[llvm-branch-commits] [llvm] [SelectionDAG] Lower vector `bitinsert` and `bitextract` (PR #240586)
Pedro Lobo via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Thu Nov 12 04:46:46 PST 2026
https://github.com/pedroclobo updated https://github.com/llvm/llvm-project/pull/240586
>From 506717c93225be84dc1a1c9b6c88702723277907 Mon Sep 17 00:00:00 2001
From: Pedro Lobo <pedro.lobo at tecnico.ulisboa.pt>
Date: Fri, 6 Nov 2026 13:08:24 +0000
Subject: [PATCH 1/3] [SelectionDAG] Lower vector `bitinsert` and `bitextract`
---
.../SelectionDAG/SelectionDAGBuilder.cpp | 4 +-
.../AArch64/bitinsert-bitextract-vector.ll | 138 +++
.../AArch64/sve-bitinsert-bitextract.ll | 27 +
.../AMDGPU/bitinsert-bitextract-vector.ll | 928 ++++++++++++++++++
.../CodeGen/RISCV/rvv/bitinsert-bitextract.ll | 34 +
.../rvv/fixed-vectors-bitinsert-bitextract.ll | 240 +++++
.../X86/bitinsert-bitextract-vector.ll | 476 +++++++++
7 files changed, 1845 insertions(+), 2 deletions(-)
create mode 100644 llvm/test/CodeGen/AArch64/bitinsert-bitextract-vector.ll
create mode 100644 llvm/test/CodeGen/AArch64/sve-bitinsert-bitextract.ll
create mode 100644 llvm/test/CodeGen/AMDGPU/bitinsert-bitextract-vector.ll
create mode 100644 llvm/test/CodeGen/RISCV/rvv/bitinsert-bitextract.ll
create mode 100644 llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitinsert-bitextract.ll
create mode 100644 llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 381f46f3f04b7..76699934d578b 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -4299,7 +4299,7 @@ void SelectionDAGBuilder::visitBitInsert(const User &I) {
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
SDLoc dl = getCurSDLoc();
- assert(BaseVT.getSizeInBits() >= ValVT.getSizeInBits() &&
+ assert(BaseVT.getScalarSizeInBits() >= ValVT.getScalarSizeInBits() &&
"bitinsert val wider than base should be rejected by verifier");
// If Val is a float, cast it to an integer of the same bitwidth
@@ -4337,7 +4337,7 @@ void SelectionDAGBuilder::visitBitExtract(const User &I) {
EVT ResultVT = TLI.getValueType(DAG.getDataLayout(), I.getType());
SDLoc dl = getCurSDLoc();
- assert(ResultVT.getSizeInBits() <= SrcVT.getSizeInBits() &&
+ assert(ResultVT.getScalarSizeInBits() <= SrcVT.getScalarSizeInBits() &&
"bitextract result wider than source should be rejected by verifier");
// Legalize shift amount to the target's shift amount type.
diff --git a/llvm/test/CodeGen/AArch64/bitinsert-bitextract-vector.ll b/llvm/test/CodeGen/AArch64/bitinsert-bitextract-vector.ll
new file mode 100644
index 0000000000000..ade116ef126b6
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/bitinsert-bitextract-vector.ll
@@ -0,0 +1,138 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=aarch64 < %s | FileCheck %s
+
+define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
+; CHECK-LABEL: test_bitextract_v4i8_const:
+; CHECK: // %bb.0:
+; CHECK-NEXT: shrn v0.4h, v0.4s, #8
+; CHECK-NEXT: ret
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
+ ret <4 x i8> %result
+}
+
+define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
+; CHECK-LABEL: test_bitextract_v4i8_var:
+; CHECK: // %bb.0:
+; CHECK-NEXT: neg v1.4s, v1.4s
+; CHECK-NEXT: ushl v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: xtn v0.4h, v0.4s
+; CHECK-NEXT: ret
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
+ ret <4 x i8> %result
+}
+
+define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
+; CHECK-LABEL: test_bitextract_v2f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: adrp x8, .LCPI2_0
+; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI2_0]
+; CHECK-NEXT: ushl v0.2d, v0.2d, v1.2d
+; CHECK-NEXT: xtn v0.2s, v0.2d
+; CHECK-NEXT: ret
+ %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
+ ret <2 x float> %result
+}
+
+define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
+; CHECK-LABEL: test_bitextract_v2ptr:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ret
+ %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
+ ret <2 x ptr> %result
+}
+
+define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
+; CHECK-LABEL: test_bitextract_v16i4_var:
+; CHECK: // %bb.0:
+; CHECK-NEXT: uzp1 v3.8h, v3.8h, v4.8h
+; CHECK-NEXT: uzp1 v1.8h, v1.8h, v2.8h
+; CHECK-NEXT: uzp1 v1.16b, v1.16b, v3.16b
+; CHECK-NEXT: neg v1.16b, v1.16b
+; CHECK-NEXT: ushl v0.16b, v0.16b, v1.16b
+; CHECK-NEXT: ret
+ %result = bitextract <16 x i4>, <16 x b8> %src, <16 x i32> %off
+ ret <16 x i4> %result
+}
+
+define <4 x b32> @test_bitinsert_v4i8_const(<4 x b32> %base, <4 x i8> %val) {
+; CHECK-LABEL: test_bitinsert_v4i8_const:
+; CHECK: // %bb.0:
+; CHECK-NEXT: bic v1.4h, #255, lsl #8
+; CHECK-NEXT: bic v0.4s, #255, lsl #8
+; CHECK-NEXT: ushll v1.4s, v1.4h, #8
+; CHECK-NEXT: orr v0.16b, v0.16b, v1.16b
+; CHECK-NEXT: ret
+ %result = bitinsert <4 x b32> %base, <4 x i8> %val, <4 x i32> splat (i32 8)
+ ret <4 x b32> %result
+}
+
+define <4 x b32> @test_bitinsert_v4i8_var(<4 x b32> %base, <4 x i8> %val, <4 x i32> %off) {
+; CHECK-LABEL: test_bitinsert_v4i8_var:
+; CHECK: // %bb.0:
+; CHECK-NEXT: movi v3.2d, #0x0000ff000000ff
+; CHECK-NEXT: bic v1.4h, #255, lsl #8
+; CHECK-NEXT: ushll v1.4s, v1.4h, #0
+; CHECK-NEXT: ushl v3.4s, v3.4s, v2.4s
+; CHECK-NEXT: ushl v1.4s, v1.4s, v2.4s
+; CHECK-NEXT: bic v0.16b, v0.16b, v3.16b
+; CHECK-NEXT: orr v0.16b, v0.16b, v1.16b
+; CHECK-NEXT: ret
+ %result = bitinsert <4 x b32> %base, <4 x i8> %val, <4 x i32> %off
+ ret <4 x b32> %result
+}
+
+define <2 x b64> @test_bitinsert_v2f32(<2 x b64> %base, <2 x float> %val) {
+; CHECK-LABEL: test_bitinsert_v2f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: adrp x8, .LCPI7_0
+; CHECK-NEXT: ushll v1.2d, v1.2s, #0
+; CHECK-NEXT: adrp x9, .LCPI7_1
+; CHECK-NEXT: ldr q2, [x8, :lo12:.LCPI7_0]
+; CHECK-NEXT: ldr q3, [x9, :lo12:.LCPI7_1]
+; CHECK-NEXT: ushl v1.2d, v1.2d, v2.2d
+; CHECK-NEXT: and v0.16b, v0.16b, v3.16b
+; CHECK-NEXT: orr v0.16b, v0.16b, v1.16b
+; CHECK-NEXT: ret
+ %result = bitinsert <2 x b64> %base, <2 x float> %val, <2 x i32> <i32 0, i32 32>
+ ret <2 x b64> %result
+}
+
+define <2 x b64> @test_bitinsert_v2ptr(<2 x b64> %base, <2 x ptr> %val) {
+; CHECK-LABEL: test_bitinsert_v2ptr:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov v0.16b, v1.16b
+; CHECK-NEXT: ret
+ %result = bitinsert <2 x b64> %base, <2 x ptr> %val, <2 x i32> zeroinitializer
+ ret <2 x b64> %result
+}
+
+define <2 x b128> @test_bitinsert_v2ptr_b128(<2 x b128> %base, <2 x ptr> %val) {
+; CHECK-LABEL: test_bitinsert_v2ptr_b128:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov x8, v0.d[1]
+; CHECK-NEXT: fmov x9, d0
+; CHECK-NEXT: bfi x0, x9, #32, #32
+; CHECK-NEXT: bfxil x1, x9, #32, #32
+; CHECK-NEXT: bfi x2, x8, #32, #32
+; CHECK-NEXT: bfxil x3, x8, #32, #32
+; CHECK-NEXT: ret
+ %result = bitinsert <2 x b128> %base, <2 x ptr> %val, <2 x i32> splat (i32 32)
+ ret <2 x b128> %result
+}
+
+define <16 x b8> @test_bitinsert_v16i4_var(<16 x b8> %base, <16 x i4> %val, <16 x i32> %off) {
+; CHECK-LABEL: test_bitinsert_v16i4_var:
+; CHECK: // %bb.0:
+; CHECK-NEXT: uzp1 v4.8h, v4.8h, v5.8h
+; CHECK-NEXT: uzp1 v2.8h, v2.8h, v3.8h
+; CHECK-NEXT: movi v3.16b, #15
+; CHECK-NEXT: uzp1 v2.16b, v2.16b, v4.16b
+; CHECK-NEXT: and v1.16b, v1.16b, v3.16b
+; CHECK-NEXT: ushl v3.16b, v3.16b, v2.16b
+; CHECK-NEXT: ushl v1.16b, v1.16b, v2.16b
+; CHECK-NEXT: bic v0.16b, v0.16b, v3.16b
+; CHECK-NEXT: orr v0.16b, v0.16b, v1.16b
+; CHECK-NEXT: ret
+ %result = bitinsert <16 x b8> %base, <16 x i4> %val, <16 x i32> %off
+ ret <16 x b8> %result
+}
diff --git a/llvm/test/CodeGen/AArch64/sve-bitinsert-bitextract.ll b/llvm/test/CodeGen/AArch64/sve-bitinsert-bitextract.ll
new file mode 100644
index 0000000000000..4b3614708cbc2
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve-bitinsert-bitextract.ll
@@ -0,0 +1,27 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=aarch64 -mattr=+sve < %s | FileCheck %s
+
+define <vscale x 4 x i8> @test_bitextract_nxv4i8(<vscale x 4 x b32> %src, <vscale x 4 x i32> %off) {
