Google Git
Sign in
cos / third_party / overlays / chromiumos-overlay / 41ae31d3a453346e1bd695573d4d4ca9e715c92a / . / sys-devel / llvm / files / cherry / 765ef4bb2af604ea2bbd6c1bffaa6e1600804c9e.patch
blob: a1bff4c93416a9684f5cdd56904c771f3a4bd5c2 [file] [log] [blame]
commit 765ef4bb2af604ea2bbd6c1bffaa6e1600804c9e
Author: Craig Topper <craig.topper@sifive.com>
Date: Wed Jun 9 07:08:16 2021 -0700
[X86] Check destination element type before forming VTRUNCS/VTRUNCUS in combineTruncateWithSat.
Fixes crash reported here https://reviews.llvm.org/D73607
Using a store to keep the trunc intact. Returning v16i24 would
cause the trunc to be optimized away in SelectionDAGBuilder.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D103940
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index b89e1674d7f3..90babf3b87b6 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -45273,7 +45273,8 @@ static SDValue combineTruncateWithSat(SDValue In, EVT VT, const SDLoc &DL,
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
if (TLI.isTypeLegal(InVT) && InVT.isVector() && SVT != MVT::i1 &&
- Subtarget.hasAVX512() && (InSVT != MVT::i16 || Subtarget.hasBWI())) {
+ Subtarget.hasAVX512() && (InSVT != MVT::i16 || Subtarget.hasBWI()) &&
+ (SVT == MVT::i32 || SVT == MVT::i16 || SVT == MVT::i8)) {
unsigned TruncOpc = 0;
SDValue SatVal;
if (auto SSatVal = detectSSatPattern(In, VT)) {
diff --git a/llvm/test/CodeGen/X86/vector-trunc-ssat.ll b/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
index b6ac19660088..924cfc169086 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
@@ -5631,3 +5631,678 @@ define <32 x i8> @trunc_ssat_v32i32_v32i8(<32 x i32>* %p0) "min-legal-vector-wid
%5 = trunc <32 x i32> %4 to <32 x i8>
ret <32 x i8> %5
}
+
+; This used to crash with avx512 due because we were missing a check for
+; unsupported element types like i24.
+define void @trunc_ssat_v16i32_v16i24(<16 x i32> %x, <16 x i24>* %y) nounwind {
+; SSE2-LABEL: trunc_ssat_v16i32_v16i24:
+; SSE2: # %bb.0:
+; SSE2-NEXT: pushq %rbp
+; SSE2-NEXT: pushq %r15
+; SSE2-NEXT: pushq %r14
+; SSE2-NEXT: pushq %r12
+; SSE2-NEXT: pushq %rbx
+; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [8388607,8388607,8388607,8388607]
+; SSE2-NEXT: movdqa %xmm5, %xmm4
+; SSE2-NEXT: pcmpgtd %xmm3, %xmm4
+; SSE2-NEXT: pand %xmm4, %xmm3
+; SSE2-NEXT: pandn %xmm5, %xmm4
+; SSE2-NEXT: por %xmm3, %xmm4
+; SSE2-NEXT: movdqa %xmm5, %xmm3
+; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
+; SSE2-NEXT: pand %xmm3, %xmm2
+; SSE2-NEXT: pandn %xmm5, %xmm3
+; SSE2-NEXT: por %xmm2, %xmm3
+; SSE2-NEXT: movdqa %xmm5, %xmm2
+; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
+; SSE2-NEXT: pand %xmm2, %xmm1
+; SSE2-NEXT: pandn %xmm5, %xmm2
+; SSE2-NEXT: por %xmm1, %xmm2
+; SSE2-NEXT: movdqa %xmm5, %xmm1
+; SSE2-NEXT: pcmpgtd %xmm0, %xmm1
+; SSE2-NEXT: pand %xmm1, %xmm0
+; SSE2-NEXT: pandn %xmm5, %xmm1
+; SSE2-NEXT: por %xmm0, %xmm1
+; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [4286578688,4286578688,4286578688,4286578688]
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: pcmpgtd %xmm5, %xmm0
+; SSE2-NEXT: pand %xmm0, %xmm1
+; SSE2-NEXT: pandn %xmm5, %xmm0
