xc2]
-; X64-NEXT: retq # encoding: [0xc3]
- %res = call <8 x float> @llvm.x86.avx512.mask.broadcastf32x2.256(<4 x float> %x0, <8 x float> %x2, i8 %x3)
- %res1 = call <8 x float> @llvm.x86.avx512.mask.broadcastf32x2.256(<4 x float> %x0, <8 x float> zeroinitializer, i8 %x3)
- %res2 = call <8 x float> @llvm.x86.avx512.mask.broadcastf32x2.256(<4 x float> %x0, <8 x float> %x2, i8 -1)
- %res3 = fadd <8 x float> %res, %res1
- %res4 = fadd <8 x float> %res3, %res2
- ret <8 x float> %res4
-}
-
-declare <8 x i32> @llvm.x86.avx512.mask.broadcasti32x2.256(<4 x i32>, <8 x i32>, i8)
-
-define <8 x i32>@test_int_x86_avx512_mask_broadcasti32x2_256(<4 x i32> %x0, <8 x i32> %x2, i8 %x3, i64 * %y_ptr) {
-; X86-LABEL: test_int_x86_avx512_mask_broadcasti32x2_256:
-; X86: # %bb.0:
-; X86-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
-; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
-; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT: vbroadcasti32x2 (%eax), %ymm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x59,0x08]
-; X86-NEXT: # ymm1 {%k1} = mem[0,1,0,1,0,1,0,1]
-; X86-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm2 # EVEX TO VEX Compression encoding: [0xc4,0xe3,0x7d,0x38,0xd0,0x01]
-; X86-NEXT: vinserti32x4 $1, %xmm0, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0xa9,0x38,0xc0,0x01]
-; X86-NEXT: vpaddd %ymm2, %ymm0, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0xfe,0xc2]
-; X86-NEXT: vpaddd %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xfe,0xc0]
-; X86-NEXT: retl # encoding: [0xc3]
-;
-; X64-LABEL: test_int_x86_avx512_mask_broadcasti32x2_256:
-; X64: # %bb.0:
-; X64-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
-; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT: vbroadcasti32x2 (%rsi), %ymm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x59,0x0e]
-; X64-NEXT: # ymm1 {%k1} = mem[0,1,0,1,0,1,0,1]
-; X64-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm2 # EVEX TO VEX Compression encoding: [0xc4,0xe3,0x7d,0x38,0xd0,0x01]
-; X64-NEXT: vinserti32x4 $1, %xmm0, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0xa9,0x38,0xc0,0x01]
-; X64-NEXT: vpaddd %ymm2, %ymm0, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0xfe,0xc2]
-; X64-NEXT: vpaddd %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xfe,0xc0]
-; X64-NEXT: retq # encoding: [0xc3]
- %y_64 = load i64, i64 * %y_ptr
- %y_v2i64 = insertelement <2 x i64> undef, i64 %y_64, i32 0
- %y = bitcast <2 x i64> %y_v2i64 to <4 x i32>
- %res = call <8 x i32> @llvm.x86.avx512.mask.broadcasti32x2.256(<4 x i32> %y, <8 x i32> %x2, i8 %x3)
- %res1 = call <8 x i32> @llvm.x86.avx512.mask.broadcasti32x2.256(<4 x i32> %x0, <8 x i32> zeroinitializer, i8 %x3)
- %res2 = call <8 x i32> @llvm.x86.avx512.mask.broadcasti32x2.256(<4 x i32> %x0, <8 x i32> %x2, i8 -1)
- %res3 = add <8 x i32> %res, %res1
- %res4 = add <8 x i32> %res3, %res2
- ret <8 x i32> %res4
-}
-
-declare <4 x i32> @llvm.x86.avx512.mask.broadcasti32x2.128(<4 x i32>, <4 x i32>, i8)
-
-define <4 x i32>@test_int_x86_avx512_mask_broadcasti32x2_128(<4 x i32> %x0, <4 x i32> %x2, i8 %x3) {
-; X86-LABEL: test_int_x86_avx512_mask_broadcasti32x2_128:
-; X86: # %bb.0:
-; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT: vmovdqa32 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x6f,0xc8]
-; X86-NEXT: vmovdqa32 %xmm0, %xmm2 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x6f,0xd0]
-; X86-NEXT: vpaddd %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xfe,0xc0]
-; X86-NEXT: vpaddd %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc0]
-; X86-NEXT: retl # encoding: [0xc3]
-;
-; X64-LABEL: test_int_x86_avx512_mask_broadcasti32x2_128:
-; X64: # %bb.0:
-; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT: vmovdqa32 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x6f,0xc8]
-; X64-NEXT: vmovdqa32 %xmm0, %xmm2 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x6f,0xd0]
-; X64-NEXT: vpaddd %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xfe,0xc0]
-; X64-NEXT: vpaddd %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc0]
-; X64-NEXT: retq # encoding: [0xc3]
- %res = call <4 x i32> @llvm.x86.avx512.mask.broadcasti32x2.128(<4 x i32> %x0, <4 x i32> %x2, i8 %x3)
- %res1 = call <4 x i32> @llvm.x86.avx512.mask.broadcasti32x2.128(<4 x i32> %x0, <4 x i32> zeroinitializer, i8 %x3)
- %res2 = call <4 x i32> @llvm.x86.avx512.mask.broadcasti32x2.128(<4 x i32> %x0, <4 x i32> %x2, i8 -1)
- %res3 = add <4 x i32> %res, %res1
- %res4 = add <4 x i32> %res3, %res2
- ret <4 x i32> %res4
-}
-
declare i8 @llvm.x86.avx512.cvtd2mask.128(<4 x i32>)
define i8 at test_int_x86_avx512_cvtd2mask_128(<4 x i32> %x0) {
More information about the llvm-commits
mailing list