+; CHECK-LABEL: test_bitextract_nxv4i8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: lsr z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT: ret
+ %result = bitextract <vscale x 4 x i8>, <vscale x 4 x b32> %src, <vscale x 4 x i32> %off
+ ret <vscale x 4 x i8> %result
+}
+
+define <vscale x 4 x b32> @test_bitinsert_nxv4i8(<vscale x 4 x b32> %base, <vscale x 4 x i8> %val, <vscale x 4 x i32> %off) {
+; CHECK-LABEL: test_bitinsert_nxv4i8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z3.s, #255 // =0xff
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: and z1.s, z1.s, #0xff
+; CHECK-NEXT: lsl z1.s, p0/m, z1.s, z2.s
+; CHECK-NEXT: lsl z3.s, p0/m, z3.s, z2.s
+; CHECK-NEXT: bic z0.d, z0.d, z3.d
+; CHECK-NEXT: orr z0.d, z0.d, z1.d
+; CHECK-NEXT: ret
+ %result = bitinsert <vscale x 4 x b32> %base, <vscale x 4 x i8> %val, <vscale x 4 x i32> %off
+ ret <vscale x 4 x b32> %result
+}
diff --git a/llvm/test/CodeGen/AMDGPU/bitinsert-bitextract-vector.ll b/llvm/test/CodeGen/AMDGPU/bitinsert-bitextract-vector.ll
new file mode 100644
index 0000000000000..9bcbddacf7a1d
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/bitinsert-bitextract-vector.ll
@@ -0,0 +1,928 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=amdgpu9.00-amd-amdhsa < %s | FileCheck -check-prefix=GFX9 %s
+; RUN: llc -mtriple=amdgpu12.00-amd-amdhsa -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-TRUE16 %s
+; RUN: llc -mtriple=amdgpu12.00-amd-amdhsa -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-FAKE16 %s
+
+define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
+; GFX9-LABEL: test_bitextract_v4i8_const:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v4, 8
+; GFX9-NEXT: v_lshrrev_b32_sdwa v3, v4, v3 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_lshrrev_b32_sdwa v1, v4, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_or_b32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX9-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX9-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX9-NEXT: v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX9-NEXT: v_lshrrev_b32_e32 v3, 24, v3
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-TRUE16-LABEL: test_bitextract_v4i8_const:
+; GFX12-TRUE16: ; %bb.0:
+; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, 8, v3
+; GFX12-TRUE16-NEXT: v_lshrrev_b16 v0.h, 8, v2.l
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX12-TRUE16-NEXT: v_lshrrev_b16 v0.l, 8, v0.l
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v2.l, 8, v3.l
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v1.l, 8, v1.l
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_or_b16 v2.l, v0.h, v2.l
+; GFX12-TRUE16-NEXT: v_or_b16 v0.l, v0.l, v1.l
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.h, v2.l
+; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, 24, v3
+; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_bitextract_v4i8_const:
+; GFX12-FAKE16: ; %bb.0:
+; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 8, v3
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX12-FAKE16-NEXT: v_lshrrev_b16 v2, 8, v2
+; GFX12-FAKE16-NEXT: v_lshrrev_b16 v0, 8, v0
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v3, 8, v3
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v1, 8, v1
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v2, v2, v3
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v1
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v2
+; GFX12-FAKE16-NEXT: v_lshl_or_b32 v3, v2, 16, v3
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 8, v3
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 24, v4
+; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
+ ret <4 x i8> %result
+}
+
+define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
+; GFX9-LABEL: test_bitextract_v4i8_var:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_lshrrev_b32_sdwa v3, v7, v3 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_lshrrev_b32_e32 v2, v6, v2
+; GFX9-NEXT: v_lshrrev_b32_sdwa v1, v5, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_lshrrev_b32_e32 v0, v4, v0
+; GFX9-NEXT: v_or_b32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX9-NEXT: v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX9-NEXT: v_lshrrev_b32_e32 v3, 24, v3
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-TRUE16-LABEL: test_bitextract_v4i8_var:
+; GFX12-TRUE16: ; %bb.0:
+; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v7, v3
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v6, v6, v2
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, v5, v1
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v0, v4, v0
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v2.l, 8, v3.l
+; GFX12-TRUE16-NEXT: v_and_b16 v2.h, v6.l, 0xff
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v1.l, 8, v1.l
+; GFX12-TRUE16-NEXT: v_and_b16 v0.l, v0.l, 0xff
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_or_b16 v2.l, v2.h, v2.l
+; GFX12-TRUE16-NEXT: v_or_b16 v0.l, v0.l, v1.l
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.h, v2.l
+; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, 24, v3
+; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_bitextract_v4i8_var:
+; GFX12-FAKE16: ; %bb.0:
+; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v7, v3
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v2, v6, v2
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, v5, v1
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v0, v4, v0
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v3, 8, v3
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v2, 0xff, v2
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v1, 8, v1
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 0xff, v0
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v2, v2, v3
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v1
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v2
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX12-FAKE16-NEXT: v_lshl_or_b32 v3, v2, 16, v3
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 8, v3
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3)
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 24, v4
+; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
+ ret <4 x i8> %result
+}
+
+define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
+; GFX9-LABEL: test_bitextract_v2f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v1, v3
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: test_bitextract_v2f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mov_b32_e32 v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
+ ret <2 x float> %result
+}
+
+define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
+; GFX9-LABEL: test_bitextract_v2ptr:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: test_bitextract_v2ptr:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
+ ret <2 x ptr> %result
+}
+
+define <4 x b32> @test_bitinsert_v4i8_const(<4 x b32> %base, <4 x i8> %val) {
+; GFX9-LABEL: test_bitinsert_v4i8_const:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v8, 8
+; GFX9-NEXT: v_lshlrev_b32_sdwa v4, v8, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT: s_mov_b32 s4, 0xffff00ff
+; GFX9-NEXT: v_and_or_b32 v0, v0, s4, v4
+; GFX9-NEXT: v_lshlrev_b32_sdwa v4, v8, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT: v_and_or_b32 v1, v1, s4, v4
+; GFX9-NEXT: v_lshlrev_b32_sdwa v4, v8, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT: v_and_or_b32 v2, v2, s4, v4
+; GFX9-NEXT: v_lshlrev_b32_sdwa v4, v8, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT: v_and_or_b32 v3, v3, s4, v4
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: test_bitinsert_v4i8_const:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_and_b32_e32 v4, 0xff, v4
+; GFX12-NEXT: v_and_b32_e32 v5, 0xff, v5
+; GFX12-NEXT: v_and_b32_e32 v6, 0xff, v6
+; GFX12-NEXT: v_and_b32_e32 v7, 0xff, v7
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-NEXT: v_lshlrev_b32_e32 v4, 8, v4
+; GFX12-NEXT: v_lshlrev_b32_e32 v5, 8, v5