+; SSE2-NEXT: por %xmm1, %xmm0
+; SSE2-NEXT: movdqa %xmm2, %xmm1
+; SSE2-NEXT: pcmpgtd %xmm5, %xmm1
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: pandn %xmm5, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm3, %xmm2
+; SSE2-NEXT: pcmpgtd %xmm5, %xmm2
+; SSE2-NEXT: pand %xmm2, %xmm3
+; SSE2-NEXT: pandn %xmm5, %xmm2
+; SSE2-NEXT: por %xmm3, %xmm2
+; SSE2-NEXT: movdqa %xmm4, %xmm3
+; SSE2-NEXT: pcmpgtd %xmm5, %xmm3
+; SSE2-NEXT: pand %xmm3, %xmm4
+; SSE2-NEXT: pandn %xmm5, %xmm3
+; SSE2-NEXT: por %xmm4, %xmm3
+; SSE2-NEXT: movd %xmm3, %edx
+; SSE2-NEXT: movw %dx, 36(%rdi)
+; SSE2-NEXT: movd %xmm2, %ecx
+; SSE2-NEXT: movw %cx, 24(%rdi)
+; SSE2-NEXT: movd %xmm1, %eax
+; SSE2-NEXT: movw %ax, 12(%rdi)
+; SSE2-NEXT: movd %xmm0, %r8d
+; SSE2-NEXT: movw %r8w, (%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[3,3,3,3]
+; SSE2-NEXT: movd %xmm4, %r9d
+; SSE2-NEXT: movw %r9w, 45(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[2,3,2,3]
+; SSE2-NEXT: movd %xmm4, %r10d
+; SSE2-NEXT: movw %r10w, 42(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,1,1]
+; SSE2-NEXT: movd %xmm3, %r11d
+; SSE2-NEXT: movw %r11w, 39(%rdi)
+; SSE2-NEXT: shrl $16, %edx
+; SSE2-NEXT: movb %dl, 38(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[3,3,3,3]
+; SSE2-NEXT: movd %xmm3, %r14d
+; SSE2-NEXT: movw %r14w, 33(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[2,3,2,3]
+; SSE2-NEXT: movd %xmm3, %r15d
+; SSE2-NEXT: movw %r15w, 30(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,1,1]
+; SSE2-NEXT: movd %xmm2, %r12d
+; SSE2-NEXT: movw %r12w, 27(%rdi)
+; SSE2-NEXT: shrl $16, %ecx
+; SSE2-NEXT: movb %cl, 26(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[3,3,3,3]
+; SSE2-NEXT: movd %xmm2, %esi
+; SSE2-NEXT: movw %si, 21(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
+; SSE2-NEXT: movd %xmm2, %ebx
+; SSE2-NEXT: movw %bx, 18(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,1,1]
+; SSE2-NEXT: movd %xmm1, %ebp
+; SSE2-NEXT: movw %bp, 15(%rdi)
+; SSE2-NEXT: shrl $16, %eax
+; SSE2-NEXT: movb %al, 14(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[3,3,3,3]
+; SSE2-NEXT: movd %xmm1, %eax
+; SSE2-NEXT: movw %ax, 9(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; SSE2-NEXT: movd %xmm1, %ecx
+; SSE2-NEXT: movw %cx, 6(%rdi)
+; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
+; SSE2-NEXT: movd %xmm0, %edx
+; SSE2-NEXT: movw %dx, 3(%rdi)
+; SSE2-NEXT: shrl $16, %r8d
+; SSE2-NEXT: movb %r8b, 2(%rdi)
+; SSE2-NEXT: shrl $16, %r9d
+; SSE2-NEXT: movb %r9b, 47(%rdi)
+; SSE2-NEXT: shrl $16, %r10d
+; SSE2-NEXT: movb %r10b, 44(%rdi)
+; SSE2-NEXT: shrl $16, %r11d
+; SSE2-NEXT: movb %r11b, 41(%rdi)
+; SSE2-NEXT: shrl $16, %r14d
+; SSE2-NEXT: movb %r14b, 35(%rdi)
+; SSE2-NEXT: shrl $16, %r15d
+; SSE2-NEXT: movb %r15b, 32(%rdi)
+; SSE2-NEXT: shrl $16, %r12d
+; SSE2-NEXT: movb %r12b, 29(%rdi)
+; SSE2-NEXT: shrl $16, %esi
+; SSE2-NEXT: movb %sil, 23(%rdi)
+; SSE2-NEXT: shrl $16, %ebx
+; SSE2-NEXT: movb %bl, 20(%rdi)
+; SSE2-NEXT: shrl $16, %ebp
+; SSE2-NEXT: movb %bpl, 17(%rdi)
+; SSE2-NEXT: shrl $16, %eax
+; SSE2-NEXT: movb %al, 11(%rdi)
+; SSE2-NEXT: shrl $16, %ecx
