[NFC][Codegen] Revisit test coverage for X % C == 0 fold

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@364642 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/test/CodeGen/AArch64/urem-seteq-vec-nonsplat.ll b/test/CodeGen/AArch64/urem-seteq-vec-nonsplat.ll
index 5fef04eb7440a4b5df71e96d08d22b6ea0bbee75..759ccf307f8d0bef9f1bfa3c48cd57322567f3d4 100644 (file)
--- a/test/CodeGen/AArch64/urem-seteq-vec-nonsplat.ll
+++ b/test/CodeGen/AArch64/urem-seteq-vec-nonsplat.ll
@@ -3,40 +3,39 @@
 
 ; At the moment, BuildUREMEqFold does not handle nonsplat vectors.
 
-define <4 x i32> @test_urem_odd_div(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_odd_div:
+; Odd+Even divisors
+define <4 x i32> @test_urem_odd_even(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_even:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    adrp x8, .LCPI0_0
 ; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI0_0]
 ; CHECK-NEXT:    adrp x8, .LCPI0_1
 ; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI0_1]
 ; CHECK-NEXT:    adrp x8, .LCPI0_2
-; CHECK-NEXT:    umull2 v3.2d, v0.4s, v1.4s
-; CHECK-NEXT:    umull v1.2d, v0.2s, v1.2s
-; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v3.4s
-; CHECK-NEXT:    sub v3.4s, v0.4s, v1.4s
-; CHECK-NEXT:    umull2 v4.2d, v3.4s, v2.4s
-; CHECK-NEXT:    umull v2.2d, v3.2s, v2.2s
 ; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI0_2]
+; CHECK-NEXT:    neg v1.4s, v1.4s
 ; CHECK-NEXT:    adrp x8, .LCPI0_3
-; CHECK-NEXT:    uzp2 v2.4s, v2.4s, v4.4s
-; CHECK-NEXT:    ldr q4, [x8, :lo12:.LCPI0_3]
+; CHECK-NEXT:    ushl v1.4s, v0.4s, v1.4s
+; CHECK-NEXT:    umull2 v4.2d, v1.4s, v2.4s
+; CHECK-NEXT:    umull v1.2d, v1.2s, v2.2s
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI0_3]
+; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
 ; CHECK-NEXT:    neg v3.4s, v3.4s
-; CHECK-NEXT:    add v1.4s, v2.4s, v1.4s
 ; CHECK-NEXT:    ushl v1.4s, v1.4s, v3.4s
-; CHECK-NEXT:    mls v0.4s, v1.4s, v4.4s
+; CHECK-NEXT:    mls v0.4s, v1.4s, v2.4s
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 3, i32 5, i32 7, i32 9>
+  %urem = urem <4 x i32> %X, <i32 5, i32 14, i32 25, i32 100>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_even_div(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_even_div:
+; One all-ones divisor in odd divisor
+define <4 x i32> @test_urem_odd_allones(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_allones:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    adrp x8, .LCPI1_0
 ; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI1_0]
@@ -44,29 +43,25 @@ define <4 x i32> @test_urem_even_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI1_1]
 ; CHECK-NEXT:    adrp x8, .LCPI1_2
 ; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI1_2]
-; CHECK-NEXT:    neg v1.4s, v1.4s
-; CHECK-NEXT:    adrp x8, .LCPI1_3
-; CHECK-NEXT:    ushl v1.4s, v0.4s, v1.4s
-; CHECK-NEXT:    umull2 v4.2d, v1.4s, v2.4s
-; CHECK-NEXT:    umull v1.2d, v1.2s, v2.2s
-; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI1_3]
+; CHECK-NEXT:    umull2 v4.2d, v0.4s, v1.4s
+; CHECK-NEXT:    umull v1.2d, v0.2s, v1.2s
 ; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
-; CHECK-NEXT:    neg v3.4s, v3.4s
-; CHECK-NEXT:    ushl v1.4s, v1.4s, v3.4s
-; CHECK-NEXT:    mls v0.4s, v1.4s, v2.4s
+; CHECK-NEXT:    neg v2.4s, v2.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v2.4s
+; CHECK-NEXT:    mls v0.4s, v1.4s, v3.4s
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 6, i32 10, i32 12, i32 14>
+  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 4294967295, i32 5>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-; Can't fold due to last line
-define <4 x i32> @test_urem_pow2(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_pow2:
+; One all-ones divisor in even divisor
+define <4 x i32> @test_urem_even_allones(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_even_allones:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    adrp x8, .LCPI2_0
 ; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI2_0]
@@ -74,25 +69,29 @@ define <4 x i32> @test_urem_pow2(<4 x i32> %X) nounwind readnone {
 ; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI2_1]
 ; CHECK-NEXT:    adrp x8, .LCPI2_2
 ; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI2_2]
-; CHECK-NEXT:    umull2 v4.2d, v0.4s, v1.4s
-; CHECK-NEXT:    umull v1.2d, v0.2s, v1.2s
+; CHECK-NEXT:    neg v1.4s, v1.4s
+; CHECK-NEXT:    adrp x8, .LCPI2_3
+; CHECK-NEXT:    ushl v1.4s, v0.4s, v1.4s
+; CHECK-NEXT:    umull2 v4.2d, v1.4s, v2.4s
+; CHECK-NEXT:    umull v1.2d, v1.2s, v2.2s
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI2_3]
 ; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
-; CHECK-NEXT:    neg v2.4s, v2.4s
-; CHECK-NEXT:    ushl v1.4s, v1.4s, v2.4s
-; CHECK-NEXT:    mls v0.4s, v1.4s, v3.4s
+; CHECK-NEXT:    neg v3.4s, v3.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v3.4s
+; CHECK-NEXT:    mls v0.4s, v1.4s, v2.4s
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 6, i32 10, i32 12, i32 16>
+  %urem = urem <4 x i32> %X, <i32 14, i32 14, i32 4294967295, i32 14>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-; Can't fold due to second line
-define <4 x i32> @test_urem_one(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_one:
+; One all-ones divisor in odd+even divisor
+define <4 x i32> @test_urem_odd_even_allones(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_even_allones:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    adrp x8, .LCPI3_0
 ; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI3_0]
@@ -106,25 +105,23 @@ define <4 x i32> @test_urem_one(<4 x i32> %X) nounwind readnone {
 ; CHECK-NEXT:    umull2 v4.2d, v1.4s, v2.4s
 ; CHECK-NEXT:    umull v1.2d, v1.2s, v2.2s
 ; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI3_3]
-; CHECK-NEXT:    adrp x8, .LCPI3_4
 ; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
-; CHECK-NEXT:    ldr q4, [x8, :lo12:.LCPI3_4]
 ; CHECK-NEXT:    neg v3.4s, v3.4s
 ; CHECK-NEXT:    ushl v1.4s, v1.4s, v3.4s
-; CHECK-NEXT:    bsl v2.16b, v0.16b, v1.16b
-; CHECK-NEXT:    mls v0.4s, v2.4s, v4.4s
+; CHECK-NEXT:    mls v0.4s, v1.4s, v2.4s
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 6, i32 1, i32 12, i32 14>
+  %urem = urem <4 x i32> %X, <i32 5, i32 14, i32 4294967295, i32 100>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_nomulinv(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_nomulinv:
+; One power-of-two divisor in odd divisor
+define <4 x i32> @test_urem_odd_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_poweroftwo:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    adrp x8, .LCPI4_0
 ; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI4_0]
@@ -132,52 +129,55 @@ define <4 x i32> @test_urem_nomulinv(<4 x i32> %X) nounwind readnone {
 ; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI4_1]
 ; CHECK-NEXT:    adrp x8, .LCPI4_2
 ; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI4_2]
-; CHECK-NEXT:    neg v1.4s, v1.4s
-; CHECK-NEXT:    adrp x8, .LCPI4_3
-; CHECK-NEXT:    ushl v1.4s, v0.4s, v1.4s
-; CHECK-NEXT:    umull2 v4.2d, v1.4s, v2.4s
-; CHECK-NEXT:    umull v1.2d, v1.2s, v2.2s
-; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI4_3]
+; CHECK-NEXT:    umull2 v4.2d, v0.4s, v1.4s
+; CHECK-NEXT:    umull v1.2d, v0.2s, v1.2s
 ; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
-; CHECK-NEXT:    neg v3.4s, v3.4s
-; CHECK-NEXT:    ushl v1.4s, v1.4s, v3.4s
-; CHECK-NEXT:    mls v0.4s, v1.4s, v2.4s
+; CHECK-NEXT:    neg v2.4s, v2.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v2.4s
+; CHECK-NEXT:    mls v0.4s, v1.4s, v3.4s
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 6, i32 100, i32 12, i32 14>
+  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 16, i32 5>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_comp(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_comp:
+; One power-of-two divisor in even divisor
+define <4 x i32> @test_urem_even_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_even_poweroftwo:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #52429
-; CHECK-NEXT:    movk w8, #52428, lsl #16
-; CHECK-NEXT:    adrp x9, .LCPI5_0
-; CHECK-NEXT:    dup v2.4s, w8
-; CHECK-NEXT:    ldr q3, [x9, :lo12:.LCPI5_0]
-; CHECK-NEXT:    umull2 v4.2d, v0.4s, v2.4s
-; CHECK-NEXT:    umull v2.2d, v0.2s, v2.2s
-; CHECK-NEXT:    uzp2 v2.4s, v2.4s, v4.4s
-; CHECK-NEXT:    movi v1.4s, #5
-; CHECK-NEXT:    ushr v2.4s, v2.4s, #2
-; CHECK-NEXT:    mls v0.4s, v2.4s, v1.4s
-; CHECK-NEXT:    cmeq v0.4s, v0.4s, v3.4s
+; CHECK-NEXT:    adrp x8, .LCPI5_0
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI5_0]
+; CHECK-NEXT:    adrp x8, .LCPI5_1
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI5_1]
+; CHECK-NEXT:    adrp x8, .LCPI5_2
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI5_2]
+; CHECK-NEXT:    neg v1.4s, v1.4s
+; CHECK-NEXT:    adrp x8, .LCPI5_3
+; CHECK-NEXT:    ushl v1.4s, v0.4s, v1.4s
+; CHECK-NEXT:    umull2 v4.2d, v1.4s, v2.4s
+; CHECK-NEXT:    umull v1.2d, v1.2s, v2.2s
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI5_3]
+; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
+; CHECK-NEXT:    neg v3.4s, v3.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v3.4s
+; CHECK-NEXT:    mls v0.4s, v1.4s, v2.4s
+; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 5, i32 5>
-  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 1, i32 0>
+  %urem = urem <4 x i32> %X, <i32 14, i32 14, i32 16, i32 14>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_both(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_both:
+; One power-of-two divisor in odd+even divisor
+define <4 x i32> @test_urem_odd_even_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_even_poweroftwo:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    adrp x8, .LCPI6_0
 ; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI6_0]
@@ -185,82 +185,194 @@ define <4 x i32> @test_urem_both(<4 x i32> %X) nounwind readnone {
 ; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI6_1]
 ; CHECK-NEXT:    adrp x8, .LCPI6_2
 ; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI6_2]
-; CHECK-NEXT:    umull2 v4.2d, v0.4s, v1.4s
-; CHECK-NEXT:    umull v1.2d, v0.2s, v1.2s
+; CHECK-NEXT:    neg v1.4s, v1.4s
+; CHECK-NEXT:    adrp x8, .LCPI6_3
+; CHECK-NEXT:    ushl v1.4s, v0.4s, v1.4s
+; CHECK-NEXT:    umull2 v4.2d, v1.4s, v2.4s
+; CHECK-NEXT:    umull v1.2d, v1.2s, v2.2s
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI6_3]
 ; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
-; CHECK-NEXT:    ushr v1.4s, v1.4s, #2
+; CHECK-NEXT:    neg v3.4s, v3.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v3.4s
 ; CHECK-NEXT:    mls v0.4s, v1.4s, v2.4s
-; CHECK-NEXT:    cmeq v0.4s, v0.4s, v3.4s
+; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 6, i32 5, i32 6, i32 5>
-  %cmp = icmp eq <4 x i32> %urem, <i32 1, i32 0, i32 1, i32 0>
+  %urem = urem <4 x i32> %X, <i32 5, i32 14, i32 16, i32 100>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_div_undef(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_div_undef:
+; One all-ones divisor and one power-of-two divisor in odd divisor
+define <4 x i32> @test_urem_odd_allones_and_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_allones_and_poweroftwo:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    movi v0.2d, #0000000000000000
+; CHECK-NEXT:    adrp x8, .LCPI7_0
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI7_0]
+; CHECK-NEXT:    adrp x8, .LCPI7_1
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI7_1]
+; CHECK-NEXT:    adrp x8, .LCPI7_2
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI7_2]
+; CHECK-NEXT:    umull2 v4.2d, v0.4s, v1.4s
+; CHECK-NEXT:    umull v1.2d, v0.2s, v1.2s
+; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
+; CHECK-NEXT:    neg v2.4s, v2.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v2.4s
+; CHECK-NEXT:    mls v0.4s, v1.4s, v3.4s
+; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
+; CHECK-NEXT:    movi v1.4s, #1
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 undef, i32 5>
+  %urem = urem <4 x i32> %X, <i32 5, i32 16, i32 4294967295, i32 5>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_comp_undef(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_comp_undef:
+; One all-ones divisor and one power-of-two divisor in even divisor
+define <4 x i32> @test_urem_even_allones_and_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_even_allones_and_poweroftwo:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #52429
-; CHECK-NEXT:    movk w8, #52428, lsl #16
-; CHECK-NEXT:    dup v2.4s, w8
-; CHECK-NEXT:    umull2 v3.2d, v0.4s, v2.4s
-; CHECK-NEXT:    umull v2.2d, v0.2s, v2.2s
-; CHECK-NEXT:    uzp2 v2.4s, v2.4s, v3.4s
-; CHECK-NEXT:    movi v1.4s, #5
-; CHECK-NEXT:    ushr v2.4s, v2.4s, #2
-; CHECK-NEXT:    mls v0.4s, v2.4s, v1.4s
+; CHECK-NEXT:    adrp x8, .LCPI8_0
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI8_0]
+; CHECK-NEXT:    adrp x8, .LCPI8_1
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI8_1]
+; CHECK-NEXT:    adrp x8, .LCPI8_2
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI8_2]
+; CHECK-NEXT:    neg v1.4s, v1.4s
+; CHECK-NEXT:    adrp x8, .LCPI8_3
+; CHECK-NEXT:    ushl v1.4s, v0.4s, v1.4s
+; CHECK-NEXT:    umull2 v4.2d, v1.4s, v2.4s
+; CHECK-NEXT:    umull v1.2d, v1.2s, v2.2s
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI8_3]
+; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
+; CHECK-NEXT:    neg v3.4s, v3.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v3.4s
+; CHECK-NEXT:    mls v0.4s, v1.4s, v2.4s
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 5, i32 5>
-  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 undef, i32 0, i32 0>
+  %urem = urem <4 x i32> %X, <i32 14, i32 16, i32 4294967295, i32 14>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_both_undef(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_both_undef:
+; One all-ones divisor and one power-of-two divisor in odd+even divisor
+define <4 x i32> @test_urem_odd_even_allones_and_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_even_allones_and_poweroftwo:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    movi v0.2d, #0000000000000000
+; CHECK-NEXT:    adrp x8, .LCPI9_0
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI9_0]
+; CHECK-NEXT:    adrp x8, .LCPI9_1
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI9_1]
+; CHECK-NEXT:    adrp x8, .LCPI9_2
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI9_2]
+; CHECK-NEXT:    umull2 v4.2d, v0.4s, v1.4s
+; CHECK-NEXT:    umull v1.2d, v0.2s, v1.2s
+; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
+; CHECK-NEXT:    neg v2.4s, v2.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v2.4s
+; CHECK-NEXT:    mls v0.4s, v1.4s, v3.4s
+; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
+; CHECK-NEXT:    movi v1.4s, #1
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 undef, i32 5>
-  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 undef, i32 0, i32 0>
+  %urem = urem <4 x i32> %X, <i32 5, i32 16, i32 4294967295, i32 100>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_div_even_odd(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_div_even_odd:
+;------------------------------------------------------------------------------;
+; Negative tests - the fold is invalid if any divisor is 1.
+;------------------------------------------------------------------------------;
+
+; One divisor in odd divisor
+define <4 x i32> @test_urem_odd_one(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_one:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    adrp x8, .LCPI10_0
 ; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI10_0]
 ; CHECK-NEXT:    adrp x8, .LCPI10_1
 ; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI10_1]
-; CHECK-NEXT:    umull2 v3.2d, v0.4s, v1.4s
+; CHECK-NEXT:    adrp x8, .LCPI10_2
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI10_2]
+; CHECK-NEXT:    adrp x8, .LCPI10_3
+; CHECK-NEXT:    umull2 v4.2d, v0.4s, v1.4s
 ; CHECK-NEXT:    umull v1.2d, v0.2s, v1.2s
-; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v3.4s
-; CHECK-NEXT:    ushr v1.4s, v1.4s, #2
-; CHECK-NEXT:    mls v0.4s, v1.4s, v2.4s
+; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
+; CHECK-NEXT:    ldr q4, [x8, :lo12:.LCPI10_3]
+; CHECK-NEXT:    neg v2.4s, v2.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v2.4s
+; CHECK-NEXT:    bsl v3.16b, v0.16b, v1.16b
+; CHECK-NEXT:    mls v0.4s, v3.4s, v4.4s
+; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
+; CHECK-NEXT:    movi v1.4s, #1
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    ret
+  %urem = urem <4 x i32> %X, <i32 25, i32 25, i32 1, i32 25>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
+}
+
+; One divisor in even divisors
+define <4 x i32> @test_urem_even_one(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_even_one:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    adrp x8, .LCPI11_0
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI11_0]
+; CHECK-NEXT:    adrp x8, .LCPI11_1
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI11_1]
+; CHECK-NEXT:    adrp x8, .LCPI11_2
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI11_2]
+; CHECK-NEXT:    adrp x8, .LCPI11_3
+; CHECK-NEXT:    umull2 v4.2d, v0.4s, v1.4s
+; CHECK-NEXT:    umull v1.2d, v0.2s, v1.2s
+; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
+; CHECK-NEXT:    ldr q4, [x8, :lo12:.LCPI11_3]
+; CHECK-NEXT:    neg v2.4s, v2.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v2.4s
+; CHECK-NEXT:    bsl v3.16b, v0.16b, v1.16b
+; CHECK-NEXT:    mls v0.4s, v3.4s, v4.4s
+; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
+; CHECK-NEXT:    movi v1.4s, #1
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    ret
+  %urem = urem <4 x i32> %X, <i32 100, i32 100, i32 1, i32 100>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
+}
+
+; One divisor in odd-even divisors
+define <4 x i32> @test_urem_odd_even_one(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_even_one:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    adrp x8, .LCPI12_0
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI12_0]
+; CHECK-NEXT:    adrp x8, .LCPI12_1
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI12_1]
+; CHECK-NEXT:    adrp x8, .LCPI12_2
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI12_2]
+; CHECK-NEXT:    adrp x8, .LCPI12_3
+; CHECK-NEXT:    umull2 v4.2d, v0.4s, v1.4s
+; CHECK-NEXT:    umull v1.2d, v0.2s, v1.2s
+; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
+; CHECK-NEXT:    ldr q4, [x8, :lo12:.LCPI12_3]
+; CHECK-NEXT:    neg v2.4s, v2.4s
+; CHECK-NEXT:    ushl v1.4s, v1.4s, v2.4s
+; CHECK-NEXT:    bsl v3.16b, v0.16b, v1.16b
+; CHECK-NEXT:    mls v0.4s, v3.4s, v4.4s
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 6, i32 6>
+  %urem = urem <4 x i32> %X, <i32 25, i32 100, i32 1, i32 25>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret

diff --git a/test/CodeGen/AArch64/urem-seteq-vec-splat.ll b/test/CodeGen/AArch64/urem-seteq-vec-splat.ll
index 7621050631e4be5596a93763cd6e3013c6d447ee..8b08f533239ebddc8c6a8edb051e2dfb75719bf2 100644 (file)
--- a/test/CodeGen/AArch64/urem-seteq-vec-splat.ll
+++ b/test/CodeGen/AArch64/urem-seteq-vec-splat.ll
@@ -1,102 +1,105 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s
 
-; Tests BuildUREMEqFold for 4 x i32 splat vectors with odd divisor.
-; See urem-seteq.ll for justification behind constants emitted.
-define <4 x i32> @test_urem_odd_vec_i32(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_odd_vec_i32:
+; Odd divisor
+define <4 x i32> @test_urem_odd_25(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_25:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #52429
-; CHECK-NEXT:    movk w8, #52428, lsl #16
-; CHECK-NEXT:    dup v2.4s, w8
-; CHECK-NEXT:    movi v1.16b, #51
-; CHECK-NEXT:    mul v0.4s, v0.4s, v2.4s
-; CHECK-NEXT:    cmhs v0.4s, v1.4s, v0.4s
+; CHECK-NEXT:    mov w8, #23593
+; CHECK-NEXT:    movk w8, #49807, lsl #16
+; CHECK-NEXT:    mov w9, #28835
+; CHECK-NEXT:    movk w9, #2621, lsl #16
+; CHECK-NEXT:    dup v1.4s, w8
+; CHECK-NEXT:    dup v2.4s, w9
+; CHECK-NEXT:    mul v0.4s, v0.4s, v1.4s
+; CHECK-NEXT:    cmhs v0.4s, v2.4s, v0.4s
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 5, i32 5>
+  %urem = urem <4 x i32> %X, <i32 25, i32 25, i32 25, i32 25>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-; Like test_urem_odd_vec_i32, but with 8 x i16 vectors.
-define <8 x i16> @test_urem_odd_vec_i16(<8 x i16> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_odd_vec_i16:
+; Even divisors
+define <4 x i32> @test_urem_even_100(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_even_100:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #52429
-; CHECK-NEXT:    dup v2.8h, w8
-; CHECK-NEXT:    movi v1.16b, #51
-; CHECK-NEXT:    mul v0.8h, v0.8h, v2.8h
-; CHECK-NEXT:    cmhs v0.8h, v1.8h, v0.8h
-; CHECK-NEXT:    movi v1.8h, #1
+; CHECK-NEXT:    mov w8, #34079
+; CHECK-NEXT:    movk w8, #20971, lsl #16
+; CHECK-NEXT:    dup v2.4s, w8
+; CHECK-NEXT:    umull2 v3.2d, v0.4s, v2.4s
+; CHECK-NEXT:    umull v2.2d, v0.2s, v2.2s
+; CHECK-NEXT:    uzp2 v2.4s, v2.4s, v3.4s
+; CHECK-NEXT:    movi v1.4s, #100
+; CHECK-NEXT:    ushr v2.4s, v2.4s, #5
+; CHECK-NEXT:    mls v0.4s, v2.4s, v1.4s
+; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
+; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <8 x i16> %X, <i16 5, i16 5, i16 5, i16 5,
-                              i16 5, i16 5, i16 5, i16 5>
-  %cmp = icmp eq <8 x i16> %urem, <i16 0, i16 0, i16 0, i16 0,
-                                   i16 0, i16 0, i16 0, i16 0>
-  %ret = zext <8 x i1> %cmp to <8 x i16>
-  ret <8 x i16> %ret
+  %urem = urem <4 x i32> %X, <i32 100, i32 100, i32 100, i32 100>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
 }
 
-; Tests BuildUREMEqFold for 4 x i32 splat vectors with even divisor.
-; The expected behavior is that the fold is _not_ applied
-; because it requires a ROTR in the even case, which has to be expanded.
-define <4 x i32> @test_urem_even_vec_i32(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_even_vec_i32:
+;------------------------------------------------------------------------------;
+; Comparison constant has undef elements.
+;------------------------------------------------------------------------------;
+
+define <4 x i32> @test_urem_odd_undef1(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_odd_undef1:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #9363
-; CHECK-NEXT:    movk w8, #37449, lsl #16
-; CHECK-NEXT:    ushr v1.4s, v0.4s, #1
-; CHECK-NEXT:    dup v3.4s, w8
-; CHECK-NEXT:    umull2 v4.2d, v1.4s, v3.4s
-; CHECK-NEXT:    umull v1.2d, v1.2s, v3.2s
-; CHECK-NEXT:    uzp2 v1.4s, v1.4s, v4.4s
-; CHECK-NEXT:    movi v2.4s, #14
-; CHECK-NEXT:    ushr v1.4s, v1.4s, #2
-; CHECK-NEXT:    mls v0.4s, v1.4s, v2.4s
+; CHECK-NEXT:    mov w8, #34079
+; CHECK-NEXT:    movk w8, #20971, lsl #16
+; CHECK-NEXT:    dup v2.4s, w8
+; CHECK-NEXT:    umull2 v3.2d, v0.4s, v2.4s
+; CHECK-NEXT:    umull v2.2d, v0.2s, v2.2s
+; CHECK-NEXT:    uzp2 v2.4s, v2.4s, v3.4s
+; CHECK-NEXT:    movi v1.4s, #25
+; CHECK-NEXT:    ushr v2.4s, v2.4s, #3
+; CHECK-NEXT:    mls v0.4s, v2.4s, v1.4s
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
 ; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <4 x i32> %X, <i32 14, i32 14, i32 14, i32 14>
-  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %urem = urem <4 x i32> %X, <i32 25, i32 25, i32 25, i32 25>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 undef, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-; Like test_urem_even_vec_i32, but with 8 x i16 vectors.
-; i16 is not legal for ROTR on AArch64, but ROTR also cannot be promoted to i32,
-; so this would crash if BuildUREMEqFold was applied.
-define <8 x i16> @test_urem_even_vec_i16(<8 x i16> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_even_vec_i16:
+define <4 x i32> @test_urem_even_undef1(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_even_undef1:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #18725
-; CHECK-NEXT:    ushr v1.8h, v0.8h, #1
-; CHECK-NEXT:    dup v3.8h, w8
-; CHECK-NEXT:    umull2 v4.4s, v1.8h, v3.8h
-; CHECK-NEXT:    umull v1.4s, v1.4h, v3.4h
-; CHECK-NEXT:    uzp2 v1.8h, v1.8h, v4.8h
-; CHECK-NEXT:    movi v2.8h, #14
-; CHECK-NEXT:    ushr v1.8h, v1.8h, #1
-; CHECK-NEXT:    mls v0.8h, v1.8h, v2.8h
-; CHECK-NEXT:    cmeq v0.8h, v0.8h, #0
-; CHECK-NEXT:    movi v1.8h, #1
+; CHECK-NEXT:    mov w8, #34079
+; CHECK-NEXT:    movk w8, #20971, lsl #16
+; CHECK-NEXT:    dup v2.4s, w8
+; CHECK-NEXT:    umull2 v3.2d, v0.4s, v2.4s
+; CHECK-NEXT:    umull v2.2d, v0.2s, v2.2s
+; CHECK-NEXT:    uzp2 v2.4s, v2.4s, v3.4s
+; CHECK-NEXT:    movi v1.4s, #100
+; CHECK-NEXT:    ushr v2.4s, v2.4s, #5
+; CHECK-NEXT:    mls v0.4s, v2.4s, v1.4s
+; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
+; CHECK-NEXT:    movi v1.4s, #1
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
-  %urem = urem <8 x i16> %X, <i16 14, i16 14, i16 14, i16 14,
-                              i16 14, i16 14, i16 14, i16 14>
-  %cmp = icmp eq <8 x i16> %urem, <i16 0, i16 0, i16 0, i16 0,
-                                   i16 0, i16 0, i16 0, i16 0>
-  %ret = zext <8 x i1> %cmp to <8 x i16>
-  ret <8 x i16> %ret
+  %urem = urem <4 x i32> %X, <i32 100, i32 100, i32 100, i32 100>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 undef, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
 }
 