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-NEXT: v_lshlrev_b32_e32 v6, 8, v6
+; GFX12-NEXT: v_lshlrev_b32_e32 v7, 8, v7
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-NEXT: v_and_or_b32 v0, v0, 0xffff00ff, v4
+; GFX12-NEXT: v_and_or_b32 v1, v1, 0xffff00ff, v5
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-NEXT: v_and_or_b32 v2, v2, 0xffff00ff, v6
+; GFX12-NEXT: v_and_or_b32 v3, v3, 0xffff00ff, v7
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %result = bitinsert <4 x b32> %base, <4 x i8> %val, <4 x i32> splat (i32 8)
+ ret <4 x b32> %result
+}
+
+define <4 x b32> @test_bitinsert_v4i8_var(<4 x b32> %base, <4 x i8> %val, <4 x i32> %off) {
+; GFX9-LABEL: test_bitinsert_v4i8_var:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_movk_i32 s4, 0xff
+; GFX9-NEXT: v_lshlrev_b32_sdwa v4, v8, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT: v_lshlrev_b32_e64 v8, v8, s4
+; GFX9-NEXT: v_not_b32_e32 v8, v8
+; GFX9-NEXT: v_and_or_b32 v0, v0, v8, v4
+; GFX9-NEXT: v_lshlrev_b32_sdwa v4, v9, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT: v_lshlrev_b32_e64 v5, v9, s4
+; GFX9-NEXT: v_not_b32_e32 v5, v5
+; GFX9-NEXT: v_and_or_b32 v1, v1, v5, v4
+; GFX9-NEXT: v_lshlrev_b32_e64 v5, v10, s4
+; GFX9-NEXT: v_lshlrev_b32_sdwa v4, v10, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT: v_not_b32_e32 v5, v5
+; GFX9-NEXT: v_and_or_b32 v2, v2, v5, v4
+; GFX9-NEXT: v_lshlrev_b32_e64 v5, v11, s4
+; GFX9-NEXT: v_lshlrev_b32_sdwa v4, v11, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT: v_not_b32_e32 v5, v5
+; GFX9-NEXT: v_and_or_b32 v3, v3, v5, v4
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: test_bitinsert_v4i8_var:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_and_b32_e32 v4, 0xff, v4
+; GFX12-NEXT: v_lshlrev_b32_e64 v12, v8, 0xff
+; GFX12-NEXT: v_and_b32_e32 v5, 0xff, v5
+; GFX12-NEXT: v_and_b32_e32 v6, 0xff, v6
+; GFX12-NEXT: v_and_b32_e32 v7, 0xff, v7
+; GFX12-NEXT: v_lshlrev_b32_e32 v4, v8, v4
+; GFX12-NEXT: v_not_b32_e32 v8, v12
+; GFX12-NEXT: v_lshlrev_b32_e32 v5, v9, v5
+; GFX12-NEXT: v_lshlrev_b32_e64 v9, v9, 0xff
+; GFX12-NEXT: v_lshlrev_b32_e64 v12, v10, 0xff
+; GFX12-NEXT: v_lshlrev_b32_e64 v13, v11, 0xff
+; GFX12-NEXT: v_lshlrev_b32_e32 v6, v10, v6
+; GFX12-NEXT: v_lshlrev_b32_e32 v7, v11, v7
+; GFX12-NEXT: v_not_b32_e32 v9, v9
+; GFX12-NEXT: v_not_b32_e32 v10, v12
+; GFX12-NEXT: v_not_b32_e32 v11, v13
+; GFX12-NEXT: v_and_or_b32 v0, v0, v8, v4
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-NEXT: v_and_or_b32 v1, v1, v9, v5
+; GFX12-NEXT: v_and_or_b32 v2, v2, v10, v6
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_4)
+; GFX12-NEXT: v_and_or_b32 v3, v3, v11, v7
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %result = bitinsert <4 x b32> %base, <4 x i8> %val, <4 x i32> %off
+ ret <4 x b32> %result
+}
+
+define <2 x b64> @test_bitinsert_v2f32(<2 x b64> %base, <2 x float> %val) {
+; GFX9-LABEL: test_bitinsert_v2f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v3, v5
+; GFX9-NEXT: v_mov_b32_e32 v0, v4
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: test_bitinsert_v2f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_mov_b32 v3, v5 :: v_dual_mov_b32 v0, v4
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %result = bitinsert <2 x b64> %base, <2 x float> %val, <2 x i32> <i32 0, i32 32>
+ ret <2 x b64> %result
+}
+
+define <2 x b64> @test_bitinsert_v2ptr(<2 x b64> %base, <2 x ptr> %val) {
+; GFX9-LABEL: test_bitinsert_v2ptr:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v3, v7
+; GFX9-NEXT: v_mov_b32_e32 v2, v6
+; GFX9-NEXT: v_mov_b32_e32 v1, v5
+; GFX9-NEXT: v_mov_b32_e32 v0, v4
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: test_bitinsert_v2ptr:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_mov_b32 v3, v7 :: v_dual_mov_b32 v2, v6
+; GFX12-NEXT: v_dual_mov_b32 v1, v5 :: v_dual_mov_b32 v0, v4
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %result = bitinsert <2 x b64> %base, <2 x ptr> %val, <2 x i32> zeroinitializer
+ ret <2 x b64> %result
+}
+
+define <2 x b128> @test_bitinsert_v2ptr_b128(<2 x b128> %base, <2 x ptr> %val) {
+; GFX9-LABEL: test_bitinsert_v2ptr_b128:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v6, v11
+; GFX9-NEXT: v_mov_b32_e32 v5, v10
+; GFX9-NEXT: v_mov_b32_e32 v2, v9
+; GFX9-NEXT: v_mov_b32_e32 v1, v8
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: test_bitinsert_v2ptr_b128:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_mov_b32 v6, v11 :: v_dual_mov_b32 v5, v10
+; GFX12-NEXT: v_dual_mov_b32 v2, v9 :: v_dual_mov_b32 v1, v8
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %result = bitinsert <2 x b128> %base, <2 x ptr> %val, <2 x i32> splat (i32 32)
+ ret <2 x b128> %result
+}
+
+define <2 x b128> @test_bitinsert_v2ptr_addrspace3_const(<2 x b128> %base, <2 x ptr addrspace(3)> %val) {
+; GFX9-LABEL: test_bitinsert_v2ptr_addrspace3_const:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v10, v8
+; GFX9-NEXT: v_mov_b32_e32 v11, 0
+; GFX9-NEXT: v_lshlrev_b64 v[12:13], 16, v[10:11]
+; GFX9-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-NEXT: v_lshlrev_b64 v[8:9], 16, v[9:10]
+; GFX9-NEXT: v_and_b32_e32 v1, 0xffff0000, v1
+; GFX9-NEXT: v_and_b32_e32 v5, 0xffff0000, v5
+; GFX9-NEXT: v_or_b32_e32 v1, v1, v13
+; GFX9-NEXT: v_or_b32_sdwa v0, v0, v12 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX9-NEXT: v_or_b32_e32 v5, v5, v9
+; GFX9-NEXT: v_or_b32_sdwa v4, v4, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: test_bitinsert_v2ptr_addrspace3_const:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_mov_b32 v11, v8 :: v_dual_mov_b32 v12, 0
+; GFX12-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX12-NEXT: v_and_b32_e32 v5, 0xffff0000, v5
+; GFX12-NEXT: v_and_b32_e32 v4, 0xffff, v4
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX12-NEXT: v_dual_mov_b32 v10, v12 :: v_dual_and_b32 v1, 0xffff0000, v1
+; GFX12-NEXT: v_lshlrev_b64_e32 v[11:12], 16, v[11:12]
+; GFX12-NEXT: v_lshlrev_b64_e32 v[8:9], 16, v[9:10]
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-NEXT: v_or_b32_e32 v0, v0, v11
+; GFX12-NEXT: v_or_b32_e32 v1, v1, v12
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-NEXT: v_or_b32_e32 v5, v5, v9
+; GFX12-NEXT: v_or_b32_e32 v4, v4, v8
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %result = bitinsert <2 x b128> %base, <2 x ptr addrspace(3)> %val, <2 x i32> splat (i32 16)
+ ret <2 x b128> %result
+}
+
+define <16 x b8> @test_bitinsert_v16i4_var(<16 x b8> %base, <16 x i4> %val, <16 x i32> %off) {
+; GFX9-LABEL: test_bitinsert_v16i4_var:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: buffer_load_ushort v31, off, s[0:3], s32
+; GFX9-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:64
+; GFX9-NEXT: v_and_b32_e32 v30, 15, v30
+; GFX9-NEXT: v_and_b32_e32 v29, 15, v29
+; GFX9-NEXT: v_and_b32_e32 v28, 15, v28
+; GFX9-NEXT: v_and_b32_e32 v27, 15, v27
+; GFX9-NEXT: v_and_b32_e32 v26, 15, v26
+; GFX9-NEXT: v_and_b32_e32 v23, 15, v23
+; GFX9-NEXT: s_waitcnt vmcnt(1)
+; GFX9-NEXT: v_and_b32_e32 v31, 15, v31
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_lshlrev_b16_e32 v31, v32, v31
+; GFX9-NEXT: v_lshlrev_b16_e64 v32, v32, 15
+; GFX9-NEXT: v_xor_b32_e32 v32, -1, v32
+; GFX9-NEXT: v_and_b32_e32 v15, v15, v32
+; GFX9-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:60
+; GFX9-NEXT: v_or_b32_sdwa v15, v15, v31 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_lshlrev_b16_e32 v30, v32, v30
+; GFX9-NEXT: v_lshlrev_b16_e64 v32, v32, 15
+; GFX9-NEXT: v_xor_b32_e32 v32, -1, v32
+; GFX9-NEXT: v_and_b32_e32 v14, v14, v32
+; GFX9-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:56
+; GFX9-NEXT: v_or_b32_e32 v14, v14, v30
+; GFX9-NEXT: v_or_b32_sdwa v14, v14, v15 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: buffer_load_dword v15, off, s[0:3], s32 offset:36
+; GFX9-NEXT: s_waitcnt vmcnt(1)
+; GFX9-NEXT: v_lshlrev_b16_e32 v29, v32, v29
+; GFX9-NEXT: v_lshlrev_b16_e64 v32, v32, 15
+; GFX9-NEXT: v_xor_b32_e32 v32, -1, v32
+; GFX9-NEXT: v_and_b32_e32 v13, v13, v32