+; SSE2-NEXT: movb %cl, 8(%rdi)
+; SSE2-NEXT: shrl $16, %edx
+; SSE2-NEXT: movb %dl, 5(%rdi)
+; SSE2-NEXT: popq %rbx
+; SSE2-NEXT: popq %r12
+; SSE2-NEXT: popq %r14
+; SSE2-NEXT: popq %r15
+; SSE2-NEXT: popq %rbp
+; SSE2-NEXT: retq
+;
+; SSSE3-LABEL: trunc_ssat_v16i32_v16i24:
+; SSSE3: # %bb.0:
+; SSSE3-NEXT: pushq %rbp
+; SSSE3-NEXT: pushq %r15
+; SSSE3-NEXT: pushq %r14
+; SSSE3-NEXT: pushq %r12
+; SSSE3-NEXT: pushq %rbx
+; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [8388607,8388607,8388607,8388607]
+; SSSE3-NEXT: movdqa %xmm5, %xmm4
+; SSSE3-NEXT: pcmpgtd %xmm3, %xmm4
+; SSSE3-NEXT: pand %xmm4, %xmm3
+; SSSE3-NEXT: pandn %xmm5, %xmm4
+; SSSE3-NEXT: por %xmm3, %xmm4
+; SSSE3-NEXT: movdqa %xmm5, %xmm3
+; SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
+; SSSE3-NEXT: pand %xmm3, %xmm2
+; SSSE3-NEXT: pandn %xmm5, %xmm3
+; SSSE3-NEXT: por %xmm2, %xmm3
+; SSSE3-NEXT: movdqa %xmm5, %xmm2
+; SSSE3-NEXT: pcmpgtd %xmm1, %xmm2
+; SSSE3-NEXT: pand %xmm2, %xmm1
+; SSSE3-NEXT: pandn %xmm5, %xmm2
+; SSSE3-NEXT: por %xmm1, %xmm2
+; SSSE3-NEXT: movdqa %xmm5, %xmm1
+; SSSE3-NEXT: pcmpgtd %xmm0, %xmm1
+; SSSE3-NEXT: pand %xmm1, %xmm0
+; SSSE3-NEXT: pandn %xmm5, %xmm1
+; SSSE3-NEXT: por %xmm0, %xmm1
+; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [4286578688,4286578688,4286578688,4286578688]
+; SSSE3-NEXT: movdqa %xmm1, %xmm0
+; SSSE3-NEXT: pcmpgtd %xmm5, %xmm0
+; SSSE3-NEXT: pand %xmm0, %xmm1
+; SSSE3-NEXT: pandn %xmm5, %xmm0
+; SSSE3-NEXT: por %xmm1, %xmm0
+; SSSE3-NEXT: movdqa %xmm2, %xmm1
+; SSSE3-NEXT: pcmpgtd %xmm5, %xmm1
+; SSSE3-NEXT: pand %xmm1, %xmm2
+; SSSE3-NEXT: pandn %xmm5, %xmm1
+; SSSE3-NEXT: por %xmm2, %xmm1
+; SSSE3-NEXT: movdqa %xmm3, %xmm2
+; SSSE3-NEXT: pcmpgtd %xmm5, %xmm2
+; SSSE3-NEXT: pand %xmm2, %xmm3
+; SSSE3-NEXT: pandn %xmm5, %xmm2
+; SSSE3-NEXT: por %xmm3, %xmm2
+; SSSE3-NEXT: movdqa %xmm4, %xmm3
+; SSSE3-NEXT: pcmpgtd %xmm5, %xmm3
+; SSSE3-NEXT: pand %xmm3, %xmm4
+; SSSE3-NEXT: pandn %xmm5, %xmm3
+; SSSE3-NEXT: por %xmm4, %xmm3
+; SSSE3-NEXT: movd %xmm3, %edx
+; SSSE3-NEXT: movw %dx, 36(%rdi)
+; SSSE3-NEXT: movd %xmm2, %ecx
+; SSSE3-NEXT: movw %cx, 24(%rdi)
+; SSSE3-NEXT: movd %xmm1, %eax
+; SSSE3-NEXT: movw %ax, 12(%rdi)
+; SSSE3-NEXT: movd %xmm0, %r8d
+; SSSE3-NEXT: movw %r8w, (%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm3[3,3,3,3]
+; SSSE3-NEXT: movd %xmm4, %r9d
+; SSSE3-NEXT: movw %r9w, 45(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm3[2,3,2,3]
+; SSSE3-NEXT: movd %xmm4, %r10d
+; SSSE3-NEXT: movw %r10w, 42(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,1,1]
+; SSSE3-NEXT: movd %xmm3, %r11d
+; SSSE3-NEXT: movw %r11w, 39(%rdi)
+; SSSE3-NEXT: shrl $16, %edx
+; SSSE3-NEXT: movb %dl, 38(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[3,3,3,3]
+; SSSE3-NEXT: movd %xmm3, %r14d
+; SSSE3-NEXT: movw %r14w, 33(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[2,3,2,3]
+; SSSE3-NEXT: movd %xmm3, %r15d
+; SSSE3-NEXT: movw %r15w, 30(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,1,1]
+; SSSE3-NEXT: movd %xmm2, %r12d
+; SSSE3-NEXT: movw %r12w, 27(%rdi)
+; SSSE3-NEXT: shrl $16, %ecx
+; SSSE3-NEXT: movb %cl, 26(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[3,3,3,3]