-; We should not proceed with this fold if the divisor is 1 or -1
-define <4 x i32> @test_urem_one_vec(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_one_vec:
+;------------------------------------------------------------------------------;
+; Negative tests
+;------------------------------------------------------------------------------;
+
+; The fold is invalid if divisor is 1.
+define <4 x i32> @test_urem_one(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_one:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    movi v0.4s, #1
 ; CHECK-NEXT:    ret
@@ -106,10 +109,24 @@ define <4 x i32> @test_urem_one_vec(<4 x i32> %X) nounwind readnone {
   ret <4 x i32> %ret
 }
 
-; BuildUREMEqFold does not work when the only odd factor of the divisor is 1.
-; This ensures we don't touch powers of two.
-define <4 x i32> @test_urem_pow2_vec(<4 x i32> %X) nounwind readnone {
-; CHECK-LABEL: test_urem_pow2_vec:
+; We can lower remainder of division by all-ones much better elsewhere.
+define <4 x i32> @test_urem_allones(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_allones:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    neg v0.4s, v0.4s
+; CHECK-NEXT:    movi v1.4s, #1
+; CHECK-NEXT:    cmhs v0.4s, v1.4s, v0.4s
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    ret
+  %urem = urem <4 x i32> %X, <i32 4294967295, i32 4294967295, i32 4294967295, i32 4294967295>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
+}
+
+; We can lower remainder of division by powers of two much better elsewhere.
+define <4 x i32> @test_urem_pow2(<4 x i32> %X) nounwind {
+; CHECK-LABEL: test_urem_pow2:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    movi v1.4s, #15
 ; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b

diff --git a/test/CodeGen/AArch64/urem-seteq.ll b/test/CodeGen/AArch64/urem-seteq.ll
index 406866c9546266b6722dd4d8189acdbc6bdb8848..164c0a5f1ec9599365206c0b9934ed15f1eb1898 100644 (file)
--- a/test/CodeGen/AArch64/urem-seteq.ll
+++ b/test/CodeGen/AArch64/urem-seteq.ll
@@ -1,11 +1,15 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s
 
+;------------------------------------------------------------------------------;
+; Odd divisors
+;------------------------------------------------------------------------------;
+
 ; This tests the BuildREMEqFold optimization with UREM, i32, odd divisor, SETEQ.
 ; The corresponding pseudocode is:
 ; Q <- [N * multInv(5, 2^32)] <=> [N * 0xCCCCCCCD] <=> [N * (-858993459)]
 ; res <- [Q <= (2^32 - 1) / 5] <=> [Q <= 858993459] <=> [Q < 858993460]
-define i32 @test_urem_odd(i32 %X) nounwind readnone {
+define i32 @test_urem_odd(i32 %X) nounwind {
 ; CHECK-LABEL: test_urem_odd:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    mov w8, #52429
@@ -22,8 +26,25 @@ define i32 @test_urem_odd(i32 %X) nounwind readnone {
   ret i32 %ret
 }
 
+define i32 @test_urem_odd_25(i32 %X) nounwind {
+; CHECK-LABEL: test_urem_odd_25:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w8, #23593
+; CHECK-NEXT:    movk w8, #49807, lsl #16
+; CHECK-NEXT:    mov w9, #28836
+; CHECK-NEXT:    mul w8, w0, w8
+; CHECK-NEXT:    movk w9, #2621, lsl #16
+; CHECK-NEXT:    cmp w8, w9
+; CHECK-NEXT:    cset w0, lo
+; CHECK-NEXT:    ret
+  %urem = urem i32 %X, 25
+  %cmp = icmp eq i32 %urem, 0
+  %ret = zext i1 %cmp to i32
+  ret i32 %ret
+}
+
 ; This is like test_urem_odd, except the divisor has bit 30 set.
-define i32 @test_urem_odd_bit30(i32 %X) nounwind readnone {
+define i32 @test_urem_odd_bit30(i32 %X) nounwind {
 ; CHECK-LABEL: test_urem_odd_bit30:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    mov w8, #43691
@@ -39,7 +60,7 @@ define i32 @test_urem_odd_bit30(i32 %X) nounwind readnone {
 }
 
 ; This is like test_urem_odd, except the divisor has bit 31 set.
-define i32 @test_urem_odd_bit31(i32 %X) nounwind readnone {
+define i32 @test_urem_odd_bit31(i32 %X) nounwind {
 ; CHECK-LABEL: test_urem_odd_bit31:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    mov w8, #43691
@@ -54,13 +75,17 @@ define i32 @test_urem_odd_bit31(i32 %X) nounwind readnone {
   ret i32 %ret
 }
 
+;------------------------------------------------------------------------------;
+; Even divisors
+;------------------------------------------------------------------------------;
+
 ; This tests the BuildREMEqFold optimization with UREM, i16, even divisor, SETNE.
 ; In this case, D <=> 14 <=> 7 * 2^1, so D0 = 7 and K = 1.
 ; The corresponding pseudocode is:
 ; Q <- [N * multInv(D0, 2^16)] <=> [N * multInv(7, 2^16)] <=> [N * 28087]
 ; Q <- [Q >>rot K] <=> [Q >>rot 1]
 ; res <- ![Q <= (2^16 - 1) / 7] <=> ![Q <= 9362] <=> [Q > 9362]
-define i16 @test_urem_even(i16 %X) nounwind readnone {
+define i16 @test_urem_even(i16 %X) nounwind {
 ; CHECK-LABEL: test_urem_even:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    mov w9, #28087
@@ -79,8 +104,26 @@ define i16 @test_urem_even(i16 %X) nounwind readnone {
   ret i16 %ret
 }
 
+define i32 @test_urem_even_100(i32 %X) nounwind {
+; CHECK-LABEL: test_urem_even_100:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w8, #23593
+; CHECK-NEXT:    movk w8, #49807, lsl #16
+; CHECK-NEXT:    mul w8, w0, w8
+; CHECK-NEXT:    mov w9, #23593
+; CHECK-NEXT:    ror w8, w8, #2
+; CHECK-NEXT:    movk w9, #655, lsl #16
+; CHECK-NEXT:    cmp w8, w9
+; CHECK-NEXT:    cset w0, lo
+; CHECK-NEXT:    ret
+  %urem = urem i32 %X, 100
+  %cmp = icmp eq i32 %urem, 0
+  %ret = zext i1 %cmp to i32
+  ret i32 %ret
+}
+
 ; This is like test_urem_even, except the divisor has bit 30 set.
-define i32 @test_urem_even_bit30(i32 %X) nounwind readnone {
+define i32 @test_urem_even_bit30(i32 %X) nounwind {
 ; CHECK-LABEL: test_urem_even_bit30:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    mov w8, #20165
@@ -97,7 +140,7 @@ define i32 @test_urem_even_bit30(i32 %X) nounwind readnone {
 }
 
 ; This is like test_urem_odd, except the divisor has bit 31 set.
-define i32 @test_urem_even_bit31(i32 %X) nounwind readnone {
+define i32 @test_urem_even_bit31(i32 %X) nounwind {
 ; CHECK-LABEL: test_urem_even_bit31:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    mov w8, #64251
@@ -113,8 +156,33 @@ define i32 @test_urem_even_bit31(i32 %X) nounwind readnone {
   ret i32 %ret
 }
 
-; We should not proceed with this fold if the divisor is 1 or -1
-define i32 @test_urem_one(i32 %X) nounwind readnone {
+;------------------------------------------------------------------------------;
+; Special case
+;------------------------------------------------------------------------------;
+
+; 'NE' predicate is fine too.
+define i32 @test_urem_odd_setne(i32 %X) nounwind {
+; CHECK-LABEL: test_urem_odd_setne:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w8, #52429
+; CHECK-NEXT:    movk w8, #52428, lsl #16
+; CHECK-NEXT:    mul w8, w0, w8
+; CHECK-NEXT:    mov w9, #858993459
+; CHECK-NEXT:    cmp w8, w9
+; CHECK-NEXT:    cset w0, hi
+; CHECK-NEXT:    ret
+  %urem = urem i32 %X, 5
+  %cmp = icmp ne i32 %urem, 0
+  %ret = zext i1 %cmp to i32
+  ret i32 %ret
+}
+
+;------------------------------------------------------------------------------;
+; Negative tests
+;------------------------------------------------------------------------------;
+
+; The fold is invalid if divisor is 1.
+define i32 @test_urem_one(i32 %X) nounwind {
 ; CHECK-LABEL: test_urem_one:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    mov w0, #1
@@ -125,28 +193,22 @@ define i32 @test_urem_one(i32 %X) nounwind readnone {
   ret i32 %ret
 }
 
-define i32 @test_urem_100(i32 %X) nounwind readnone {
-; CHECK-LABEL: test_urem_100:
+; We can lower remainder of division by all-ones much better elsewhere.
+define i32 @test_urem_allones(i32 %X) nounwind {
+; CHECK-LABEL: test_urem_allones:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #23593
-; CHECK-NEXT:    movk w8, #49807, lsl #16
-; CHECK-NEXT:    mul w8, w0, w8
-; CHECK-NEXT:    mov w9, #23593
-; CHECK-NEXT:    ror w8, w8, #2
-; CHECK-NEXT:    movk w9, #655, lsl #16
-; CHECK-NEXT:    cmp w8, w9
+; CHECK-NEXT:    neg w8, w0
+; CHECK-NEXT:    cmp w8, #2 // =2
 ; CHECK-NEXT:    cset w0, lo
 ; CHECK-NEXT:    ret
-  %urem = urem i32 %X, 100
+  %urem = urem i32 %X, 4294967295
   %cmp = icmp eq i32 %urem, 0
   %ret = zext i1 %cmp to i32
   ret i32 %ret
 }
 
-; We can lower remainder of division by powers of two much better elsewhere;
-; also, BuildREMEqFold does not work when the only odd factor of the divisor is 1.
-; This ensures we don't touch powers of two.
-define i32 @test_urem_pow2(i32 %X) nounwind readnone {
+; We can lower remainder of division by powers of two much better elsewhere.
+define i32 @test_urem_pow2(i32 %X) nounwind {
 ; CHECK-LABEL: test_urem_pow2:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    tst w0, #0xf

diff --git a/test/CodeGen/X86/urem-seteq-vec-nonsplat.ll b/test/CodeGen/X86/urem-seteq-vec-nonsplat.ll
index 92087da1a1577764a4f036f4a1d555703ea393dc..39a65c47a78a89df9b37dac99e7e389663e75fc9 100644 (file)
--- a/test/CodeGen/X86/urem-seteq-vec-nonsplat.ll
+++ b/test/CodeGen/X86/urem-seteq-vec-nonsplat.ll
@@ -7,36 +7,36 @@
 
 ; At the moment, BuildUREMEqFold does not handle nonsplat vectors.
 