+; GFX9-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:52
+; GFX9-NEXT: v_or_b32_sdwa v13, v13, v29 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_lshlrev_b16_e32 v28, v32, v28
+; GFX9-NEXT: v_lshlrev_b16_e64 v32, v32, 15
+; GFX9-NEXT: v_xor_b32_e32 v32, -1, v32
+; GFX9-NEXT: v_and_b32_e32 v12, v12, v32
+; GFX9-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:48
+; GFX9-NEXT: v_or_b32_e32 v12, v12, v28
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_lshlrev_b16_e32 v27, v32, v27
+; GFX9-NEXT: v_lshlrev_b16_e64 v32, v32, 15
+; GFX9-NEXT: v_xor_b32_e32 v32, -1, v32
+; GFX9-NEXT: v_and_b32_e32 v11, v11, v32
+; GFX9-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:44
+; GFX9-NEXT: v_or_b32_sdwa v11, v11, v27 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_lshlrev_b16_e32 v26, v32, v26
+; GFX9-NEXT: v_lshlrev_b16_e64 v32, v32, 15
+; GFX9-NEXT: v_xor_b32_e32 v32, -1, v32
+; GFX9-NEXT: v_and_b32_e32 v10, v10, v32
+; GFX9-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:32
+; GFX9-NEXT: v_or_b32_e32 v10, v10, v26
+; GFX9-NEXT: v_or_b32_sdwa v10, v10, v11 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: v_and_b32_e32 v11, 15, v22
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_lshlrev_b16_e32 v23, v32, v23
+; GFX9-NEXT: v_lshlrev_b16_e64 v32, v32, 15
+; GFX9-NEXT: v_xor_b32_e32 v32, -1, v32
+; GFX9-NEXT: v_and_b32_e32 v7, v7, v32
+; GFX9-NEXT: v_or_b32_sdwa v7, v7, v23 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_or_b32_sdwa v23, v12, v13 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: buffer_load_dword v12, off, s[0:3], s32 offset:28
+; GFX9-NEXT: buffer_load_dword v13, off, s[0:3], s32 offset:40
+; GFX9-NEXT: s_waitcnt vmcnt(1)
+; GFX9-NEXT: v_lshlrev_b16_e32 v11, v12, v11
+; GFX9-NEXT: v_lshlrev_b16_e64 v12, v12, 15
+; GFX9-NEXT: v_xor_b32_e32 v12, -1, v12
+; GFX9-NEXT: v_and_b32_e32 v6, v6, v12
+; GFX9-NEXT: v_or_b32_e32 v6, v6, v11
+; GFX9-NEXT: v_or_b32_sdwa v6, v6, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:24
+; GFX9-NEXT: v_and_b32_e32 v11, 15, v21
+; GFX9-NEXT: v_and_b32_e32 v12, 15, v20
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_lshlrev_b16_e32 v11, v7, v11
+; GFX9-NEXT: v_lshlrev_b16_e64 v7, v7, 15
+; GFX9-NEXT: v_xor_b32_e32 v7, -1, v7
+; GFX9-NEXT: v_and_b32_e32 v5, v5, v7
+; GFX9-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:20
+; GFX9-NEXT: v_or_b32_sdwa v5, v5, v11 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: buffer_load_dword v11, off, s[0:3], s32 offset:16
+; GFX9-NEXT: s_waitcnt vmcnt(1)
+; GFX9-NEXT: v_lshlrev_b16_e32 v12, v7, v12
+; GFX9-NEXT: v_lshlrev_b16_e64 v7, v7, 15
+; GFX9-NEXT: v_xor_b32_e32 v7, -1, v7
+; GFX9-NEXT: v_and_b32_e32 v4, v4, v7
+; GFX9-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:12
+; GFX9-NEXT: v_or_b32_e32 v4, v4, v12
+; GFX9-NEXT: buffer_load_dword v12, off, s[0:3], s32 offset:8
+; GFX9-NEXT: v_or_b32_sdwa v20, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:4
+; GFX9-NEXT: v_and_b32_e32 v5, 15, v19
+; GFX9-NEXT: s_waitcnt vmcnt(3)
+; GFX9-NEXT: v_lshlrev_b16_e32 v5, v11, v5
+; GFX9-NEXT: v_lshlrev_b16_e64 v11, v11, 15
+; GFX9-NEXT: v_xor_b32_e32 v11, -1, v11
+; GFX9-NEXT: v_and_b32_e32 v3, v3, v11
+; GFX9-NEXT: v_or_b32_sdwa v3, v3, v5 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_and_b32_e32 v5, 15, v18
+; GFX9-NEXT: v_lshlrev_b32_e32 v18, 16, v14
+; GFX9-NEXT: v_lshlrev_b32_e32 v11, 16, v10
+; GFX9-NEXT: s_waitcnt vmcnt(2)
+; GFX9-NEXT: v_lshlrev_b16_e32 v5, v7, v5
+; GFX9-NEXT: v_lshlrev_b16_e64 v7, v7, 15
+; GFX9-NEXT: v_xor_b32_e32 v7, -1, v7
+; GFX9-NEXT: v_and_b32_e32 v2, v2, v7
+; GFX9-NEXT: v_or_b32_e32 v2, v2, v5
+; GFX9-NEXT: s_waitcnt vmcnt(1)
+; GFX9-NEXT: v_lshlrev_b16_e64 v5, v12, 15
+; GFX9-NEXT: v_or_b32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: v_and_b32_e32 v3, 15, v17
+; GFX9-NEXT: v_xor_b32_e32 v5, -1, v5
+; GFX9-NEXT: v_lshlrev_b16_e32 v3, v12, v3
+; GFX9-NEXT: v_and_b32_e32 v1, v1, v5
+; GFX9-NEXT: v_or_b32_sdwa v1, v1, v3 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_and_b32_e32 v3, 15, v16
+; GFX9-NEXT: v_and_b32_e32 v5, 15, v25
+; GFX9-NEXT: v_and_b32_e32 v16, 15, v24
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_lshlrev_b16_e32 v3, v4, v3
+; GFX9-NEXT: v_lshlrev_b16_e64 v4, v4, 15
+; GFX9-NEXT: v_lshlrev_b16_e32 v5, v13, v5
+; GFX9-NEXT: v_lshlrev_b16_e64 v13, v13, 15
+; GFX9-NEXT: v_lshlrev_b16_e32 v16, v15, v16
+; GFX9-NEXT: v_lshlrev_b16_e64 v15, v15, 15
+; GFX9-NEXT: v_xor_b32_e32 v4, -1, v4
+; GFX9-NEXT: v_xor_b32_e32 v13, -1, v13
+; GFX9-NEXT: v_xor_b32_e32 v15, -1, v15
+; GFX9-NEXT: v_lshlrev_b32_e32 v7, 16, v6
+; GFX9-NEXT: v_and_b32_e32 v0, v0, v4
+; GFX9-NEXT: v_and_b32_e32 v9, v9, v13
+; GFX9-NEXT: v_and_b32_e32 v8, v8, v15
+; GFX9-NEXT: v_or_b32_sdwa v12, v18, v23 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT: v_or_b32_e32 v0, v0, v3
+; GFX9-NEXT: v_or_b32_sdwa v4, v7, v20 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX9-NEXT: v_or_b32_sdwa v9, v9, v5 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_or_b32_e32 v8, v8, v16
+; GFX9-NEXT: v_lshrrev_b32_e32 v13, 8, v12
+; GFX9-NEXT: v_lshrrev_b32_e32 v5, 8, v4
+; GFX9-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: v_or_b32_sdwa v8, v8, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: v_lshl_or_b32 v9, v10, 16, v9
+; GFX9-NEXT: v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT: v_lshrrev_b64 v[3:4], 24, v[3:4]
+; GFX9-NEXT: v_lshrrev_b64 v[11:12], 24, v[11:12]
+; GFX9-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX9-NEXT: v_lshrrev_b32_e32 v7, 24, v7
+; GFX9-NEXT: v_lshrrev_b32_e32 v9, 8, v9
+; GFX9-NEXT: v_lshrrev_b32_e32 v15, 24, v18
+; GFX9-NEXT: v_mov_b32_e32 v4, v20
+; GFX9-NEXT: v_mov_b32_e32 v12, v23
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-TRUE16-LABEL: test_bitinsert_v16i4_var:
+; GFX12-TRUE16: ; %bb.0:
+; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT: s_clause 0x10
+; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v31, off, s32
+; GFX12-TRUE16-NEXT: scratch_load_b32 v32, off, s32 offset:64
+; GFX12-TRUE16-NEXT: scratch_load_b32 v33, off, s32 offset:60
+; GFX12-TRUE16-NEXT: scratch_load_b32 v34, off, s32 offset:56
+; GFX12-TRUE16-NEXT: scratch_load_b32 v35, off, s32 offset:52
+; GFX12-TRUE16-NEXT: scratch_load_b32 v36, off, s32 offset:48
+; GFX12-TRUE16-NEXT: scratch_load_b32 v37, off, s32 offset:44
+; GFX12-TRUE16-NEXT: scratch_load_b32 v38, off, s32 offset:32
+; GFX12-TRUE16-NEXT: scratch_load_b32 v39, off, s32 offset:28
+; GFX12-TRUE16-NEXT: scratch_load_b32 v48, off, s32 offset:24
+; GFX12-TRUE16-NEXT: scratch_load_b32 v49, off, s32 offset:20
+; GFX12-TRUE16-NEXT: scratch_load_b32 v50, off, s32 offset:16
+; GFX12-TRUE16-NEXT: scratch_load_b32 v51, off, s32 offset:12
+; GFX12-TRUE16-NEXT: scratch_load_b32 v52, off, s32 offset:8
+; GFX12-TRUE16-NEXT: scratch_load_b32 v53, off, s32 offset:4
+; GFX12-TRUE16-NEXT: scratch_load_b32 v54, off, s32 offset:40
+; GFX12-TRUE16-NEXT: scratch_load_b32 v55, off, s32 offset:36
+; GFX12-TRUE16-NEXT: v_and_b16 v10.h, v18.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v9.h, v19.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v12.h, v16.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v3.h, v27.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v4.h, v26.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v7.h, v21.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v11.h, v17.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v0.h, v30.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v5.h, v23.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v6.h, v22.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v8.h, v20.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v1.h, v29.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v2.h, v28.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v13.h, v25.l, 15
+; GFX12-TRUE16-NEXT: v_and_b16 v14.h, v24.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x10