+; SSSE3-NEXT: movd %xmm2, %esi
+; SSSE3-NEXT: movw %si, 21(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
+; SSSE3-NEXT: movd %xmm2, %ebx
+; SSSE3-NEXT: movw %bx, 18(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,1,1]
+; SSSE3-NEXT: movd %xmm1, %ebp
+; SSSE3-NEXT: movw %bp, 15(%rdi)
+; SSSE3-NEXT: shrl $16, %eax
+; SSSE3-NEXT: movb %al, 14(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm0[3,3,3,3]
+; SSSE3-NEXT: movd %xmm1, %eax
+; SSSE3-NEXT: movw %ax, 9(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; SSSE3-NEXT: movd %xmm1, %ecx
+; SSSE3-NEXT: movw %cx, 6(%rdi)
+; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
+; SSSE3-NEXT: movd %xmm0, %edx
+; SSSE3-NEXT: movw %dx, 3(%rdi)
+; SSSE3-NEXT: shrl $16, %r8d
+; SSSE3-NEXT: movb %r8b, 2(%rdi)
+; SSSE3-NEXT: shrl $16, %r9d
+; SSSE3-NEXT: movb %r9b, 47(%rdi)
+; SSSE3-NEXT: shrl $16, %r10d
+; SSSE3-NEXT: movb %r10b, 44(%rdi)
+; SSSE3-NEXT: shrl $16, %r11d
+; SSSE3-NEXT: movb %r11b, 41(%rdi)
+; SSSE3-NEXT: shrl $16, %r14d
+; SSSE3-NEXT: movb %r14b, 35(%rdi)
+; SSSE3-NEXT: shrl $16, %r15d
+; SSSE3-NEXT: movb %r15b, 32(%rdi)
+; SSSE3-NEXT: shrl $16, %r12d
+; SSSE3-NEXT: movb %r12b, 29(%rdi)
+; SSSE3-NEXT: shrl $16, %esi
+; SSSE3-NEXT: movb %sil, 23(%rdi)
+; SSSE3-NEXT: shrl $16, %ebx
+; SSSE3-NEXT: movb %bl, 20(%rdi)
+; SSSE3-NEXT: shrl $16, %ebp
+; SSSE3-NEXT: movb %bpl, 17(%rdi)
+; SSSE3-NEXT: shrl $16, %eax
+; SSSE3-NEXT: movb %al, 11(%rdi)
+; SSSE3-NEXT: shrl $16, %ecx
+; SSSE3-NEXT: movb %cl, 8(%rdi)
+; SSSE3-NEXT: shrl $16, %edx
+; SSSE3-NEXT: movb %dl, 5(%rdi)
+; SSSE3-NEXT: popq %rbx
+; SSSE3-NEXT: popq %r12
+; SSSE3-NEXT: popq %r14
+; SSSE3-NEXT: popq %r15
+; SSSE3-NEXT: popq %rbp
+; SSSE3-NEXT: retq
+;
+; SSE41-LABEL: trunc_ssat_v16i32_v16i24:
+; SSE41: # %bb.0:
+; SSE41-NEXT: movdqa {{.*#+}} xmm4 = [8388607,8388607,8388607,8388607]
+; SSE41-NEXT: pminsd %xmm4, %xmm3
+; SSE41-NEXT: pminsd %xmm4, %xmm2
+; SSE41-NEXT: pminsd %xmm4, %xmm1
+; SSE41-NEXT: pminsd %xmm4, %xmm0
+; SSE41-NEXT: movdqa {{.*#+}} xmm4 = [4286578688,4286578688,4286578688,4286578688]
+; SSE41-NEXT: pmaxsd %xmm4, %xmm0
+; SSE41-NEXT: pmaxsd %xmm4, %xmm1
+; SSE41-NEXT: pmaxsd %xmm4, %xmm2
+; SSE41-NEXT: pmaxsd %xmm4, %xmm3
+; SSE41-NEXT: pextrd $3, %xmm3, %eax
+; SSE41-NEXT: movw %ax, 45(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 47(%rdi)
+; SSE41-NEXT: pextrd $2, %xmm3, %eax
+; SSE41-NEXT: movw %ax, 42(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 44(%rdi)
+; SSE41-NEXT: pextrd $1, %xmm3, %eax
+; SSE41-NEXT: movw %ax, 39(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 41(%rdi)
+; SSE41-NEXT: movd %xmm3, %eax
+; SSE41-NEXT: movw %ax, 36(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 38(%rdi)
+; SSE41-NEXT: pextrd $3, %xmm2, %eax
+; SSE41-NEXT: movw %ax, 33(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 35(%rdi)
+; SSE41-NEXT: pextrd $2, %xmm2, %eax
+; SSE41-NEXT: movw %ax, 30(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 32(%rdi)
+; SSE41-NEXT: pextrd $1, %xmm2, %eax
+; SSE41-NEXT: movw %ax, 27(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 29(%rdi)