-define <4 x i32> @test_urem_odd_div(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_odd_div:
+; Odd+Even divisors
+define <4 x i32> @test_urem_odd_even(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_even:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [2863311531,3435973837,613566757,954437177]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
 ; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; CHECK-SSE2-NEXT:    psubd %xmm2, %xmm1
-; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm1
-; CHECK-SSE2-NEXT:    psrlq $32, %xmm1
-; CHECK-SSE2-NEXT:    paddd %xmm2, %xmm1
+; CHECK-SSE2-NEXT:    psrld $1, %xmm1
 ; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
-; CHECK-SSE2-NEXT:    psrld $1, %xmm2
-; CHECK-SSE2-NEXT:    psrld $2, %xmm1
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,1],xmm0[3,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [3435973837,2454267027,1374389535,1374389535]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm4
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[1,3,2,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm1
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    psrld $5, %xmm2
 ; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    psrld $2, %xmm3
+; CHECK-SSE2-NEXT:    psrld $3, %xmm1
+; CHECK-SSE2-NEXT:    movsd {{.*#+}} xmm1 = xmm3[0],xmm1[1]
 ; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm2[3,3]
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [3,5,7,9]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm5
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[0,2,2,3]
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,3],xmm1[1,2]
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2,3,1]
-; CHECK-SSE2-NEXT:    pmuludq %xmm4, %xmm2
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [5,14,25,100]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm4
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm1
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
 ; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
@@ -44,24 +44,245 @@ define <4 x i32> @test_urem_odd_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_odd_div:
+; CHECK-SSE41-LABEL: test_urem_odd_even:
 ; CHECK-SSE41:       # %bb.0:
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [2863311531,3435973837,613566757,954437177]
+; CHECK-SSE41-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-SSE41-NEXT:    psrld $1, %xmm1
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm3
-; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm1
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [3435973837,2454267027,1374389535,1374389535]
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
+; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm4
+; CHECK-SSE41-NEXT:    pmuludq %xmm3, %xmm1
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3],xmm1[4,5],xmm4[6,7]
+; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE41-NEXT:    psrld $5, %xmm2
+; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-SSE41-NEXT:    psrld $2, %xmm3
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-SSE41-NEXT:    psrld $3, %xmm1
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm1
+; CHECK-SSE41-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    psrld $31, %xmm0
+; CHECK-SSE41-NEXT:    retq
+;
+; CHECK-AVX1-LABEL: test_urem_odd_even:
+; CHECK-AVX1:       # %bb.0:
+; CHECK-AVX1-NEXT:    vpsrld $1, %xmm0, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [3435973837,2454267027,1374389535,1374389535]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpmuludq %xmm4, %xmm2, %xmm2
+; CHECK-AVX1-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $5, %xmm1, %xmm2
+; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm3
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpsrld $3, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_urem_odd_even:
+; CHECK-AVX2:       # %bb.0:
+; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,1374389535,1374389535]
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm4, %xmm2
+; CHECK-AVX2-NEXT:    vpmuludq %xmm1, %xmm3, %xmm1
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
+; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512VL-LABEL: test_urem_odd_even:
+; CHECK-AVX512VL:       # %bb.0:
+; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,1374389535,1374389535]
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm4, %xmm2
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm1, %xmm3, %xmm1
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
+; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    retq
+  %urem = urem <4 x i32> %X, <i32 5, i32 14, i32 25, i32 100>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
+}
+
+; One all-ones divisor in odd divisor
+define <4 x i32> @test_urem_odd_allones(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_allones:
+; CHECK-SSE2:       # %bb.0:
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = <3435973837,u,2147483649,u>
+; CHECK-SSE2-NEXT:    pmuludq %xmm0, %xmm1
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    psrld $2, %xmm2
+; CHECK-SSE2-NEXT:    psrld $31, %xmm1
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    psrld $31, %xmm0
+; CHECK-SSE2-NEXT:    retq
+;
+; CHECK-SSE41-LABEL: test_urem_odd_allones:
+; CHECK-SSE41:       # %bb.0:
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-SSE41-NEXT:    pmuludq {{.*}}(%rip), %xmm1
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm2 = <3435973837,u,2147483649,u>
+; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm2
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-SSE41-NEXT:    movdqa %xmm2, %xmm1
+; CHECK-SSE41-NEXT:    psrld $31, %xmm1
+; CHECK-SSE41-NEXT:    psrld $2, %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm2
+; CHECK-SSE41-NEXT:    psubd %xmm2, %xmm0
+; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    psrld $31, %xmm0
+; CHECK-SSE41-NEXT:    retq
+;
+; CHECK-AVX1-LABEL: test_urem_odd_allones:
+; CHECK-AVX1:       # %bb.0:
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm0, %xmm2
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm1, %xmm2
+; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_urem_odd_allones:
+; CHECK-AVX2:       # %bb.0:
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpmuludq {{.*}}(%rip), %xmm0, %xmm2
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512VL-LABEL: test_urem_odd_allones:
+; CHECK-AVX512VL:       # %bb.0:
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmuludq {{.*}}(%rip), %xmm0, %xmm2
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    retq
+  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 4294967295, i32 5>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
+}
+
+; One all-ones divisor in even divisor
+define <4 x i32> @test_urem_even_allones(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_even_allones:
+; CHECK-SSE2:       # %bb.0:
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-SSE2-NEXT:    psrld $1, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm1
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,3,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    psrld $2, %xmm2
+; CHECK-SSE2-NEXT:    psrld $31, %xmm1
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    psrld $31, %xmm0
+; CHECK-SSE2-NEXT:    retq
+;
+; CHECK-SSE41-LABEL: test_urem_even_allones:
+; CHECK-SSE41:       # %bb.0:
+; CHECK-SSE41-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-SSE41-NEXT:    psrld $1, %xmm1
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
+; CHECK-SSE41-NEXT:    pmuludq {{.*}}(%rip), %xmm1
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
-; CHECK-SSE41-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-SSE41-NEXT:    psubd %xmm1, %xmm2
 ; CHECK-SSE41-NEXT:    pmuludq {{.*}}(%rip), %xmm2
-; CHECK-SSE41-NEXT:    psrlq $32, %xmm2
-; CHECK-SSE41-NEXT:    paddd %xmm1, %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
 ; CHECK-SSE41-NEXT:    movdqa %xmm2, %xmm1
-; CHECK-SSE41-NEXT:    psrld $2, %xmm1
-; CHECK-SSE41-NEXT:    psrld $1, %xmm2
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3,4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    psrld $31, %xmm1
+; CHECK-SSE41-NEXT:    psrld $2, %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7]
 ; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm2
 ; CHECK-SSE41-NEXT:    psubd %xmm2, %xmm0
 ; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
@@ -69,22 +290,18 @@ define <4 x i32> @test_urem_odd_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_odd_div:
+; CHECK-AVX1-LABEL: test_urem_even_allones:
 ; CHECK-AVX1:       # %bb.0:
-; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,3435973837,613566757,954437177]
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
-; CHECK-AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
-; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm2
+; CHECK-AVX1-NEXT:    vpsrld $1, %xmm0, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
 ; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm2, %xmm2
-; CHECK-AVX1-NEXT:    vpsrlq $32, %xmm2, %xmm2
-; CHECK-AVX1-NEXT:    vpaddd %xmm1, %xmm2, %xmm1
-; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm2
-; CHECK-AVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3,4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm1, %xmm2
+; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -92,19 +309,15 @@ define <4 x i32> @test_urem_odd_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_odd_div:
+; CHECK-AVX2-LABEL: test_urem_even_allones:
 ; CHECK-AVX2:       # %bb.0:
-; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,3435973837,613566757,954437177]
+; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm1
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
-; CHECK-AVX2-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
+; CHECK-AVX2-NEXT:    vpmuludq {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
-; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm2
-; CHECK-AVX2-NEXT:    vpmuludq {{.*}}(%rip), %xmm2, %xmm2
-; CHECK-AVX2-NEXT:    vpsrlq $32, %xmm2, %xmm2
-; CHECK-AVX2-NEXT:    vpaddd %xmm1, %xmm2, %xmm1
 ; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
@@ -113,19 +326,15 @@ define <4 x i32> @test_urem_odd_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_odd_div:
+; CHECK-AVX512VL-LABEL: test_urem_even_allones:
 ; CHECK-AVX512VL:       # %bb.0:
-; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,3435973837,613566757,954437177]
+; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
-; CHECK-AVX512VL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
+; CHECK-AVX512VL-NEXT:    vpmuludq {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
-; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm2
-; CHECK-AVX512VL-NEXT:    vpmuludq {{.*}}(%rip), %xmm2, %xmm2
-; CHECK-AVX512VL-NEXT:    vpsrlq $32, %xmm2, %xmm2
-; CHECK-AVX512VL-NEXT:    vpaddd %xmm1, %xmm2, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
@@ -133,39 +342,41 @@ define <4 x i32> @test_urem_odd_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 3, i32 5, i32 7, i32 9>
+  %urem = urem <4 x i32> %X, <i32 14, i32 14, i32 4294967295, i32 14>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_even_div(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_even_div:
+; One all-ones divisor in odd+even divisor
+define <4 x i32> @test_urem_odd_even_allones(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_even_allones:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [2863311531,3435973837,2863311531,2454267027]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm3
-; CHECK-SSE2-NEXT:    psrld $1, %xmm3
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm4
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm4 = xmm4[1,1],xmm3[3,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm4, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-SSE2-NEXT:    psrld $1, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,1],xmm0[3,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [3435973837,2454267027,2147483649,1374389535]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm4
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[1,3,2,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm1
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm1
-; CHECK-SSE2-NEXT:    psrld $2, %xmm1
-; CHECK-SSE2-NEXT:    psrld $3, %xmm2
-; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm3
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm1[3,3]
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [6,10,12,14]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm5
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[0,2,2,3]
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm2[1,2]
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2,3,1]
-; CHECK-SSE2-NEXT:    pmuludq %xmm4, %xmm1
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    psrld $5, %xmm2
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    psrld $2, %xmm3
+; CHECK-SSE2-NEXT:    psrld $31, %xmm1
+; CHECK-SSE2-NEXT:    movsd {{.*#+}} xmm1 = xmm3[0],xmm1[1]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm2[3,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [5,14,4294967295,100]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm4
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm1
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
 ; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
@@ -174,22 +385,26 @@ define <4 x i32> @test_urem_even_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_even_div:
+; CHECK-SSE41-LABEL: test_urem_odd_even_allones:
 ; CHECK-SSE41:       # %bb.0:
 ; CHECK-SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; CHECK-SSE41-NEXT:    psrld $1, %xmm1
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5],xmm1[6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [2863311531,3435973837,2863311531,2454267027]
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [3435973837,2454267027,2147483649,1374389535]
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
 ; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm4
 ; CHECK-SSE41-NEXT:    pmuludq %xmm3, %xmm1
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3],xmm1[4,5],xmm4[6,7]
 ; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm2
-; CHECK-SSE41-NEXT:    psrld $3, %xmm2
-; CHECK-SSE41-NEXT:    psrld $2, %xmm1
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3,4,5],xmm1[6,7]
+; CHECK-SSE41-NEXT:    psrld $5, %xmm2
+; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-SSE41-NEXT:    psrld $2, %xmm3
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-SSE41-NEXT:    psrld $31, %xmm1
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
 ; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm1
 ; CHECK-SSE41-NEXT:    psubd %xmm1, %xmm0
 ; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
@@ -197,20 +412,23 @@ define <4 x i32> @test_urem_even_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_even_div:
+; CHECK-AVX1-LABEL: test_urem_odd_even_allones:
 ; CHECK-AVX1:       # %bb.0:
 ; CHECK-AVX1-NEXT:    vpsrld $1, %xmm0, %xmm1
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [2863311531,3435973837,2863311531,2454267027]
+; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [3435973837,2454267027,2147483649,1374389535]
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
 ; CHECK-AVX1-NEXT:    vpmuludq %xmm4, %xmm2, %xmm2
 ; CHECK-AVX1-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
-; CHECK-AVX1-NEXT:    vpsrld $3, %xmm1, %xmm2
-; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3,4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $5, %xmm1, %xmm2
+; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm3
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -218,9 +436,9 @@ define <4 x i32> @test_urem_even_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_even_div:
+; CHECK-AVX2-LABEL: test_urem_odd_even_allones:
 ; CHECK-AVX2:       # %bb.0:
-; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,3435973837,2863311531,2454267027]
+; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,2147483649,1374389535]
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
 ; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
@@ -236,9 +454,9 @@ define <4 x i32> @test_urem_even_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_even_div:
+; CHECK-AVX512VL-LABEL: test_urem_odd_even_allones:
 ; CHECK-AVX512VL:       # %bb.0:
-; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,3435973837,2863311531,2454267027]
+; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,2147483649,1374389535]
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
 ; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
@@ -253,60 +471,49 @@ define <4 x i32> @test_urem_even_div(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 6, i32 10, i32 12, i32 14>
+  %urem = urem <4 x i32> %X, <i32 5, i32 14, i32 4294967295, i32 100>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-; Can't fold due to last line
-define <4 x i32> @test_urem_pow2(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_pow2:
+; One power-of-two divisor in odd divisor
+define <4 x i32> @test_urem_odd_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_poweroftwo:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [2863311531,3435973837,2863311531,268435456]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = <3435973837,u,268435456,u>
+; CHECK-SSE2-NEXT:    pmuludq %xmm0, %xmm1
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm2
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm1
-; CHECK-SSE2-NEXT:    psrld $3, %xmm1
-; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm3
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm2[3,3]
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [6,10,12,16]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm5
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
 ; CHECK-SSE2-NEXT:    psrld $2, %xmm2
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,3],xmm1[2,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm4, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm2
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
 ; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_pow2:
+; CHECK-SSE41-LABEL: test_urem_odd_poweroftwo:
 ; CHECK-SSE41:       # %bb.0:
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [2863311531,3435973837,2863311531,268435456]
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm3
-; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm1
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
-; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm2
-; CHECK-SSE41-NEXT:    psrld $3, %xmm2
-; CHECK-SSE41-NEXT:    movdqa %xmm2, %xmm3
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm1[4,5,6,7]
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-SSE41-NEXT:    pmuludq {{.*}}(%rip), %xmm1
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm2 = <3435973837,u,268435456,u>
+; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm2
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
 ; CHECK-SSE41-NEXT:    psrld $2, %xmm1
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5,6,7]
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
 ; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm1
 ; CHECK-SSE41-NEXT:    psubd %xmm1, %xmm0
 ; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
@@ -314,20 +521,120 @@ define <4 x i32> @test_urem_pow2(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_pow2:
+; CHECK-AVX1-LABEL: test_urem_odd_poweroftwo:
 ; CHECK-AVX1:       # %bb.0:
-; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,3435973837,2863311531,268435456]
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
-; CHECK-AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm0, %xmm2
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_urem_odd_poweroftwo:
+; CHECK-AVX2:       # %bb.0:
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpmuludq {{.*}}(%rip), %xmm0, %xmm2
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512VL-LABEL: test_urem_odd_poweroftwo:
+; CHECK-AVX512VL:       # %bb.0:
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmuludq {{.*}}(%rip), %xmm0, %xmm2
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    retq
+  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 16, i32 5>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
+}
+
+; One power-of-two divisor in even divisor
+define <4 x i32> @test_urem_even_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_even_poweroftwo:
+; CHECK-SSE2:       # %bb.0:
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-SSE2-NEXT:    psrld $1, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm1
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,3,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    psrld $2, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    psrld $31, %xmm0
+; CHECK-SSE2-NEXT:    retq
+;
+; CHECK-SSE41-LABEL: test_urem_even_poweroftwo:
+; CHECK-SSE41:       # %bb.0:
+; CHECK-SSE41-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-SSE41-NEXT:    psrld $1, %xmm1
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
+; CHECK-SSE41-NEXT:    pmuludq {{.*}}(%rip), %xmm1
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-SSE41-NEXT:    pmuludq {{.*}}(%rip), %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    psrld $2, %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm2
+; CHECK-SSE41-NEXT:    psubd %xmm2, %xmm0
+; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    psrld $31, %xmm0
+; CHECK-SSE41-NEXT:    retq
+;
+; CHECK-AVX1-LABEL: test_urem_even_poweroftwo:
+; CHECK-AVX1:       # %bb.0:
+; CHECK-AVX1-NEXT:    vpsrld $1, %xmm0, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm2, %xmm2
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
-; CHECK-AVX1-NEXT:    vpsrld $3, %xmm1, %xmm2
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm2[0,1,2,3],xmm1[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
 ; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5,6,7]
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -335,13 +642,13 @@ define <4 x i32> @test_urem_pow2(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_pow2:
+; CHECK-AVX2-LABEL: test_urem_even_poweroftwo:
 ; CHECK-AVX2:       # %bb.0:
-; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,3435973837,2863311531,268435456]
+; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm1
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
-; CHECK-AVX2-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
+; CHECK-AVX2-NEXT:    vpmuludq {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
@@ -352,13 +659,13 @@ define <4 x i32> @test_urem_pow2(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_pow2:
+; CHECK-AVX512VL-LABEL: test_urem_even_poweroftwo:
 ; CHECK-AVX512VL:       # %bb.0:
-; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,3435973837,2863311531,268435456]
+; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
-; CHECK-AVX512VL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
+; CHECK-AVX512VL-NEXT:    vpmuludq {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
@@ -368,89 +675,90 @@ define <4 x i32> @test_urem_pow2(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 6, i32 10, i32 12, i32 16>
+  %urem = urem <4 x i32> %X, <i32 14, i32 14, i32 16, i32 14>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-; Can't fold due to second line
-define <4 x i32> @test_urem_one(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_one:
+; One power-of-two divisor in odd+even divisor
+define <4 x i32> @test_urem_odd_even_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_even_poweroftwo:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [2863311531,0,2863311531,2454267027]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm3
-; CHECK-SSE2-NEXT:    psrld $1, %xmm3
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm4
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm4 = xmm4[1,1],xmm3[3,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm4, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-SSE2-NEXT:    psrld $1, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,1],xmm0[3,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [3435973837,2454267027,268435456,1374389535]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm4
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[1,3,2,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm1
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm1
-; CHECK-SSE2-NEXT:    psrld $2, %xmm1
-; CHECK-SSE2-NEXT:    psrld $3, %xmm2
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm3
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,0],xmm1[0,0]
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[2,0],xmm2[0,2]
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [6,1,12,14]
-; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm3
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm4
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm4 = xmm4[1,1],xmm1[3,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm4, %xmm1
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    psrld $5, %xmm2
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    psrld $2, %xmm3
+; CHECK-SSE2-NEXT:    movsd {{.*#+}} xmm1 = xmm3[0],xmm1[1]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm2[3,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [5,14,16,100]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm4
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm1
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
-; CHECK-SSE2-NEXT:    psubd %xmm3, %xmm0
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
+; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
 ; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_one:
+; CHECK-SSE41-LABEL: test_urem_odd_even_poweroftwo:
 ; CHECK-SSE41:       # %bb.0:
 ; CHECK-SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; CHECK-SSE41-NEXT:    psrld $1, %xmm1
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5],xmm1[6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [2863311531,0,2863311531,2454267027]
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [3435973837,2454267027,268435456,1374389535]
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
 ; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm4
 ; CHECK-SSE41-NEXT:    pmuludq %xmm3, %xmm1
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3],xmm1[4,5],xmm4[6,7]
 ; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm2
-; CHECK-SSE41-NEXT:    psrld $3, %xmm2
-; CHECK-SSE41-NEXT:    psrld $2, %xmm1
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3],xmm1[4,5,6,7]
-; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm1
-; CHECK-SSE41-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    psrld $5, %xmm2
+; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-SSE41-NEXT:    psrld $2, %xmm3
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm1[4,5,6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm3
+; CHECK-SSE41-NEXT:    psubd %xmm3, %xmm0
 ; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
 ; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_one:
+; CHECK-AVX1-LABEL: test_urem_odd_even_poweroftwo:
 ; CHECK-AVX1:       # %bb.0:
 ; CHECK-AVX1-NEXT:    vpsrld $1, %xmm0, %xmm1
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [2863311531,0,2863311531,2454267027]
+; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [3435973837,2454267027,268435456,1374389535]
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
 ; CHECK-AVX1-NEXT:    vpmuludq %xmm4, %xmm2, %xmm2
 ; CHECK-AVX1-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
-; CHECK-AVX1-NEXT:    vpsrld $3, %xmm1, %xmm2
-; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3],xmm1[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpsrld $5, %xmm1, %xmm2
+; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm3
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -458,9 +766,9 @@ define <4 x i32> @test_urem_one(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_one:
+; CHECK-AVX2-LABEL: test_urem_odd_even_poweroftwo:
 ; CHECK-AVX2:       # %bb.0:
-; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,0,2863311531,2454267027]
+; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,268435456,1374389535]
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
 ; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
@@ -469,7 +777,6 @@ define <4 x i32> @test_urem_one(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
-; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm0[1],xmm1[2,3]
 ; CHECK-AVX2-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -477,9 +784,9 @@ define <4 x i32> @test_urem_one(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_one:
+; CHECK-AVX512VL-LABEL: test_urem_odd_even_poweroftwo:
 ; CHECK-AVX512VL:       # %bb.0:
-; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,0,2863311531,2454267027]
+; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,268435456,1374389535]
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
 ; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
@@ -488,47 +795,162 @@ define <4 x i32> @test_urem_one(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
-; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm0[1],xmm1[2,3]
 ; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 6, i32 1, i32 12, i32 14>
+  %urem = urem <4 x i32> %X, <i32 5, i32 14, i32 16, i32 100>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_nomulinv(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_nomulinv:
+; One all-ones divisor and one power-of-two divisor in odd divisor
+define <4 x i32> @test_urem_odd_allones_and_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_allones_and_poweroftwo:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [2863311531,1374389535,2863311531,2454267027]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [3435973837,268435456,2147483649,3435973837]
 ; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
 ; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm3
-; CHECK-SSE2-NEXT:    psrld $1, %xmm3
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm4
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm4 = xmm4[1,1],xmm3[3,3]
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm1
+; CHECK-SSE2-NEXT:    psrld $2, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm3
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm1[3,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [5,16,4294967295,5]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm5
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[0,2,2,3]
+; CHECK-SSE2-NEXT:    psrld $31, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm2[2,3]
 ; CHECK-SSE2-NEXT:    pmuludq %xmm4, %xmm1
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
+; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    psrld $31, %xmm0
+; CHECK-SSE2-NEXT:    retq
+;
+; CHECK-SSE41-LABEL: test_urem_odd_allones_and_poweroftwo:
+; CHECK-SSE41:       # %bb.0:
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [3435973837,268435456,2147483649,3435973837]
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm3
+; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm1
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE41-NEXT:    psrld $2, %xmm2
+; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-SSE41-NEXT:    psrld $31, %xmm1
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm1
+; CHECK-SSE41-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    psrld $31, %xmm0
+; CHECK-SSE41-NEXT:    retq
+;
+; CHECK-AVX1-LABEL: test_urem_odd_allones_and_poweroftwo:
+; CHECK-AVX1:       # %bb.0:
+; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,268435456,2147483649,3435973837]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
+; CHECK-AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm2
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm1[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_urem_odd_allones_and_poweroftwo:
+; CHECK-AVX2:       # %bb.0:
+; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,268435456,2147483649,3435973837]
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
+; CHECK-AVX2-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
+; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512VL-LABEL: test_urem_odd_allones_and_poweroftwo:
+; CHECK-AVX512VL:       # %bb.0:
+; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,268435456,2147483649,3435973837]
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
+; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    retq
+  %urem = urem <4 x i32> %X, <i32 5, i32 16, i32 4294967295, i32 5>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
+}
+
+; One all-ones divisor and one power-of-two divisor in even divisor
+define <4 x i32> @test_urem_even_allones_and_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_even_allones_and_poweroftwo:
+; CHECK-SSE2:       # %bb.0:
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-SSE2-NEXT:    psrld $1, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,1],xmm1[3,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [2454267027,268435456,2147483649,2454267027]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm4
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[1,3,2,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[1,2]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2,3,1]
+; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm1
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm1
-; CHECK-SSE2-NEXT:    psrld $2, %xmm1
-; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm3
-; CHECK-SSE2-NEXT:    psrld $5, %xmm3
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm1[3,3]
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [6,100,12,14]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    psrld $2, %xmm2
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm2[3,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [14,16,4294967295,14]
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
 ; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm5
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[0,2,2,3]
-; CHECK-SSE2-NEXT:    psrld $3, %xmm2
-; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm2[2,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm4, %xmm1
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; CHECK-SSE2-NEXT:    psrld $31, %xmm1
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,3],xmm1[2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm4, %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
 ; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
 ; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
@@ -536,13 +958,13 @@ define <4 x i32> @test_urem_nomulinv(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_nomulinv:
+; CHECK-SSE41-LABEL: test_urem_even_allones_and_poweroftwo:
 ; CHECK-SSE41:       # %bb.0:
 ; CHECK-SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; CHECK-SSE41-NEXT:    psrld $1, %xmm1
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5],xmm1[6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [2863311531,1374389535,2863311531,2454267027]
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [2454267027,268435456,2147483649,2454267027]
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
 ; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm4
 ; CHECK-SSE41-NEXT:    pmuludq %xmm3, %xmm1
@@ -551,9 +973,8 @@ define <4 x i32> @test_urem_nomulinv(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm2
 ; CHECK-SSE41-NEXT:    psrld $2, %xmm2
 ; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm3
-; CHECK-SSE41-NEXT:    psrld $5, %xmm3
 ; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
-; CHECK-SSE41-NEXT:    psrld $3, %xmm1
+; CHECK-SSE41-NEXT:    psrld $31, %xmm1
 ; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
 ; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
 ; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm1
@@ -563,21 +984,20 @@ define <4 x i32> @test_urem_nomulinv(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_nomulinv:
+; CHECK-AVX1-LABEL: test_urem_even_allones_and_poweroftwo:
 ; CHECK-AVX1:       # %bb.0:
 ; CHECK-AVX1-NEXT:    vpsrld $1, %xmm0, %xmm1
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [2863311531,1374389535,2863311531,2454267027]
+; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [2454267027,268435456,2147483649,2454267027]
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
 ; CHECK-AVX1-NEXT:    vpmuludq %xmm4, %xmm2, %xmm2
 ; CHECK-AVX1-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
 ; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm2
-; CHECK-AVX1-NEXT:    vpsrld $5, %xmm1, %xmm3
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
-; CHECK-AVX1-NEXT:    vpsrld $3, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm1[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
 ; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
@@ -587,9 +1007,9 @@ define <4 x i32> @test_urem_nomulinv(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_nomulinv:
+; CHECK-AVX2-LABEL: test_urem_even_allones_and_poweroftwo:
 ; CHECK-AVX2:       # %bb.0:
-; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,1374389535,2863311531,2454267027]
+; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [2454267027,268435456,2147483649,2454267027]
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
 ; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
@@ -605,9 +1025,9 @@ define <4 x i32> @test_urem_nomulinv(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_nomulinv:
+; CHECK-AVX512VL-LABEL: test_urem_even_allones_and_poweroftwo:
 ; CHECK-AVX512VL:       # %bb.0:
-; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [2863311531,1374389535,2863311531,2454267027]
+; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [2454267027,268435456,2147483649,2454267027]
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
 ; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
@@ -622,232 +1042,281 @@ define <4 x i32> @test_urem_nomulinv(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 6, i32 100, i32 12, i32 14>
+  %urem = urem <4 x i32> %X, <i32 14, i32 16, i32 4294967295, i32 14>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_comp(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_comp:
+; One all-ones divisor and one power-of-two divisor in odd+even divisor
+define <4 x i32> @test_urem_odd_even_allones_and_poweroftwo(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_even_allones_and_poweroftwo:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [3435973837,268435456,2147483649,1374389535]
 ; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
 ; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
 ; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
 ; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-SSE2-NEXT:    psrld $2, %xmm2
 ; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm1
-; CHECK-SSE2-NEXT:    pslld $2, %xmm1
-; CHECK-SSE2-NEXT:    paddd %xmm2, %xmm1
+; CHECK-SSE2-NEXT:    psrld $5, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm3
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm1[3,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [5,16,4294967295,100]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm1[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm4
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
+; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm4
+; CHECK-SSE2-NEXT:    psrld $31, %xmm4
+; CHECK-SSE2-NEXT:    psrld $2, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,3],xmm4[2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
 ; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
-; CHECK-SSE2-NEXT:    pcmpeqd {{.*}}(%rip), %xmm0
+; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_comp:
+; CHECK-SSE41-LABEL: test_urem_odd_even_allones_and_poweroftwo:
 ; CHECK-SSE41:       # %bb.0:
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
-; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm1
-; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm2
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
-; CHECK-SSE41-NEXT:    psrld $2, %xmm2
-; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm2
-; CHECK-SSE41-NEXT:    psubd %xmm2, %xmm0
-; CHECK-SSE41-NEXT:    pcmpeqd {{.*}}(%rip), %xmm0
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [3435973837,268435456,2147483649,1374389535]
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm3
+; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm1
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE41-NEXT:    psrld $5, %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm1[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-SSE41-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-SSE41-NEXT:    psrld $31, %xmm3
+; CHECK-SSE41-NEXT:    psrld $2, %xmm1
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5,6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm1
+; CHECK-SSE41-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_comp:
+; CHECK-AVX1-LABEL: test_urem_odd_even_allones_and_poweroftwo:
 ; CHECK-AVX1:       # %bb.0:
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
-; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
-; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,268435456,2147483649,1374389535]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
+; CHECK-AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $5, %xmm1, %xmm2
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm1[0,1,2,3],xmm2[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm1, %xmm3
 ; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
-; CHECK-AVX1-NEXT:    vpcmpeqd {{.*}}(%rip), %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_comp:
+; CHECK-AVX2-LABEL: test_urem_odd_even_allones_and_poweroftwo:
 ; CHECK-AVX2:       # %bb.0:
-; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
-; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
-; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
-; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
-; CHECK-AVX2-NEXT:    vpsrld $2, %xmm1, %xmm1
-; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [5,5,5,5]
-; CHECK-AVX2-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,268435456,2147483649,1374389535]
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
+; CHECK-AVX2-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
+; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
-; CHECK-AVX2-NEXT:    vpcmpeqd {{.*}}(%rip), %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_comp:
+; CHECK-AVX512VL-LABEL: test_urem_odd_even_allones_and_poweroftwo:
 ; CHECK-AVX512VL:       # %bb.0:
-; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
-; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
-; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
-; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
-; CHECK-AVX512VL-NEXT:    vpsrld $2, %xmm1, %xmm1
-; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip){1to4}, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,268435456,2147483649,1374389535]
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
+; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
-; CHECK-AVX512VL-NEXT:    vpcmpeqd {{.*}}(%rip), %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 5, i32 5>
-  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 1, i32 0>
+  %urem = urem <4 x i32> %X, <i32 5, i32 16, i32 4294967295, i32 100>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_both(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_both:
+;------------------------------------------------------------------------------;
+; Negative tests - the fold is invalid if any divisor is 1.
+;------------------------------------------------------------------------------;
+
+; One divisor in odd divisor
+define <4 x i32> @test_urem_odd_one(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_one:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = <2863311531,u,2863311531,u>
+; CHECK-SSE2-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-SSE2-NEXT:    movd %eax, %xmm1
 ; CHECK-SSE2-NEXT:    pmuludq %xmm0, %xmm1
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
 ; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm2
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
 ; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; CHECK-SSE2-NEXT:    psrld $2, %xmm1
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; CHECK-SSE2-NEXT:    psrld $3, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
 ; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm1
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm2
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
 ; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
-; CHECK-SSE2-NEXT:    pcmpeqd {{.*}}(%rip), %xmm0
+; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_both:
+; CHECK-SSE41-LABEL: test_urem_odd_one:
 ; CHECK-SSE41:       # %bb.0:
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-SSE41-NEXT:    pmuludq {{.*}}(%rip), %xmm1
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm2 = <2863311531,u,2863311531,u>
-; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm2
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
-; CHECK-SSE41-NEXT:    psrld $2, %xmm2
+; CHECK-SSE41-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-SSE41-NEXT:    movd %eax, %xmm1
+; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm1
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; CHECK-SSE41-NEXT:    pmuludq {{.*}}(%rip), %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    psrld $3, %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm0[4,5],xmm2[6,7]
 ; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm2
 ; CHECK-SSE41-NEXT:    psubd %xmm2, %xmm0
-; CHECK-SSE41-NEXT:    pcmpeqd {{.*}}(%rip), %xmm0
+; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_both:
+; CHECK-AVX1-LABEL: test_urem_odd_one:
 ; CHECK-AVX1:       # %bb.0:
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm1, %xmm1
-; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm0, %xmm2
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
-; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-AVX1-NEXT:    vmovd %eax, %xmm1
+; CHECK-AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm2, %xmm2
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $3, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
-; CHECK-AVX1-NEXT:    vpcmpeqd {{.*}}(%rip), %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_both:
+; CHECK-AVX2-LABEL: test_urem_odd_one:
 ; CHECK-AVX2:       # %bb.0:
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
 ; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
-; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [2863311531,2863311531,2863311531,2863311531]
+; CHECK-AVX2-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-AVX2-NEXT:    vmovd %eax, %xmm2
 ; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
-; CHECK-AVX2-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpsrld $3, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
 ; CHECK-AVX2-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
-; CHECK-AVX2-NEXT:    vpcmpeqd {{.*}}(%rip), %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_both:
+; CHECK-AVX512VL-LABEL: test_urem_odd_one:
 ; CHECK-AVX512VL:       # %bb.0:
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
 ; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
-; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [2863311531,2863311531,2863311531,2863311531]
+; CHECK-AVX512VL-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-AVX512VL-NEXT:    vmovd %eax, %xmm2
 ; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
-; CHECK-AVX512VL-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpsrld $3, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
 ; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
-; CHECK-AVX512VL-NEXT:    vpcmpeqd {{.*}}(%rip), %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 6, i32 5, i32 6, i32 5>
-  %cmp = icmp eq <4 x i32> %urem, <i32 1, i32 0, i32 1, i32 0>
-  %ret = zext <4 x i1> %cmp to <4 x i32>
-  ret <4 x i32> %ret
-}
-
-define <4 x i32> @test_urem_div_undef(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE-LABEL: test_urem_div_undef:
-; CHECK-SSE:       # %bb.0:
-; CHECK-SSE-NEXT:    xorps %xmm0, %xmm0
-; CHECK-SSE-NEXT:    retq
-;
-; CHECK-AVX-LABEL: test_urem_div_undef:
-; CHECK-AVX:       # %bb.0:
-; CHECK-AVX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; CHECK-AVX-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 undef, i32 5>
+  %urem = urem <4 x i32> %X, <i32 25, i32 25, i32 1, i32 25>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_comp_undef(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_comp_undef:
+; One divisor in even divisors
+define <4 x i32> @test_urem_even_one(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_even_one:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-SSE2-NEXT:    movd %eax, %xmm1
+; CHECK-SSE2-NEXT:    pmuludq %xmm0, %xmm1
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm2
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-SSE2-NEXT:    psrld $2, %xmm2
-; CHECK-SSE2-NEXT:    movdqa %xmm2, %xmm1
-; CHECK-SSE2-NEXT:    pslld $2, %xmm1
-; CHECK-SSE2-NEXT:    paddd %xmm2, %xmm1
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    psrld $5, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm1
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
 ; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_comp_undef:
+; CHECK-SSE41-LABEL: test_urem_even_one:
 ; CHECK-SSE41:       # %bb.0:
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
-; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm1
-; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm2
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
-; CHECK-SSE41-NEXT:    psrld $2, %xmm2
+; CHECK-SSE41-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-SSE41-NEXT:    movd %eax, %xmm1
+; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm1
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; CHECK-SSE41-NEXT:    pmuludq {{.*}}(%rip), %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    psrld $5, %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm0[4,5],xmm2[6,7]
 ; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm2
 ; CHECK-SSE41-NEXT:    psubd %xmm2, %xmm0
 ; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
@@ -855,15 +1324,17 @@ define <4 x i32> @test_urem_comp_undef(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_comp_undef:
+; CHECK-AVX1-LABEL: test_urem_even_one:
 ; CHECK-AVX1:       # %bb.0:
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
-; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
-; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
-; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-AVX1-NEXT:    vmovd %eax, %xmm1
+; CHECK-AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm2, %xmm2
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $5, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -871,114 +1342,117 @@ define <4 x i32> @test_urem_comp_undef(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_comp_undef:
+; CHECK-AVX2-LABEL: test_urem_even_one:
 ; CHECK-AVX2:       # %bb.0:
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
 ; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-AVX2-NEXT:    vmovd %eax, %xmm2
 ; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
 ; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
-; CHECK-AVX2-NEXT:    vpsrld $2, %xmm1, %xmm1
-; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [5,5,5,5]
-; CHECK-AVX2-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpsrld $5, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
+; CHECK-AVX2-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_comp_undef:
+; CHECK-AVX512VL-LABEL: test_urem_even_one:
 ; CHECK-AVX512VL:       # %bb.0:
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
 ; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-AVX512VL-NEXT:    vmovd %eax, %xmm2
 ; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
 ; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
-; CHECK-AVX512VL-NEXT:    vpsrld $2, %xmm1, %xmm1
-; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip){1to4}, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpsrld $5, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
+; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 5, i32 5>
-  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 undef, i32 0, i32 0>
-  %ret = zext <4 x i1> %cmp to <4 x i32>
-  ret <4 x i32> %ret
-}
-
-define <4 x i32> @test_urem_both_undef(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE-LABEL: test_urem_both_undef:
-; CHECK-SSE:       # %bb.0:
-; CHECK-SSE-NEXT:    xorps %xmm0, %xmm0
-; CHECK-SSE-NEXT:    retq
-;
-; CHECK-AVX-LABEL: test_urem_both_undef:
-; CHECK-AVX:       # %bb.0:
-; CHECK-AVX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; CHECK-AVX-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 undef, i32 5>
-  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 undef, i32 0, i32 0>
+  %urem = urem <4 x i32> %X, <i32 100, i32 100, i32 1, i32 100>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-define <4 x i32> @test_urem_div_even_odd(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_div_even_odd:
+; One divisor in odd-even divisors
+define <4 x i32> @test_urem_odd_even_one(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_even_one:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [3435973837,3435973837,2863311531,2863311531]
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-SSE2-NEXT:    movd %eax, %xmm1
+; CHECK-SSE2-NEXT:    pmuludq %xmm0, %xmm1
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm2
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-SSE2-NEXT:    psrld $2, %xmm2
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [5,5,6,6]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm1
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    psrld $3, %xmm2
+; CHECK-SSE2-NEXT:    psrld $5, %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm2[3,3]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [25,100,1,25]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm3, %xmm5
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[0,2,2,3]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm2[0,0]
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm5
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm5 = xmm5[2,0],xmm2[3,0]
+; CHECK-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm5[0,2]
+; CHECK-SSE2-NEXT:    pmuludq %xmm4, %xmm1
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-SSE2-NEXT:    psubd %xmm2, %xmm0
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
+; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
 ; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_div_even_odd:
+; CHECK-SSE41-LABEL: test_urem_odd_even_one:
 ; CHECK-SSE41:       # %bb.0:
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [3435973837,3435973837,2863311531,2863311531]
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm3
+; CHECK-SSE41-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-SSE41-NEXT:    movd %eax, %xmm1
 ; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm1
 ; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
-; CHECK-SSE41-NEXT:    psrld $2, %xmm1
-; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm1
-; CHECK-SSE41-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; CHECK-SSE41-NEXT:    pmuludq {{.*}}(%rip), %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    movdqa %xmm2, %xmm1
+; CHECK-SSE41-NEXT:    psrld $5, %xmm1
+; CHECK-SSE41-NEXT:    psrld $3, %xmm2
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5,6,7]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm0[4,5],xmm2[6,7]
+; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm2
+; CHECK-SSE41-NEXT:    psubd %xmm2, %xmm0
 ; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
 ; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_div_even_odd:
+; CHECK-AVX1-LABEL: test_urem_odd_even_one:
 ; CHECK-AVX1:       # %bb.0:
-; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,3435973837,2863311531,2863311531]
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
+; CHECK-AVX1-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-AVX1-NEXT:    vmovd %eax, %xmm1
 ; CHECK-AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
 ; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpmuludq {{.*}}(%rip), %xmm2, %xmm2
 ; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
-; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpsrld $5, %xmm1, %xmm2
+; CHECK-AVX1-NEXT:    vpsrld $3, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5,6,7]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -986,16 +1460,18 @@ define <4 x i32> @test_urem_div_even_odd(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_div_even_odd:
+; CHECK-AVX2-LABEL: test_urem_odd_even_one:
 ; CHECK-AVX2:       # %bb.0:
-; CHECK-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,3435973837,2863311531,2863311531]
-; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
-; CHECK-AVX2-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
-; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
-; CHECK-AVX2-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-AVX2-NEXT:    vmovd %eax, %xmm2
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
 ; CHECK-AVX2-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -1003,23 +1479,25 @@ define <4 x i32> @test_urem_div_even_odd(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_div_even_odd:
+; CHECK-AVX512VL-LABEL: test_urem_odd_even_one:
 ; CHECK-AVX512VL:       # %bb.0:
-; CHECK-AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [3435973837,3435973837,2863311531,2863311531]
-; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
-; CHECK-AVX512VL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm1
-; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
-; CHECK-AVX512VL-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    movl $1374389535, %eax # imm = 0x51EB851F
+; CHECK-AVX512VL-NEXT:    vmovd %eax, %xmm2
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
 ; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 6, i32 6>
+  %urem = urem <4 x i32> %X, <i32 25, i32 100, i32 1, i32 25>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret

diff --git a/test/CodeGen/X86/urem-seteq-vec-splat.ll b/test/CodeGen/X86/urem-seteq-vec-splat.ll
index bc96920504158662c596eecfc6edb20d0e3292ad..e7475eec3441f40313bcc7c36b7657e95c3bef6a 100644 (file)
--- a/test/CodeGen/X86/urem-seteq-vec-splat.ll
+++ b/test/CodeGen/X86/urem-seteq-vec-splat.ll
@@ -5,12 +5,11 @@
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-AVX,CHECK-AVX2
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl < %s | FileCheck %s --check-prefixes=CHECK,CHECK-AVX,CHECK-AVX512VL
 
-; Tests BuildUREMEqFold for 4 x i32 splat vectors with odd divisor.
-; See urem-seteq.ll for justification behind constants emitted.
-define <4 x i32> @test_urem_odd_vec_i32(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_odd_vec_i32:
+; Odd divisor
+define <4 x i32> @test_urem_odd_25(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_25:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [3264175145,3264175145,3264175145,3264175145]
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
 ; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
@@ -22,16 +21,16 @@ define <4 x i32> @test_urem_odd_vec_i32(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE2-NEXT:    pandn {{.*}}(%rip), %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_odd_vec_i32:
+; CHECK-SSE41-LABEL: test_urem_odd_25:
 ; CHECK-SSE41:       # %bb.0:
 ; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm0
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [858993459,858993459,858993459,858993459]
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [171798691,171798691,171798691,171798691]
 ; CHECK-SSE41-NEXT:    pminud %xmm0, %xmm1
 ; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_odd_vec_i32:
+; CHECK-AVX1-LABEL: test_urem_odd_25:
 ; CHECK-AVX1:       # %bb.0:
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpminud {{.*}}(%rip), %xmm0, %xmm1
@@ -39,121 +38,172 @@ define <4 x i32> @test_urem_odd_vec_i32(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_odd_vec_i32:
+; CHECK-AVX2-LABEL: test_urem_odd_25:
 ; CHECK-AVX2:       # %bb.0:
-; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837]
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [3264175145,3264175145,3264175145,3264175145]
 ; CHECK-AVX2-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
-; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [858993459,858993459,858993459,858993459]
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [171798691,171798691,171798691,171798691]
 ; CHECK-AVX2-NEXT:    vpminud %xmm1, %xmm0, %xmm1
 ; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_odd_vec_i32:
+; CHECK-AVX512VL-LABEL: test_urem_odd_25:
 ; CHECK-AVX512VL:       # %bb.0:
 ; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpminud {{.*}}(%rip){1to4}, %xmm0, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 5, i32 5, i32 5, i32 5>
+  %urem = urem <4 x i32> %X, <i32 25, i32 25, i32 25, i32 25>
   %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-; Like test_urem_odd_vec_i32, but with 8 x i16 vectors.
-define <8 x i16> @test_urem_odd_vec_i16(<8 x i16> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_odd_vec_i16:
+; Even divisors
+define <4 x i32> @test_urem_even_100(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_even_100:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm0
-; CHECK-SSE2-NEXT:    psubusw {{.*}}(%rip), %xmm0
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; CHECK-SSE2-NEXT:    psrld $5, %xmm2
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [100,100,100,100]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; CHECK-SSE2-NEXT:    psubd %xmm2, %xmm0
 ; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
-; CHECK-SSE2-NEXT:    pcmpeqw %xmm1, %xmm0
-; CHECK-SSE2-NEXT:    psrlw $15, %xmm0
+; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_odd_vec_i16:
+; CHECK-SSE41-LABEL: test_urem_even_100:
 ; CHECK-SSE41:       # %bb.0:
-; CHECK-SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm0
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [13107,13107,13107,13107,13107,13107,13107,13107]
-; CHECK-SSE41-NEXT:    pminuw %xmm0, %xmm1
-; CHECK-SSE41-NEXT:    pcmpeqw %xmm1, %xmm0
-; CHECK-SSE41-NEXT:    psrlw $15, %xmm0
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm1
+; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm2
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-SSE41-NEXT:    psrld $5, %xmm2
+; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm2
+; CHECK-SSE41-NEXT:    psubd %xmm2, %xmm0
+; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX-LABEL: test_urem_odd_vec_i16:
-; CHECK-AVX:       # %bb.0:
-; CHECK-AVX-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
-; CHECK-AVX-NEXT:    vpminuw {{.*}}(%rip), %xmm0, %xmm1
-; CHECK-AVX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
-; CHECK-AVX-NEXT:    vpsrlw $15, %xmm0, %xmm0
-; CHECK-AVX-NEXT:    retq
-  %urem = urem <8 x i16> %X, <i16 5, i16 5, i16 5, i16 5,
-                              i16 5, i16 5, i16 5, i16 5>
-  %cmp = icmp eq <8 x i16> %urem, <i16 0, i16 0, i16 0, i16 0,
-                                   i16 0, i16 0, i16 0, i16 0>
-  %ret = zext <8 x i1> %cmp to <8 x i16>
-  ret <8 x i16> %ret
+; CHECK-AVX1-LABEL: test_urem_even_100:
+; CHECK-AVX1:       # %bb.0:
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $5, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_urem_even_100:
+; CHECK-AVX2:       # %bb.0:
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX2-NEXT:    vpsrld $5, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [100,100,100,100]
+; CHECK-AVX2-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512VL-LABEL: test_urem_even_100:
+; CHECK-AVX512VL:       # %bb.0:
+; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vprord $2, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpminud {{.*}}(%rip){1to4}, %xmm0, %xmm1
+; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    retq
+  %urem = urem <4 x i32> %X, <i32 100, i32 100, i32 100, i32 100>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
 }
 