+; GFX12-TRUE16-NEXT: v_and_b16 v15.h, v31.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0xf
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v16.l, v32.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0xe
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v16.h, v33.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0xd
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v17.l, v34.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0xc
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v17.h, v35.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0xb
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v18.l, v36.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0xa
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v18.h, v37.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x9
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v19.l, v38.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x8
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v19.h, v39.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x7
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v20.l, v48.l, 15
+; GFX12-TRUE16-NEXT: v_xor_b16 v18.l, v18.l, -1
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x5
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v21.l, v50.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x4
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v21.h, v51.l, 15
+; GFX12-TRUE16-NEXT: v_xor_b16 v18.h, v18.h, -1
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x3
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v22.l, v52.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x1
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v23.l, v54.l, 15
+; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v23.h, v55.l, 15
+; GFX12-TRUE16-NEXT: v_xor_b16 v16.l, v16.l, -1
+; GFX12-TRUE16-NEXT: v_xor_b16 v16.h, v16.h, -1
+; GFX12-TRUE16-NEXT: v_xor_b16 v19.l, v19.l, -1
+; GFX12-TRUE16-NEXT: v_xor_b16 v19.h, v19.h, -1
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v3.h, v36.l, v3.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v4.h, v37.l, v4.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v20.h, v49.l, 15
+; GFX12-TRUE16-NEXT: v_xor_b16 v17.l, v17.l, -1
+; GFX12-TRUE16-NEXT: v_xor_b16 v17.h, v17.h, -1
+; GFX12-TRUE16-NEXT: v_xor_b16 v21.l, v21.l, -1
+; GFX12-TRUE16-NEXT: v_xor_b16 v21.h, v21.h, -1
+; GFX12-TRUE16-NEXT: v_and_b16 v11.l, v11.l, v18.l
+; GFX12-TRUE16-NEXT: v_and_b16 v10.l, v10.l, v18.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v0.h, v33.l, v0.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v5.h, v38.l, v5.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v6.h, v39.l, v6.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v22.h, v53.l, 15
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v15.h, v32.l, v15.h
+; GFX12-TRUE16-NEXT: v_xor_b16 v20.l, v20.l, -1
+; GFX12-TRUE16-NEXT: v_xor_b16 v22.l, v22.l, -1
+; GFX12-TRUE16-NEXT: v_xor_b16 v23.l, v23.l, -1
+; GFX12-TRUE16-NEXT: v_xor_b16 v23.h, v23.h, -1
+; GFX12-TRUE16-NEXT: v_and_b16 v15.l, v15.l, v16.l
+; GFX12-TRUE16-NEXT: v_and_b16 v14.l, v14.l, v16.h
+; GFX12-TRUE16-NEXT: v_and_b16 v7.l, v7.l, v19.l
+; GFX12-TRUE16-NEXT: v_and_b16 v6.l, v6.l, v19.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v1.h, v34.l, v1.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v2.h, v35.l, v2.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v9.h, v50.l, v9.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v10.h, v51.l, v10.h
+; GFX12-TRUE16-NEXT: v_xor_b16 v20.h, v20.h, -1
+; GFX12-TRUE16-NEXT: v_and_b16 v13.l, v13.l, v17.l
+; GFX12-TRUE16-NEXT: v_and_b16 v12.l, v12.l, v17.h
+; GFX12-TRUE16-NEXT: v_and_b16 v3.l, v3.l, v21.l
+; GFX12-TRUE16-NEXT: v_and_b16 v2.l, v2.l, v21.h
+; GFX12-TRUE16-NEXT: v_or_b16 v3.h, v11.l, v3.h
+; GFX12-TRUE16-NEXT: v_or_b16 v4.h, v10.l, v4.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v7.h, v48.l, v7.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v11.h, v52.l, v11.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v13.h, v54.l, v13.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v14.h, v55.l, v14.h
+; GFX12-TRUE16-NEXT: v_xor_b16 v22.h, v22.h, -1
+; GFX12-TRUE16-NEXT: v_and_b16 v5.l, v5.l, v20.l
+; GFX12-TRUE16-NEXT: v_and_b16 v1.l, v1.l, v22.l
+; GFX12-TRUE16-NEXT: v_and_b16 v9.l, v9.l, v23.l
+; GFX12-TRUE16-NEXT: v_and_b16 v8.l, v8.l, v23.h
+; GFX12-TRUE16-NEXT: v_or_b16 v15.l, v15.l, v15.h
+; GFX12-TRUE16-NEXT: v_or_b16 v0.h, v14.l, v0.h
+; GFX12-TRUE16-NEXT: v_or_b16 v5.h, v7.l, v5.h
+; GFX12-TRUE16-NEXT: v_or_b16 v6.l, v6.l, v6.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v8.h, v49.l, v8.h
+; GFX12-TRUE16-NEXT: v_and_b16 v4.l, v4.l, v20.h
+; GFX12-TRUE16-NEXT: v_or_b16 v1.h, v13.l, v1.h
+; GFX12-TRUE16-NEXT: v_or_b16 v2.h, v12.l, v2.h
+; GFX12-TRUE16-NEXT: v_or_b16 v3.l, v3.l, v9.h
+; GFX12-TRUE16-NEXT: v_or_b16 v2.l, v2.l, v10.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v3.h, 8, v3.h
+; GFX12-TRUE16-NEXT: v_and_b16 v4.h, v4.h, 0xff
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v12.h, v53.l, v12.h
+; GFX12-TRUE16-NEXT: v_and_b16 v0.l, v0.l, v22.h
+; GFX12-TRUE16-NEXT: v_or_b16 v5.l, v5.l, v7.h
+; GFX12-TRUE16-NEXT: v_or_b16 v1.l, v1.l, v11.h
+; GFX12-TRUE16-NEXT: v_or_b16 v6.h, v9.l, v13.h
+; GFX12-TRUE16-NEXT: v_or_b16 v7.l, v8.l, v14.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v7.h, 8, v15.l
+; GFX12-TRUE16-NEXT: v_and_b16 v0.h, v0.h, 0xff
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v5.h, 8, v5.h
+; GFX12-TRUE16-NEXT: v_and_b16 v6.l, v6.l, 0xff
+; GFX12-TRUE16-NEXT: v_or_b16 v4.l, v4.l, v8.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v1.h, 8, v1.h
+; GFX12-TRUE16-NEXT: v_and_b16 v2.h, v2.h, 0xff
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v3.l, 8, v3.l
+; GFX12-TRUE16-NEXT: v_and_b16 v2.l, v2.l, 0xff
+; GFX12-TRUE16-NEXT: v_or_b16 v10.l, v4.h, v3.h
+; GFX12-TRUE16-NEXT: v_or_b16 v0.l, v0.l, v12.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v9.l, 8, v1.l
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v14.l, 8, v6.h
+; GFX12-TRUE16-NEXT: v_and_b16 v1.l, v7.l, 0xff
+; GFX12-TRUE16-NEXT: v_or_b16 v16.h, v0.h, v7.h
+; GFX12-TRUE16-NEXT: v_or_b16 v12.h, v6.l, v5.h
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v5.l, 8, v5.l
+; GFX12-TRUE16-NEXT: v_and_b16 v4.l, v4.l, 0xff
+; GFX12-TRUE16-NEXT: v_or_b16 v16.l, v2.h, v1.h
+; GFX12-TRUE16-NEXT: v_or_b16 v2.l, v2.l, v3.l
+; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v15, 16, v10
+; GFX12-TRUE16-NEXT: v_or_b16 v8.l, v1.l, v14.l
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.l, v16.h
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, v12.h
+; GFX12-TRUE16-NEXT: v_or_b16 v12.l, v4.l, v5.l
+; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v11, 16, v2
+; GFX12-TRUE16-NEXT: v_lshrrev_b64 v[17:18], 24, v[15:16]
+; GFX12-TRUE16-NEXT: v_and_b16 v0.l, v0.l, 0xff
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v14.h, v10.l
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v9.h, v2.l
+; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v6, 16, v1
+; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v7, 16, v3
+; GFX12-TRUE16-NEXT: v_lshrrev_b64 v[3:4], 24, v[11:12]
+; GFX12-TRUE16-NEXT: v_or_b16 v0.l, v0.l, v9.l
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v13, 8, v16
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v5, 8, v12
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v9
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v9, 8, v14
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v7, 24, v7
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v15, 24, v6
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.l, v12.l
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v6.l, v12.h
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v12.l, v16.l
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v14.l, v16.h
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v11.l, v17.l
+; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_bitinsert_v16i4_var:
+; GFX12-FAKE16: ; %bb.0:
+; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT: s_clause 0x10
+; GFX12-FAKE16-NEXT: scratch_load_u16 v31, off, s32
+; GFX12-FAKE16-NEXT: scratch_load_b32 v32, off, s32 offset:64
+; GFX12-FAKE16-NEXT: scratch_load_b32 v33, off, s32 offset:60
+; GFX12-FAKE16-NEXT: scratch_load_b32 v34, off, s32 offset:56
+; GFX12-FAKE16-NEXT: scratch_load_b32 v35, off, s32 offset:52
+; GFX12-FAKE16-NEXT: scratch_load_b32 v36, off, s32 offset:48
+; GFX12-FAKE16-NEXT: scratch_load_b32 v37, off, s32 offset:44
+; GFX12-FAKE16-NEXT: scratch_load_b32 v38, off, s32 offset:32
+; GFX12-FAKE16-NEXT: scratch_load_b32 v39, off, s32 offset:28
+; GFX12-FAKE16-NEXT: scratch_load_b32 v48, off, s32 offset:24
+; GFX12-FAKE16-NEXT: scratch_load_b32 v49, off, s32 offset:20
+; GFX12-FAKE16-NEXT: scratch_load_b32 v50, off, s32 offset:16
+; GFX12-FAKE16-NEXT: scratch_load_b32 v51, off, s32 offset:12
+; GFX12-FAKE16-NEXT: scratch_load_b32 v52, off, s32 offset:8
+; GFX12-FAKE16-NEXT: scratch_load_b32 v53, off, s32 offset:4
+; GFX12-FAKE16-NEXT: scratch_load_b32 v54, off, s32 offset:40
+; GFX12-FAKE16-NEXT: scratch_load_b32 v55, off, s32 offset:36
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v21, 15, v21
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v20, 15, v20
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v17, 15, v17
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v28, 15, v28
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v16, 15, v16
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v30, 15, v30
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v26, 15, v26
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v23, 15, v23
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v22, 15, v22
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v19, 15, v19
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v18, 15, v18
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v25, 15, v25
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v24, 15, v24
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x10
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v31, 15, v31
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0xf
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v64, v32, 15
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0xe
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v30, v33, v30
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v33, v33, 15
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0xc
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v28, v35, v28
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v35, v35, 15
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0xa
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v26, v37, v26
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v37, v37, 15
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x9
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v23, v38, v23
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x7
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v21, v48, v21
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v48, v48, 15
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x6
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v20, v49, v20
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v49, v49, 15
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x3
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v17, v52, v17
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v52, v52, 15
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v48, -1, v48
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v35, -1, v35
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v49, -1, v49
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x2
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v16, v53, v16
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v52, -1, v52
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, v5, v48
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v53, v53, 15
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, v4, v49
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v12, v12, v35
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, v1, v52
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v5, v5, v21
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v53, -1, v53
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v4, v4, v20
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v12, v12, v28
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v1, v1, v17
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v5, 8, v5
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, v0, v53
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, 0xff, v4
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v12, 0xff, v12
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v38, v38, 15
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v22, v39, v22
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v0, v0, v16
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v17, v4, v5
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v29, 15, v29
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v39, v39, 15
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v19, v50, v19
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v50, v50, 15
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v18, v51, v18
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v29, v34, v29
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v34, v34, 15
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v51, v51, 15
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x1
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v25, v54, v25
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v54, v54, 15
+; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x0
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v24, v55, v24
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v34, -1, v34
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v55, v55, 15
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v31, v32, v31
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v32, -1, v64
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v33, -1, v33
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v13, v13, v34
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v37, -1, v37
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v38, -1, v38
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v39, -1, v39
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v50, -1, v50
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v13, v13, v29
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v51, -1, v51
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v54, -1, v54
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v55, -1, v55
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v15, v15, v32
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v13, 8, v13
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v14, v14, v33
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v10, v10, v37
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v7, v7, v38
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v6, v6, v39
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v16, v12, v13
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v27, 15, v27
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, v3, v50
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v2, v2, v51
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v9, v9, v54
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v8, v8, v55
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v27, v36, v27
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v36, v36, 15
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v15, v15, v31
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v14, v14, v30
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v10, v10, v26