+; SSE41-NEXT: movd %xmm2, %eax
+; SSE41-NEXT: movw %ax, 24(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 26(%rdi)
+; SSE41-NEXT: pextrd $3, %xmm1, %eax
+; SSE41-NEXT: movw %ax, 21(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 23(%rdi)
+; SSE41-NEXT: pextrd $2, %xmm1, %eax
+; SSE41-NEXT: movw %ax, 18(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 20(%rdi)
+; SSE41-NEXT: pextrd $1, %xmm1, %eax
+; SSE41-NEXT: movw %ax, 15(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 17(%rdi)
+; SSE41-NEXT: movd %xmm1, %eax
+; SSE41-NEXT: movw %ax, 12(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 14(%rdi)
+; SSE41-NEXT: pextrd $3, %xmm0, %eax
+; SSE41-NEXT: movw %ax, 9(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 11(%rdi)
+; SSE41-NEXT: pextrd $2, %xmm0, %eax
+; SSE41-NEXT: movw %ax, 6(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 8(%rdi)
+; SSE41-NEXT: pextrd $1, %xmm0, %eax
+; SSE41-NEXT: movw %ax, 3(%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 5(%rdi)
+; SSE41-NEXT: movd %xmm0, %eax
+; SSE41-NEXT: movw %ax, (%rdi)
+; SSE41-NEXT: shrl $16, %eax
+; SSE41-NEXT: movb %al, 2(%rdi)
+; SSE41-NEXT: retq
+;
+; AVX1-LABEL: trunc_ssat_v16i32_v16i24:
+; AVX1: # %bb.0:
+; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [8388607,8388607,8388607,8388607]
+; AVX1-NEXT: vpminsd %xmm3, %xmm2, %xmm4
+; AVX1-NEXT: vpminsd %xmm3, %xmm1, %xmm2
+; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT: vpminsd %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpminsd %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [4286578688,4286578688,4286578688,4286578688]
+; AVX1-NEXT: vpmaxsd %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpmaxsd %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpmaxsd %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vpmaxsd %xmm3, %xmm4, %xmm3
+; AVX1-NEXT: vpextrd $3, %xmm3, %eax
+; AVX1-NEXT: movw %ax, 45(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 47(%rdi)
+; AVX1-NEXT: vpextrd $2, %xmm3, %eax
+; AVX1-NEXT: movw %ax, 42(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 44(%rdi)
+; AVX1-NEXT: vpextrd $1, %xmm3, %eax
+; AVX1-NEXT: movw %ax, 39(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 41(%rdi)
+; AVX1-NEXT: vmovd %xmm3, %eax
+; AVX1-NEXT: movw %ax, 36(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 38(%rdi)
+; AVX1-NEXT: vpextrd $3, %xmm2, %eax
+; AVX1-NEXT: movw %ax, 33(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 35(%rdi)
+; AVX1-NEXT: vpextrd $2, %xmm2, %eax
+; AVX1-NEXT: movw %ax, 30(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 32(%rdi)
+; AVX1-NEXT: vpextrd $1, %xmm2, %eax
+; AVX1-NEXT: movw %ax, 27(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 29(%rdi)
+; AVX1-NEXT: vmovd %xmm2, %eax
+; AVX1-NEXT: movw %ax, 24(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 26(%rdi)
+; AVX1-NEXT: vpextrd $3, %xmm1, %eax
+; AVX1-NEXT: movw %ax, 21(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 23(%rdi)
+; AVX1-NEXT: vpextrd $2, %xmm1, %eax
+; AVX1-NEXT: movw %ax, 18(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 20(%rdi)
+; AVX1-NEXT: vpextrd $1, %xmm1, %eax