-; Tests BuildUREMEqFold for 4 x i32 splat vectors with even divisor.
-; The expected behavior is that the fold is _not_ applied
-; because it requires a ROTR in the even case, which has to be expanded.
-define <4 x i32> @test_urem_even_vec_i32(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE2-LABEL: test_urem_even_vec_i32:
+;------------------------------------------------------------------------------;
+; Comparison constant has undef elements.
+;------------------------------------------------------------------------------;
+
+define <4 x i32> @test_urem_odd_undef1(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_odd_undef1:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; CHECK-SSE2-NEXT:    psrld $1, %xmm1
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2454267027,2454267027,2454267027,2454267027]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm1
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm3
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,3,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; CHECK-SSE2-NEXT:    psrld $2, %xmm1
-; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [14,14,14,14]
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm1
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; CHECK-SSE2-NEXT:    pmuludq %xmm2, %xmm3
-; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
-; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; CHECK-SSE2-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; CHECK-SSE2-NEXT:    psrld $3, %xmm2
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [25,25,25,25]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; CHECK-SSE2-NEXT:    psubd %xmm2, %xmm0
 ; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
 ; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
-; CHECK-SSE41-LABEL: test_urem_even_vec_i32:
+; CHECK-SSE41-LABEL: test_urem_odd_undef1:
 ; CHECK-SSE41:       # %bb.0:
-; CHECK-SSE41-NEXT:    movdqa %xmm0, %xmm1
-; CHECK-SSE41-NEXT:    psrld $1, %xmm1
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
-; CHECK-SSE41-NEXT:    pmuludq %xmm3, %xmm2
-; CHECK-SSE41-NEXT:    pmuludq %xmm3, %xmm1
-; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
-; CHECK-SSE41-NEXT:    psrld $2, %xmm1
-; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm1
-; CHECK-SSE41-NEXT:    psubd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm1
+; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm2
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-SSE41-NEXT:    psrld $3, %xmm2
+; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm2
+; CHECK-SSE41-NEXT:    psubd %xmm2, %xmm0
 ; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
 ; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
 ; CHECK-SSE41-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE41-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_even_vec_i32:
+; CHECK-AVX1-LABEL: test_urem_odd_undef1:
 ; CHECK-AVX1:       # %bb.0:
-; CHECK-AVX1-NEXT:    vpsrld $1, %xmm0, %xmm1
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
-; CHECK-AVX1-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
-; CHECK-AVX1-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
-; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
-; CHECK-AVX1-NEXT:    vpsrld $2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $3, %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
 ; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -161,17 +211,16 @@ define <4 x i32> @test_urem_even_vec_i32(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_even_vec_i32:
+; CHECK-AVX2-LABEL: test_urem_odd_undef1:
 ; CHECK-AVX2:       # %bb.0:
-; CHECK-AVX2-NEXT:    vpsrld $1, %xmm0, %xmm1
-; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
-; CHECK-AVX2-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
-; CHECK-AVX2-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
-; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
-; CHECK-AVX2-NEXT:    vpsrld $2, %xmm1, %xmm1
-; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [14,14,14,14]
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX2-NEXT:    vpsrld $3, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [25,25,25,25]
 ; CHECK-AVX2-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
 ; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -179,72 +228,144 @@ define <4 x i32> @test_urem_even_vec_i32(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_even_vec_i32:
+; CHECK-AVX512VL-LABEL: test_urem_odd_undef1:
 ; CHECK-AVX512VL:       # %bb.0:
-; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0
-; CHECK-AVX512VL-NEXT:    vprord $1, %xmm0, %xmm0
-; CHECK-AVX512VL-NEXT:    vpminud {{.*}}(%rip){1to4}, %xmm0, %xmm1
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX512VL-NEXT:    vpsrld $3, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip){1to4}, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    retq
-  %urem = urem <4 x i32> %X, <i32 14, i32 14, i32 14, i32 14>
-  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %urem = urem <4 x i32> %X, <i32 25, i32 25, i32 25, i32 25>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 undef, i32 0>
   %ret = zext <4 x i1> %cmp to <4 x i32>
   ret <4 x i32> %ret
 }
 
-; Like test_urem_even_vec_i32, but with 8 x i16 vectors.
-define <8 x i16> @test_urem_even_vec_i16(<8 x i16> %X) nounwind readnone {
-; CHECK-SSE-LABEL: test_urem_even_vec_i16:
-; CHECK-SSE:       # %bb.0:
-; CHECK-SSE-NEXT:    movdqa %xmm0, %xmm1
-; CHECK-SSE-NEXT:    psrlw $1, %xmm1
-; CHECK-SSE-NEXT:    pmulhuw {{.*}}(%rip), %xmm1
-; CHECK-SSE-NEXT:    psrlw $1, %xmm1
-; CHECK-SSE-NEXT:    pmullw {{.*}}(%rip), %xmm1
-; CHECK-SSE-NEXT:    psubw %xmm1, %xmm0
-; CHECK-SSE-NEXT:    pxor %xmm1, %xmm1
-; CHECK-SSE-NEXT:    pcmpeqw %xmm1, %xmm0
-; CHECK-SSE-NEXT:    psrlw $15, %xmm0
-; CHECK-SSE-NEXT:    retq
+define <4 x i32> @test_urem_even_undef1(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_even_undef1:
+; CHECK-SSE2:       # %bb.0:
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; CHECK-SSE2-NEXT:    psrld $5, %xmm2
+; CHECK-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [100,100,100,100]
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm2
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; CHECK-SSE2-NEXT:    pmuludq %xmm1, %xmm3
+; CHECK-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
+; CHECK-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; CHECK-SSE2-NEXT:    psubd %xmm2, %xmm0
+; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    psrld $31, %xmm0
+; CHECK-SSE2-NEXT:    retq
+;
+; CHECK-SSE41-LABEL: test_urem_even_undef1:
+; CHECK-SSE41:       # %bb.0:
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-SSE41-NEXT:    pmuludq %xmm2, %xmm1
+; CHECK-SSE41-NEXT:    pmuludq %xmm0, %xmm2
+; CHECK-SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-SSE41-NEXT:    psrld $5, %xmm2
+; CHECK-SSE41-NEXT:    pmulld {{.*}}(%rip), %xmm2
+; CHECK-SSE41-NEXT:    psubd %xmm2, %xmm0
+; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    psrld $31, %xmm0
+; CHECK-SSE41-NEXT:    retq
+;
+; CHECK-AVX1-LABEL: test_urem_even_undef1:
+; CHECK-AVX1:       # %bb.0:
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
+; CHECK-AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
+; CHECK-AVX1-NEXT:    vpsrld $5, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX-LABEL: test_urem_even_vec_i16:
-; CHECK-AVX:       # %bb.0:
-; CHECK-AVX-NEXT:    vpsrlw $1, %xmm0, %xmm1
-; CHECK-AVX-NEXT:    vpmulhuw {{.*}}(%rip), %xmm1, %xmm1
-; CHECK-AVX-NEXT:    vpsrlw $1, %xmm1, %xmm1
-; CHECK-AVX-NEXT:    vpmullw {{.*}}(%rip), %xmm1, %xmm1
-; CHECK-AVX-NEXT:    vpsubw %xmm1, %xmm0, %xmm0
-; CHECK-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; CHECK-AVX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
-; CHECK-AVX-NEXT:    vpsrlw $15, %xmm0, %xmm0
-; CHECK-AVX-NEXT:    retq
-  %urem = urem <8 x i16> %X, <i16 14, i16 14, i16 14, i16 14,
-                              i16 14, i16 14, i16 14, i16 14>
-  %cmp = icmp eq <8 x i16> %urem, <i16 0, i16 0, i16 0, i16 0,
-                                   i16 0, i16 0, i16 0, i16 0>
-  %ret = zext <8 x i1> %cmp to <8 x i16>
-  ret <8 x i16> %ret
+; CHECK-AVX2-LABEL: test_urem_even_undef1:
+; CHECK-AVX2:       # %bb.0:
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
+; CHECK-AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX2-NEXT:    vpsrld $5, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [100,100,100,100]
+; CHECK-AVX2-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512VL-LABEL: test_urem_even_undef1:
+; CHECK-AVX512VL:       # %bb.0:
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1374389535,1374389535,1374389535,1374389535]
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmuludq %xmm2, %xmm0, %xmm2
+; CHECK-AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3]
+; CHECK-AVX512VL-NEXT:    vpsrld $5, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpmulld {{.*}}(%rip){1to4}, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    retq
+  %urem = urem <4 x i32> %X, <i32 100, i32 100, i32 100, i32 100>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 undef, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
 }
 
-; We should not proceed with this fold if the divisor is 1 or -1
-define <4 x i32> @test_urem_one_vec(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE-LABEL: test_urem_one_vec:
+;------------------------------------------------------------------------------;
+; Negative tests
+;------------------------------------------------------------------------------;
+
+; The fold is invalid if divisor is 1.
+define <4 x i32> @test_urem_one(<4 x i32> %X) nounwind {
+; CHECK-SSE-LABEL: test_urem_one:
 ; CHECK-SSE:       # %bb.0:
 ; CHECK-SSE-NEXT:    movaps {{.*#+}} xmm0 = [1,1,1,1]
 ; CHECK-SSE-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_one_vec:
+; CHECK-AVX1-LABEL: test_urem_one:
 ; CHECK-AVX1:       # %bb.0:
 ; CHECK-AVX1-NEXT:    vmovaps {{.*#+}} xmm0 = [1,1,1,1]
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_one_vec:
+; CHECK-AVX2-LABEL: test_urem_one:
 ; CHECK-AVX2:       # %bb.0:
 ; CHECK-AVX2-NEXT:    vbroadcastss {{.*#+}} xmm0 = [1,1,1,1]
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_one_vec:
+; CHECK-AVX512VL-LABEL: test_urem_one:
 ; CHECK-AVX512VL:       # %bb.0:
 ; CHECK-AVX512VL-NEXT:    vbroadcastss {{.*#+}} xmm0 = [1,1,1,1]
 ; CHECK-AVX512VL-NEXT:    retq
@@ -254,10 +375,64 @@ define <4 x i32> @test_urem_one_vec(<4 x i32> %X) nounwind readnone {
   ret <4 x i32> %ret
 }
 