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v7, v7, v23
+; GFX12-FAKE16-NEXT: v_xor_b32_e32 v36, -1, v36
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, v6, v22
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v3, v3, v19
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v2, v2, v18
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v9, v9, v25
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v11, v11, v36
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, v8, v24
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v15, 8, v15
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v14, 0xff, v14
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v10, 0xff, v10
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v11, v11, v27
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v7, 8, v7
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v6, 0xff, v6
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v3, 8, v3
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v2, 0xff, v2
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v11, 8, v11
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v9, 8, v9
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v1, 8, v1
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 0xff, v0
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v8, 0xff, v8
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v14, v14, v15
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v10, v10, v11
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, v6, v7
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v2, v2, v3
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v9
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff, v16
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v7, 0xffff, v17
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, v8, v9
+; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v11, 16, v10
+; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v13, 16, v14
+; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v9, 16, v6
+; GFX12-FAKE16-NEXT: v_lshl_or_b32 v15, v10, 16, v4
+; GFX12-FAKE16-NEXT: v_lshl_or_b32 v12, v14, 16, v5
+; GFX12-FAKE16-NEXT: v_lshl_or_b32 v4, v6, 16, v7
+; GFX12-FAKE16-NEXT: v_lshl_or_b32 v1, v2, 16, v1
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 24, v9
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v9, 8, v15
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v15, 24, v13
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v13, 8, v12
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 8, v4
+; GFX12-FAKE16-NEXT: v_lshrrev_b64 v[3:4], 24, v[3:4]
+; GFX12-FAKE16-NEXT: v_lshrrev_b64 v[11:12], 24, v[11:12]
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v17
+; GFX12-FAKE16-NEXT: v_mov_b32_e32 v12, v16
+; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
+ %result = bitinsert <16 x b8> %base, <16 x i4> %val, <16 x i32> %off
+ ret <16 x b8> %result
+}
diff --git a/llvm/test/CodeGen/RISCV/rvv/bitinsert-bitextract.ll b/llvm/test/CodeGen/RISCV/rvv/bitinsert-bitextract.ll
new file mode 100644
index 0000000000000..7ed246cfafaf1
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/bitinsert-bitextract.ll
@@ -0,0 +1,34 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=riscv32 -mattr=+v < %s | FileCheck %s
+; RUN: llc -mtriple=riscv64 -mattr=+v < %s | FileCheck %s
+
+define <vscale x 4 x i8> @test_bitextract_nxv4i8(<vscale x 4 x b32> %src, <vscale x 4 x i32> %off) {
+; CHECK-LABEL: test_bitextract_nxv4i8:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
+; CHECK-NEXT: vsrl.vv v8, v8, v10
+; CHECK-NEXT: vsetvli zero, zero, e16, m1, ta, ma
+; CHECK-NEXT: vnsrl.wi v10, v8, 0
+; CHECK-NEXT: vsetvli zero, zero, e8, mf2, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v10, 0
+; CHECK-NEXT: ret
+ %result = bitextract <vscale x 4 x i8>, <vscale x 4 x b32> %src, <vscale x 4 x i32> %off
+ ret <vscale x 4 x i8> %result
+}
+
+define <vscale x 4 x b32> @test_bitinsert_nxv4i8(<vscale x 4 x b32> %base, <vscale x 4 x i8> %val, <vscale x 4 x i32> %off) {
+; CHECK-LABEL: test_bitinsert_nxv4i8:
+; CHECK: # %bb.0:
+; CHECK-NEXT: li a0, 255
+; CHECK-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; CHECK-NEXT: vmv.v.x v14, a0
+; CHECK-NEXT: vzext.vf4 v16, v10
+; CHECK-NEXT: vsll.vv v10, v16, v12
+; CHECK-NEXT: vsll.vv v12, v14, v12
+; CHECK-NEXT: vnot.v v12, v12
+; CHECK-NEXT: vand.vv v8, v8, v12
+; CHECK-NEXT: vor.vv v8, v8, v10
+; CHECK-NEXT: ret
+ %result = bitinsert <vscale x 4 x b32> %base, <vscale x 4 x i8> %val, <vscale x 4 x i32> %off
+ ret <vscale x 4 x b32> %result
+}
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitinsert-bitextract.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitinsert-bitextract.ll
new file mode 100644
index 0000000000000..7b38d23cf3394
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitinsert-bitextract.ll
@@ -0,0 +1,240 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=riscv32 -mattr=+v < %s | FileCheck %s --check-prefixes=CHECK,RV32
+; RUN: llc -mtriple=riscv64 -mattr=+v < %s | FileCheck %s --check-prefixes=CHECK,RV64
+
+define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
+; CHECK-LABEL: test_bitextract_v4i8_const:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v8, 8
+; CHECK-NEXT: vsetvli zero, zero, e8, mf4, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v8, 0
+; CHECK-NEXT: ret
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
+ ret <4 x i8> %result
+}
+
+define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
+; CHECK-LABEL: test_bitextract_v4i8_var:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-NEXT: vsrl.vv v8, v8, v9
+; CHECK-NEXT: vsetvli zero, zero, e16, mf2, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v8, 0
+; CHECK-NEXT: vsetvli zero, zero, e8, mf4, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v8, 0
+; CHECK-NEXT: ret
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
+ ret <4 x i8> %result
+}
+
+define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
+; CHECK-LABEL: test_bitextract_v2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e64, m1, ta, ma
+; CHECK-NEXT: vid.v v9
+; CHECK-NEXT: vsll.vi v9, v9, 5
+; CHECK-NEXT: vsrl.vv v8, v8, v9
+; CHECK-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v8, 0
+; CHECK-NEXT: ret
+ %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
+ ret <2 x float> %result
+}
+
+define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
+; RV32-LABEL: test_bitextract_v2ptr:
+; RV32: # %bb.0:
+; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV32-NEXT: vnsrl.wi v8, v8, 0
+; RV32-NEXT: ret
+;
+; RV64-LABEL: test_bitextract_v2ptr:
+; RV64: # %bb.0:
+; RV64-NEXT: ret
+ %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
+ ret <2 x ptr> %result
+}
+
+define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
+; CHECK-LABEL: test_bitextract_v16i4_var:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, ma
+; CHECK-NEXT: vnsrl.wi v10, v12, 0
+; CHECK-NEXT: vsetvli zero, zero, e8, m1, ta, ma
+; CHECK-NEXT: vnsrl.wi v9, v10, 0
+; CHECK-NEXT: vsrl.vv v8, v8, v9
+; CHECK-NEXT: ret
+ %result = bitextract <16 x i4>, <16 x b8> %src, <16 x i32> %off
+ ret <16 x i4> %result
+}
+
+define <4 x b32> @test_bitinsert_v4i8_const(<4 x b32> %base, <4 x i8> %val) {
+; CHECK-LABEL: test_bitinsert_v4i8_const:
+; CHECK: # %bb.0:
+; CHECK-NEXT: lui a0, 1048560
+; CHECK-NEXT: addi a0, a0, 255
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-NEXT: vzext.vf4 v10, v9
+; CHECK-NEXT: vand.vx v8, v8, a0
+; CHECK-NEXT: vsll.vi v9, v10, 8
+; CHECK-NEXT: vor.vv v8, v8, v9
+; CHECK-NEXT: ret
+ %result = bitinsert <4 x b32> %base, <4 x i8> %val, <4 x i32> splat (i32 8)
+ ret <4 x b32> %result
+}
+
+define <4 x b32> @test_bitinsert_v4i8_var(<4 x b32> %base, <4 x i8> %val, <4 x i32> %off) {
+; CHECK-LABEL: test_bitinsert_v4i8_var:
+; CHECK: # %bb.0:
+; CHECK-NEXT: li a0, 255
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-NEXT: vmv.v.x v11, a0
+; CHECK-NEXT: vzext.vf4 v12, v9
+; CHECK-NEXT: vsll.vv v9, v11, v10