+; AVX1-NEXT: movw %ax, 15(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 17(%rdi)
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: movw %ax, 12(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 14(%rdi)
+; AVX1-NEXT: vpextrd $3, %xmm0, %eax
+; AVX1-NEXT: movw %ax, 9(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 11(%rdi)
+; AVX1-NEXT: vpextrd $2, %xmm0, %eax
+; AVX1-NEXT: movw %ax, 6(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 8(%rdi)
+; AVX1-NEXT: vpextrd $1, %xmm0, %eax
+; AVX1-NEXT: movw %ax, 3(%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 5(%rdi)
+; AVX1-NEXT: vmovd %xmm0, %eax
+; AVX1-NEXT: movw %ax, (%rdi)
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: movb %al, 2(%rdi)
+; AVX1-NEXT: vzeroupper
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: trunc_ssat_v16i32_v16i24:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm2 = [8388607,8388607,8388607,8388607,8388607,8388607,8388607,8388607]
+; AVX2-NEXT: vpminsd %ymm2, %ymm1, %ymm1
+; AVX2-NEXT: vpminsd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm2 = [4286578688,4286578688,4286578688,4286578688,4286578688,4286578688,4286578688,4286578688]
+; AVX2-NEXT: vpmaxsd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT: vpmaxsd %ymm2, %ymm1, %ymm1
+; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT: vpextrd $3, %xmm2, %eax
+; AVX2-NEXT: movw %ax, 45(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 47(%rdi)
+; AVX2-NEXT: vpextrd $2, %xmm2, %eax
+; AVX2-NEXT: movw %ax, 42(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 44(%rdi)
+; AVX2-NEXT: vpextrd $1, %xmm2, %eax
+; AVX2-NEXT: movw %ax, 39(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 41(%rdi)
+; AVX2-NEXT: vmovd %xmm2, %eax
+; AVX2-NEXT: movw %ax, 36(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 38(%rdi)
+; AVX2-NEXT: vpextrd $3, %xmm1, %eax
+; AVX2-NEXT: movw %ax, 33(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 35(%rdi)
+; AVX2-NEXT: vpextrd $2, %xmm1, %eax
+; AVX2-NEXT: movw %ax, 30(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 32(%rdi)
+; AVX2-NEXT: vpextrd $1, %xmm1, %eax
+; AVX2-NEXT: movw %ax, 27(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 29(%rdi)
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: movw %ax, 24(%rdi)
+; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 26(%rdi)
+; AVX2-NEXT: vpextrd $3, %xmm1, %eax
+; AVX2-NEXT: movw %ax, 21(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 23(%rdi)
+; AVX2-NEXT: vpextrd $2, %xmm1, %eax
+; AVX2-NEXT: movw %ax, 18(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 20(%rdi)
+; AVX2-NEXT: vpextrd $1, %xmm1, %eax
+; AVX2-NEXT: movw %ax, 15(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 17(%rdi)
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: movw %ax, 12(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 14(%rdi)
+; AVX2-NEXT: vpextrd $3, %xmm0, %eax
+; AVX2-NEXT: movw %ax, 9(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 11(%rdi)
+; AVX2-NEXT: vpextrd $2, %xmm0, %eax
+; AVX2-NEXT: movw %ax, 6(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 8(%rdi)