-; BuildUREMEqFold does not work when the only odd factor of the divisor is 1.
-; This ensures we don't touch powers of two.
-define <4 x i32> @test_urem_pow2_vec(<4 x i32> %X) nounwind readnone {
-; CHECK-SSE-LABEL: test_urem_pow2_vec:
+; We can lower remainder of division by all-ones much better elsewhere.
+define <4 x i32> @test_urem_allones(<4 x i32> %X) nounwind {
+; CHECK-SSE2-LABEL: test_urem_allones:
+; CHECK-SSE2:       # %bb.0:
+; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE2-NEXT:    psubd %xmm0, %xmm1
+; CHECK-SSE2-NEXT:    pxor {{.*}}(%rip), %xmm1
+; CHECK-SSE2-NEXT:    pcmpgtd {{.*}}(%rip), %xmm1
+; CHECK-SSE2-NEXT:    pandn {{.*}}(%rip), %xmm1
+; CHECK-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-SSE2-NEXT:    retq
+;
+; CHECK-SSE41-LABEL: test_urem_allones:
+; CHECK-SSE41:       # %bb.0:
+; CHECK-SSE41-NEXT:    pxor %xmm1, %xmm1
+; CHECK-SSE41-NEXT:    psubd %xmm0, %xmm1
+; CHECK-SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [1,1,1,1]
+; CHECK-SSE41-NEXT:    pminud %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-SSE41-NEXT:    psrld $31, %xmm0
+; CHECK-SSE41-NEXT:    retq
+;
+; CHECK-AVX1-LABEL: test_urem_allones:
+; CHECK-AVX1:       # %bb.0:
+; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX1-NEXT:    vpsubd %xmm0, %xmm1, %xmm0
+; CHECK-AVX1-NEXT:    vpminud {{.*}}(%rip), %xmm0, %xmm1
+; CHECK-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX1-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_urem_allones:
+; CHECK-AVX2:       # %bb.0:
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpsubd %xmm0, %xmm1, %xmm0
+; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [1,1,1,1]
+; CHECK-AVX2-NEXT:    vpminud %xmm1, %xmm0, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512VL-LABEL: test_urem_allones:
+; CHECK-AVX512VL:       # %bb.0:
+; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX512VL-NEXT:    vpsubd %xmm0, %xmm1, %xmm0
+; CHECK-AVX512VL-NEXT:    vpminud {{.*}}(%rip){1to4}, %xmm0, %xmm1
+; CHECK-AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    vpsrld $31, %xmm0, %xmm0
+; CHECK-AVX512VL-NEXT:    retq
+  %urem = urem <4 x i32> %X, <i32 4294967295, i32 4294967295, i32 4294967295, i32 4294967295>
+  %cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
+  %ret = zext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %ret
+}
+
+; We can lower remainder of division by powers of two much better elsewhere.
+define <4 x i32> @test_urem_pow2(<4 x i32> %X) nounwind {
+; CHECK-SSE-LABEL: test_urem_pow2:
 ; CHECK-SSE:       # %bb.0:
 ; CHECK-SSE-NEXT:    pand {{.*}}(%rip), %xmm0
 ; CHECK-SSE-NEXT:    pxor %xmm1, %xmm1
@@ -265,7 +440,7 @@ define <4 x i32> @test_urem_pow2_vec(<4 x i32> %X) nounwind readnone {
 ; CHECK-SSE-NEXT:    psrld $31, %xmm0
 ; CHECK-SSE-NEXT:    retq
 ;
-; CHECK-AVX1-LABEL: test_urem_pow2_vec:
+; CHECK-AVX1-LABEL: test_urem_pow2:
 ; CHECK-AVX1:       # %bb.0:
 ; CHECK-AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -273,7 +448,7 @@ define <4 x i32> @test_urem_pow2_vec(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX1-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX1-NEXT:    retq
 ;
-; CHECK-AVX2-LABEL: test_urem_pow2_vec:
+; CHECK-AVX2-LABEL: test_urem_pow2:
 ; CHECK-AVX2:       # %bb.0:
 ; CHECK-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [15,15,15,15]
 ; CHECK-AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
@@ -282,7 +457,7 @@ define <4 x i32> @test_urem_pow2_vec(<4 x i32> %X) nounwind readnone {
 ; CHECK-AVX2-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; CHECK-AVX2-NEXT:    retq
 ;
-; CHECK-AVX512VL-LABEL: test_urem_pow2_vec:
+; CHECK-AVX512VL-LABEL: test_urem_pow2:
 ; CHECK-AVX512VL:       # %bb.0:
 ; CHECK-AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm0, %xmm0
 ; CHECK-AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1

diff --git a/test/CodeGen/X86/urem-seteq.ll b/test/CodeGen/X86/urem-seteq.ll
index cd1cb3715a29c2c9190118bece8d2c2f638e66be..26b9e85feb2b25a6de45d23fd766a1096741f13b 100644 (file)
--- a/test/CodeGen/X86/urem-seteq.ll
+++ b/test/CodeGen/X86/urem-seteq.ll
@@ -2,11 +2,15 @@
 ; RUN: llc -mtriple=i686-unknown-linux-gnu < %s | FileCheck %s --check-prefixes=CHECK,X86
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s --check-prefixes=CHECK,X64
 
+;------------------------------------------------------------------------------;
+; Odd divisors
+;------------------------------------------------------------------------------;
+
 ; This tests the BuildREMEqFold optimization with UREM, i32, odd divisor, SETEQ.
 ; The corresponding pseudocode is:
 ; Q <- [N * multInv(5, 2^32)] <=> [N * 0xCCCCCCCD] <=> [N * (-858993459)]
 ; res <- [Q <= (2^32 - 1) / 5] <=> [Q <= 858993459] <=> [Q < 858993460]
-define i32 @test_urem_odd(i32 %X) nounwind readnone {
+define i32 @test_urem_odd(i32 %X) nounwind {
 ; X86-LABEL: test_urem_odd:
 ; X86:       # %bb.0:
 ; X86-NEXT:    imull $-858993459, {{[0-9]+}}(%esp), %ecx # imm = 0xCCCCCCCD
@@ -28,8 +32,30 @@ define i32 @test_urem_odd(i32 %X) nounwind readnone {
   ret i32 %ret
 }
 
+define i32 @test_urem_odd_25(i32 %X) nounwind {
+; X86-LABEL: test_urem_odd_25:
+; X86:       # %bb.0:
+; X86-NEXT:    imull $-1030792151, {{[0-9]+}}(%esp), %ecx # imm = 0xC28F5C29
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    cmpl $171798692, %ecx # imm = 0xA3D70A4
+; X86-NEXT:    setb %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_urem_odd_25:
+; X64:       # %bb.0:
+; X64-NEXT:    imull $-1030792151, %edi, %ecx # imm = 0xC28F5C29
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    cmpl $171798692, %ecx # imm = 0xA3D70A4
+; X64-NEXT:    setb %al
+; X64-NEXT:    retq
+  %urem = urem i32 %X, 25
+  %cmp = icmp eq i32 %urem, 0
+  %ret = zext i1 %cmp to i32
+  ret i32 %ret
+}
+
 ; This is like test_urem_odd, except the divisor has bit 30 set.
-define i32 @test_urem_odd_bit30(i32 %X) nounwind readnone {
+define i32 @test_urem_odd_bit30(i32 %X) nounwind {
 ; X86-LABEL: test_urem_odd_bit30:
 ; X86:       # %bb.0:
 ; X86-NEXT:    imull $1789569707, {{[0-9]+}}(%esp), %ecx # imm = 0x6AAAAAAB
@@ -52,7 +78,7 @@ define i32 @test_urem_odd_bit30(i32 %X) nounwind readnone {
 }
 
 ; This is like test_urem_odd, except the divisor has bit 31 set.
-define i32 @test_urem_odd_bit31(i32 %X) nounwind readnone {
+define i32 @test_urem_odd_bit31(i32 %X) nounwind {
 ; X86-LABEL: test_urem_odd_bit31:
 ; X86:       # %bb.0:
 ; X86-NEXT:    imull $715827883, {{[0-9]+}}(%esp), %ecx # imm = 0x2AAAAAAB
@@ -74,13 +100,17 @@ define i32 @test_urem_odd_bit31(i32 %X) nounwind readnone {
   ret i32 %ret
 }
 
+;------------------------------------------------------------------------------;
+; Even divisors
+;------------------------------------------------------------------------------;
+
 ; This tests the BuildREMEqFold optimization with UREM, i16, even divisor, SETNE.
 ; In this case, D <=> 14 <=> 7 * 2^1, so D0 = 7 and K = 1.
 ; The corresponding pseudocode is:
 ; Q <- [N * multInv(D0, 2^16)] <=> [N * multInv(7, 2^16)] <=> [N * 28087]
 ; Q <- [Q >>rot K] <=> [Q >>rot 1]
 ; res <- ![Q <= (2^16 - 1) / 7] <=> ![Q <= 9362] <=> [Q > 9362]
-define i16 @test_urem_even(i16 %X) nounwind readnone {
+define i16 @test_urem_even(i16 %X) nounwind {
 ; X86-LABEL: test_urem_even:
 ; X86:       # %bb.0:
 ; X86-NEXT:    imull $28087, {{[0-9]+}}(%esp), %eax # imm = 0x6DB7
@@ -108,8 +138,32 @@ define i16 @test_urem_even(i16 %X) nounwind readnone {
   ret i16 %ret
 }
 
+define i32 @test_urem_even_100(i32 %X) nounwind {
+; X86-LABEL: test_urem_even_100:
+; X86:       # %bb.0:
+; X86-NEXT:    imull $-1030792151, {{[0-9]+}}(%esp), %ecx # imm = 0xC28F5C29
+; X86-NEXT:    rorl $2, %ecx
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    cmpl $42949673, %ecx # imm = 0x28F5C29
+; X86-NEXT:    setb %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_urem_even_100:
+; X64:       # %bb.0:
+; X64-NEXT:    imull $-1030792151, %edi, %ecx # imm = 0xC28F5C29
+; X64-NEXT:    rorl $2, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    cmpl $42949673, %ecx # imm = 0x28F5C29
+; X64-NEXT:    setb %al
+; X64-NEXT:    retq
+  %urem = urem i32 %X, 100
+  %cmp = icmp eq i32 %urem, 0
+  %ret = zext i1 %cmp to i32
+  ret i32 %ret
+}
+
 ; This is like test_urem_even, except the divisor has bit 30 set.
-define i32 @test_urem_even_bit30(i32 %X) nounwind readnone {
+define i32 @test_urem_even_bit30(i32 %X) nounwind {
 ; X86-LABEL: test_urem_even_bit30:
 ; X86:       # %bb.0:
 ; X86-NEXT:    imull $-51622203, {{[0-9]+}}(%esp), %ecx # imm = 0xFCEC4EC5
@@ -134,7 +188,7 @@ define i32 @test_urem_even_bit30(i32 %X) nounwind readnone {
 }
 
 ; This is like test_urem_odd, except the divisor has bit 31 set.
-define i32 @test_urem_even_bit31(i32 %X) nounwind readnone {
+define i32 @test_urem_even_bit31(i32 %X) nounwind {
 ; X86-LABEL: test_urem_even_bit31:
 ; X86:       # %bb.0:
 ; X86-NEXT:    imull $-1157956869, {{[0-9]+}}(%esp), %ecx # imm = 0xBAFAFAFB
@@ -158,8 +212,39 @@ define i32 @test_urem_even_bit31(i32 %X) nounwind readnone {
   ret i32 %ret
 }
 
-; We should not proceed with this fold if the divisor is 1 or -1
-define i32 @test_urem_one(i32 %X) nounwind readnone {
+;------------------------------------------------------------------------------;
+; Special case
+;------------------------------------------------------------------------------;
+
+; 'NE' predicate is fine too.
+define i32 @test_urem_odd_setne(i32 %X) nounwind {
+; X86-LABEL: test_urem_odd_setne:
+; X86:       # %bb.0:
+; X86-NEXT:    imull $-858993459, {{[0-9]+}}(%esp), %ecx # imm = 0xCCCCCCCD
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    cmpl $858993459, %ecx # imm = 0x33333333
+; X86-NEXT:    seta %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_urem_odd_setne:
+; X64:       # %bb.0:
+; X64-NEXT:    imull $-858993459, %edi, %ecx # imm = 0xCCCCCCCD
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    cmpl $858993459, %ecx # imm = 0x33333333
+; X64-NEXT:    seta %al
+; X64-NEXT:    retq
+  %urem = urem i32 %X, 5
+  %cmp = icmp ne i32 %urem, 0
+  %ret = zext i1 %cmp to i32
+  ret i32 %ret
+}
+
+;------------------------------------------------------------------------------;
+; Negative tests
+;------------------------------------------------------------------------------;
+
+; The fold is invalid if divisor is 1.
+define i32 @test_urem_one(i32 %X) nounwind {
 ; CHECK-LABEL: test_urem_one:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl $1, %eax
@@ -170,34 +255,32 @@ define i32 @test_urem_one(i32 %X) nounwind readnone {
   ret i32 %ret
 }
 
-define i32 @test_urem_100(i32 %X) nounwind readnone {
-; X86-LABEL: test_urem_100:
+; We can lower remainder of division by all-ones much better elsewhere.
+define i32 @test_urem_allones(i32 %X) nounwind {
+; X86-LABEL: test_urem_allones:
 ; X86:       # %bb.0:
-; X86-NEXT:    imull $-1030792151, {{[0-9]+}}(%esp), %ecx # imm = 0xC28F5C29
-; X86-NEXT:    rorl $2, %ecx
+; X86-NEXT:    xorl %ecx, %ecx
+; X86-NEXT:    subl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    xorl %eax, %eax
-; X86-NEXT:    cmpl $42949673, %ecx # imm = 0x28F5C29
+; X86-NEXT:    cmpl $2, %ecx
 ; X86-NEXT:    setb %al
 ; X86-NEXT:    retl
 ;
-; X64-LABEL: test_urem_100:
+; X64-LABEL: test_urem_allones:
 ; X64:       # %bb.0:
-; X64-NEXT:    imull $-1030792151, %edi, %ecx # imm = 0xC28F5C29
-; X64-NEXT:    rorl $2, %ecx
+; X64-NEXT:    negl %edi
 ; X64-NEXT:    xorl %eax, %eax
-; X64-NEXT:    cmpl $42949673, %ecx # imm = 0x28F5C29
+; X64-NEXT:    cmpl $2, %edi
 ; X64-NEXT:    setb %al
 ; X64-NEXT:    retq
-  %urem = urem i32 %X, 100
+  %urem = urem i32 %X, 4294967295
   %cmp = icmp eq i32 %urem, 0
   %ret = zext i1 %cmp to i32
   ret i32 %ret
 }
 
-; We can lower remainder of division by powers of two much better elsewhere;
-; also, BuildREMEqFold does not work when the only odd factor of the divisor is 1.
-; This ensures we don't touch powers of two.
-define i32 @test_urem_pow2(i32 %X) nounwind readnone {
+; We can lower remainder of division by powers of two much better elsewhere.
+define i32 @test_urem_pow2(i32 %X) nounwind {
 ; X86-LABEL: test_urem_pow2:
 ; X86:       # %bb.0:
 ; X86-NEXT:    xorl %eax, %eax