+; AVX2-NEXT: vpextrd $1, %xmm0, %eax
+; AVX2-NEXT: movw %ax, 3(%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 5(%rdi)
+; AVX2-NEXT: vmovd %xmm0, %eax
+; AVX2-NEXT: movw %ax, (%rdi)
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: movb %al, 2(%rdi)
+; AVX2-NEXT: vzeroupper
+; AVX2-NEXT: retq
+;
+; AVX512-LABEL: trunc_ssat_v16i32_v16i24:
+; AVX512: # %bb.0:
+; AVX512-NEXT: pushq %rbp
+; AVX512-NEXT: pushq %r15
+; AVX512-NEXT: pushq %r14
+; AVX512-NEXT: pushq %rbx
+; AVX512-NEXT: vpminsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
+; AVX512-NEXT: vpmaxsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
+; AVX512-NEXT: vextracti32x4 $3, %zmm0, %xmm1
+; AVX512-NEXT: vpextrd $3, %xmm1, %ecx
+; AVX512-NEXT: movw %cx, 45(%rdi)
+; AVX512-NEXT: vpextrd $2, %xmm1, %eax
+; AVX512-NEXT: movw %ax, 42(%rdi)
+; AVX512-NEXT: vpextrd $1, %xmm1, %ebp
+; AVX512-NEXT: movw %bp, 39(%rdi)
+; AVX512-NEXT: vmovd %xmm1, %esi
+; AVX512-NEXT: movw %si, 36(%rdi)
+; AVX512-NEXT: vextracti32x4 $2, %zmm0, %xmm1
+; AVX512-NEXT: vpextrd $3, %xmm1, %ebx
+; AVX512-NEXT: movw %bx, 33(%rdi)
+; AVX512-NEXT: vpextrd $2, %xmm1, %edx
+; AVX512-NEXT: movw %dx, 30(%rdi)
+; AVX512-NEXT: vpextrd $1, %xmm1, %r15d
+; AVX512-NEXT: movw %r15w, 27(%rdi)
+; AVX512-NEXT: vmovd %xmm1, %r14d
+; AVX512-NEXT: movw %r14w, 24(%rdi)
+; AVX512-NEXT: vpextrd $3, %xmm0, %r11d
+; AVX512-NEXT: movw %r11w, 9(%rdi)
+; AVX512-NEXT: vpextrd $2, %xmm0, %r10d
+; AVX512-NEXT: movw %r10w, 6(%rdi)
+; AVX512-NEXT: vpextrd $1, %xmm0, %r9d
+; AVX512-NEXT: movw %r9w, 3(%rdi)
+; AVX512-NEXT: vmovd %xmm0, %r8d
+; AVX512-NEXT: movw %r8w, (%rdi)
+; AVX512-NEXT: shrl $16, %ecx
+; AVX512-NEXT: movb %cl, 47(%rdi)
+; AVX512-NEXT: shrl $16, %eax
+; AVX512-NEXT: movb %al, 44(%rdi)
+; AVX512-NEXT: shrl $16, %ebp
+; AVX512-NEXT: movb %bpl, 41(%rdi)
+; AVX512-NEXT: shrl $16, %esi
+; AVX512-NEXT: movb %sil, 38(%rdi)
+; AVX512-NEXT: shrl $16, %ebx
+; AVX512-NEXT: movb %bl, 35(%rdi)
+; AVX512-NEXT: shrl $16, %edx
+; AVX512-NEXT: movb %dl, 32(%rdi)
+; AVX512-NEXT: shrl $16, %r15d
+; AVX512-NEXT: movb %r15b, 29(%rdi)
+; AVX512-NEXT: shrl $16, %r14d
+; AVX512-NEXT: movb %r14b, 26(%rdi)
+; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm0
+; AVX512-NEXT: vpextrd $3, %xmm0, %edx
+; AVX512-NEXT: movw %dx, 21(%rdi)
+; AVX512-NEXT: vpextrd $2, %xmm0, %esi
+; AVX512-NEXT: movw %si, 18(%rdi)
+; AVX512-NEXT: vpextrd $1, %xmm0, %ecx
+; AVX512-NEXT: movw %cx, 15(%rdi)
+; AVX512-NEXT: vmovd %xmm0, %eax
+; AVX512-NEXT: movw %ax, 12(%rdi)
+; AVX512-NEXT: shrl $16, %r11d
+; AVX512-NEXT: movb %r11b, 11(%rdi)
+; AVX512-NEXT: shrl $16, %r10d
+; AVX512-NEXT: movb %r10b, 8(%rdi)
+; AVX512-NEXT: shrl $16, %r9d
+; AVX512-NEXT: movb %r9b, 5(%rdi)
+; AVX512-NEXT: shrl $16, %r8d
+; AVX512-NEXT: movb %r8b, 2(%rdi)
+; AVX512-NEXT: shrl $16, %edx
+; AVX512-NEXT: movb %dl, 23(%rdi)
+; AVX512-NEXT: shrl $16, %esi
+; AVX512-NEXT: movb %sil, 20(%rdi)
+; AVX512-NEXT: shrl $16, %ecx
+; AVX512-NEXT: movb %cl, 17(%rdi)
+; AVX512-NEXT: shrl $16, %eax
+; AVX512-NEXT: movb %al, 14(%rdi)
+; AVX512-NEXT: popq %rbx
+; AVX512-NEXT: popq %r14
+; AVX512-NEXT: popq %r15
+; AVX512-NEXT: popq %rbp
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
+;
+; SKX-LABEL: trunc_ssat_v16i32_v16i24:
+; SKX: # %bb.0:
+; SKX-NEXT: pushq %rbp
+; SKX-NEXT: pushq %r15
+; SKX-NEXT: pushq %r14
+; SKX-NEXT: pushq %rbx
+; SKX-NEXT: vpminsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
+; SKX-NEXT: vpmaxsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
+; SKX-NEXT: vextracti32x4 $3, %zmm0, %xmm1
+; SKX-NEXT: vpextrd $3, %xmm1, %ecx
+; SKX-NEXT: movw %cx, 45(%rdi)
+; SKX-NEXT: vpextrd $2, %xmm1, %eax
+; SKX-NEXT: movw %ax, 42(%rdi)
+; SKX-NEXT: vpextrd $1, %xmm1, %ebp
+; SKX-NEXT: movw %bp, 39(%rdi)
+; SKX-NEXT: vmovd %xmm1, %esi
+; SKX-NEXT: movw %si, 36(%rdi)
+; SKX-NEXT: vextracti32x4 $2, %zmm0, %xmm1
+; SKX-NEXT: vpextrd $3, %xmm1, %ebx
+; SKX-NEXT: movw %bx, 33(%rdi)
+; SKX-NEXT: vpextrd $2, %xmm1, %edx
+; SKX-NEXT: movw %dx, 30(%rdi)
+; SKX-NEXT: vpextrd $1, %xmm1, %r15d
+; SKX-NEXT: movw %r15w, 27(%rdi)
+; SKX-NEXT: vmovd %xmm1, %r14d
+; SKX-NEXT: vpextrd $3, %xmm0, %r11d
+; SKX-NEXT: movw %r14w, 24(%rdi)
+; SKX-NEXT: movw %r11w, 9(%rdi)
+; SKX-NEXT: vpextrd $2, %xmm0, %r10d
+; SKX-NEXT: vpextrd $1, %xmm0, %r9d
+; SKX-NEXT: movw %r10w, 6(%rdi)
+; SKX-NEXT: movw %r9w, 3(%rdi)
+; SKX-NEXT: vmovd %xmm0, %r8d
+; SKX-NEXT: movw %r8w, (%rdi)
+; SKX-NEXT: shrl $16, %ecx
+; SKX-NEXT: movb %cl, 47(%rdi)
+; SKX-NEXT: shrl $16, %eax
+; SKX-NEXT: movb %al, 44(%rdi)
+; SKX-NEXT: shrl $16, %ebp
+; SKX-NEXT: movb %bpl, 41(%rdi)
+; SKX-NEXT: shrl $16, %esi
+; SKX-NEXT: movb %sil, 38(%rdi)
+; SKX-NEXT: shrl $16, %ebx
+; SKX-NEXT: movb %bl, 35(%rdi)
+; SKX-NEXT: shrl $16, %edx
+; SKX-NEXT: movb %dl, 32(%rdi)
+; SKX-NEXT: shrl $16, %r15d
+; SKX-NEXT: movb %r15b, 29(%rdi)
+; SKX-NEXT: shrl $16, %r14d
+; SKX-NEXT: movb %r14b, 26(%rdi)
+; SKX-NEXT: vextracti128 $1, %ymm0, %xmm0
+; SKX-NEXT: vpextrd $3, %xmm0, %edx
+; SKX-NEXT: movw %dx, 21(%rdi)
+; SKX-NEXT: vpextrd $2, %xmm0, %esi
+; SKX-NEXT: movw %si, 18(%rdi)
+; SKX-NEXT: vpextrd $1, %xmm0, %ecx
+; SKX-NEXT: movw %cx, 15(%rdi)
+; SKX-NEXT: vmovd %xmm0, %eax
+; SKX-NEXT: movw %ax, 12(%rdi)
+; SKX-NEXT: shrl $16, %r11d
+; SKX-NEXT: movb %r11b, 11(%rdi)
+; SKX-NEXT: shrl $16, %r10d
+; SKX-NEXT: movb %r10b, 8(%rdi)
+; SKX-NEXT: shrl $16, %r9d
+; SKX-NEXT: movb %r9b, 5(%rdi)
+; SKX-NEXT: shrl $16, %r8d
+; SKX-NEXT: movb %r8b, 2(%rdi)
+; SKX-NEXT: shrl $16, %edx
+; SKX-NEXT: movb %dl, 23(%rdi)
+; SKX-NEXT: shrl $16, %esi
+; SKX-NEXT: movb %sil, 20(%rdi)
+; SKX-NEXT: shrl $16, %ecx
+; SKX-NEXT: movb %cl, 17(%rdi)
+; SKX-NEXT: shrl $16, %eax
+; SKX-NEXT: movb %al, 14(%rdi)
+; SKX-NEXT: popq %rbx
+; SKX-NEXT: popq %r14
+; SKX-NEXT: popq %r15
+; SKX-NEXT: popq %rbp
+; SKX-NEXT: vzeroupper
+; SKX-NEXT: retq
+ %a = call <16 x i32> @llvm.smin.v16i32(<16 x i32> %x, <16 x i32> <i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607, i32 8388607>)
+ %b = call <16 x i32> @llvm.smax.v16i32(<16 x i32> %a, <16 x i32> <i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608>)
+ %c = trunc <16 x i32> %b to <16 x i24>
+ store <16 x i24> %c, <16 x i24>* %y
+ ret void
+}
+declare <16 x i32> @llvm.smax.v16i32(<16 x i32>, <16 x i32>)
+declare <16 x i32> @llvm.smin.v16i32(<16 x i32>, <16 x i32>)