+; CHECK-NEXT: vsll.vv v10, v12, v10
+; CHECK-NEXT: vnot.v v9, v9
+; CHECK-NEXT: vand.vv v8, v8, v9
+; CHECK-NEXT: vor.vv v8, v8, v10
+; CHECK-NEXT: ret
+ %result = bitinsert <4 x b32> %base, <4 x i8> %val, <4 x i32> %off
+ ret <4 x b32> %result
+}
+
+define <2 x b64> @test_bitinsert_v2f32(<2 x b64> %base, <2 x float> %val) {
+; RV32-LABEL: test_bitinsert_v2f32:
+; RV32: # %bb.0:
+; RV32-NEXT: vsetivli zero, 2, e64, m1, ta, ma
+; RV32-NEXT: vid.v v10
+; RV32-NEXT: lui a0, 4096
+; RV32-NEXT: addi a0, a0, -256
+; RV32-NEXT: vmv.s.x v11, a0
+; RV32-NEXT: vsll.vi v10, v10, 5
+; RV32-NEXT: vzext.vf2 v12, v9
+; RV32-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV32-NEXT: vsext.vf4 v9, v11
+; RV32-NEXT: vsetivli zero, 2, e64, m1, ta, ma
+; RV32-NEXT: vsll.vv v10, v12, v10
+; RV32-NEXT: vand.vv v8, v8, v9
+; RV32-NEXT: vor.vv v8, v8, v10
+; RV32-NEXT: ret
+;
+; RV64-LABEL: test_bitinsert_v2f32:
+; RV64: # %bb.0:
+; RV64-NEXT: vsetivli zero, 2, e64, m1, ta, ma
+; RV64-NEXT: vid.v v10
+; RV64-NEXT: li a0, -1
+; RV64-NEXT: vsll.vi v10, v10, 5
+; RV64-NEXT: srli a1, a0, 32
+; RV64-NEXT: vmv.v.x v11, a1
+; RV64-NEXT: vzext.vf2 v12, v9
+; RV64-NEXT: slli a0, a0, 32
+; RV64-NEXT: vsetvli zero, zero, e64, m1, tu, ma
+; RV64-NEXT: vmv.s.x v11, a0
+; RV64-NEXT: vsetvli zero, zero, e64, m1, ta, ma
+; RV64-NEXT: vsll.vv v9, v12, v10
+; RV64-NEXT: vand.vv v8, v8, v11
+; RV64-NEXT: vor.vv v8, v8, v9
+; RV64-NEXT: ret
+ %result = bitinsert <2 x b64> %base, <2 x float> %val, <2 x i32> <i32 0, i32 32>
+ ret <2 x b64> %result
+}
+
+define <2 x b64> @test_bitinsert_v2ptr(<2 x b64> %base, <2 x ptr> %val) {
+; RV32-LABEL: test_bitinsert_v2ptr:
+; RV32: # %bb.0:
+; RV32-NEXT: addi sp, sp, -16
+; RV32-NEXT: .cfi_def_cfa_offset 16
+; RV32-NEXT: li a0, -1
+; RV32-NEXT: sw zero, 8(sp)
+; RV32-NEXT: sw a0, 12(sp)
+; RV32-NEXT: addi a0, sp, 8
+; RV32-NEXT: vsetivli zero, 2, e64, m1, ta, ma
+; RV32-NEXT: vlse64.v v10, (a0), zero
+; RV32-NEXT: vand.vv v8, v8, v10
+; RV32-NEXT: vzext.vf2 v10, v9
+; RV32-NEXT: vor.vv v8, v8, v10
+; RV32-NEXT: addi sp, sp, 16
+; RV32-NEXT: .cfi_def_cfa_offset 0
+; RV32-NEXT: ret
+;
+; RV64-LABEL: test_bitinsert_v2ptr:
+; RV64: # %bb.0:
+; RV64-NEXT: vsetivli zero, 1, e8, m1, ta, ma
+; RV64-NEXT: vmv1r.v v8, v9
+; RV64-NEXT: ret
+ %result = bitinsert <2 x b64> %base, <2 x ptr> %val, <2 x i32> zeroinitializer
+ ret <2 x b64> %result
+}
+
+define <2 x b128> @test_bitinsert_v2ptr_b128(<2 x b128> %base, <2 x ptr> %val) {
+; RV32-LABEL: test_bitinsert_v2ptr_b128:
+; RV32: # %bb.0:
+; RV32-NEXT: lw a2, 24(a1)
+; RV32-NEXT: lw a3, 28(a1)
+; RV32-NEXT: lw a4, 0(a1)
+; RV32-NEXT: lw a5, 8(a1)
+; RV32-NEXT: lw a6, 12(a1)
+; RV32-NEXT: lw a1, 16(a1)
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vslidedown.vi v9, v8, 1
+; RV32-NEXT: addi a7, a0, 20
+; RV32-NEXT: sw a2, 24(a0)
+; RV32-NEXT: sw a3, 28(a0)
+; RV32-NEXT: sw a4, 0(a0)
+; RV32-NEXT: sw a5, 8(a0)
+; RV32-NEXT: sw a6, 12(a0)
+; RV32-NEXT: sw a1, 16(a0)
+; RV32-NEXT: vse32.v v9, (a7)
+; RV32-NEXT: addi a0, a0, 4
+; RV32-NEXT: vse32.v v8, (a0)
+; RV32-NEXT: ret
+;
+; RV64-LABEL: test_bitinsert_v2ptr_b128:
+; RV64: # %bb.0:
+; RV64-NEXT: vsetivli zero, 1, e64, m1, ta, ma
+; RV64-NEXT: vmv.x.s a2, v8
+; RV64-NEXT: vslidedown.vi v8, v8, 1
+; RV64-NEXT: lwu a3, 0(a1)
+; RV64-NEXT: vmv.x.s a4, v8
+; RV64-NEXT: lw a5, 12(a1)
+; RV64-NEXT: lw a6, 28(a1)
+; RV64-NEXT: slli a7, a2, 32
+; RV64-NEXT: lwu a1, 16(a1)
+; RV64-NEXT: or a3, a3, a7
+; RV64-NEXT: slli a5, a5, 32
+; RV64-NEXT: srli a2, a2, 32
+; RV64-NEXT: slli a6, a6, 32
+; RV64-NEXT: or a2, a5, a2
+; RV64-NEXT: slli a5, a4, 32
+; RV64-NEXT: srli a4, a4, 32
+; RV64-NEXT: or a1, a1, a5
+; RV64-NEXT: or a4, a6, a4
+; RV64-NEXT: sd a3, 0(a0)
+; RV64-NEXT: sd a2, 8(a0)
+; RV64-NEXT: sd a1, 16(a0)
+; RV64-NEXT: sd a4, 24(a0)
+; RV64-NEXT: ret
+ %result = bitinsert <2 x b128> %base, <2 x ptr> %val, <2 x i32> splat (i32 32)
+ ret <2 x b128> %result
+}
+
+define <16 x b8> @test_bitinsert_v16i4_var(<16 x b8> %base, <16 x i4> %val, <16 x i32> %off) {
+; CHECK-LABEL: test_bitinsert_v16i4_var:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e8, m1, ta, ma
+; CHECK-NEXT: vmv.v.i v16, 15
+; CHECK-NEXT: vand.vi v9, v9, 15
+; CHECK-NEXT: vsetvli zero, zero, e16, m2, ta, ma
+; CHECK-NEXT: vnsrl.wi v10, v12, 0
+; CHECK-NEXT: vsetvli zero, zero, e8, m1, ta, ma
+; CHECK-NEXT: vnsrl.wi v12, v10, 0
+; CHECK-NEXT: vsll.vv v10, v16, v12
+; CHECK-NEXT: vnot.v v10, v10
+; CHECK-NEXT: vsll.vv v9, v9, v12
+; CHECK-NEXT: vand.vv v8, v8, v10
+; CHECK-NEXT: vor.vv v8, v8, v9
+; CHECK-NEXT: ret
+ %result = bitinsert <16 x b8> %base, <16 x i4> %val, <16 x i32> %off
+ ret <16 x b8> %result
+}
diff --git a/llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll b/llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll
new file mode 100644
index 0000000000000..227f8c6d97a29
--- /dev/null
+++ b/llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll
@@ -0,0 +1,476 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s --check-prefix=SSE2
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 < %s | FileCheck %s --check-prefix=AVX2
+
+define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
+; SSE2-LABEL: test_bitextract_v4i8_const:
+; SSE2: # %bb.0:
+; SSE2-NEXT: psrld $8, %xmm0
+; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: packuswb %xmm0, %xmm0
+; SSE2-NEXT: packuswb %xmm0, %xmm0
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitextract_v4i8_const:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-NEXT: retq
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
+ ret <4 x i8> %result
+}
+
+define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
+; SSE2-LABEL: test_bitextract_v4i8_var:
+; SSE2: # %bb.0:
+; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7]
+; SSE2-NEXT: movdqa %xmm0, %xmm3
+; SSE2-NEXT: psrld %xmm2, %xmm3
+; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm1[0,1,1,1,4,5,6,7]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: psrld %xmm4, %xmm2
+; SSE2-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
+; SSE2-NEXT: pshuflw {{.*#+}} xmm3 = xmm1[2,3,3,3,4,5,6,7]
+; SSE2-NEXT: movdqa %xmm0, %xmm4
+; SSE2-NEXT: psrld %xmm3, %xmm4
+; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7]
+; SSE2-NEXT: psrld %xmm1, %xmm0
+; SSE2-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1]
+; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm0[0,3]
+; SSE2-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: packuswb %xmm2, %xmm2
+; SSE2-NEXT: packuswb %xmm2, %xmm2
+; SSE2-NEXT: movdqa %xmm2, %xmm0
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitextract_v4i8_var:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-NEXT: retq
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
+ ret <4 x i8> %result
+}
+
+define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
+; SSE2-LABEL: test_bitextract_v2f32:
+; SSE2: # %bb.0:
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlq $32, %xmm1
+; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
+; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitextract_v2f32:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX2-NEXT: retq
+ %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
+ ret <2 x float> %result
+}
+
+define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
+; SSE2-LABEL: test_bitextract_v2ptr:
+; SSE2: # %bb.0:
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitextract_v2ptr:
+; AVX2: # %bb.0:
+; AVX2-NEXT: retq
+ %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
+ ret <2 x ptr> %result
+}
+
+define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
+; SSE2-LABEL: test_bitextract_v16i4_var:
+; SSE2: # %bb.0:
+; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; SSE2-NEXT: pand %xmm5, %xmm4
+; SSE2-NEXT: pand %xmm5, %xmm3
+; SSE2-NEXT: packuswb %xmm4, %xmm3
+; SSE2-NEXT: pand %xmm5, %xmm2
+; SSE2-NEXT: pand %xmm5, %xmm1
+; SSE2-NEXT: packuswb %xmm2, %xmm1
+; SSE2-NEXT: packuswb %xmm3, %xmm1
+; SSE2-NEXT: psllw $5, %xmm1
+; SSE2-NEXT: pxor %xmm2, %xmm2
+; SSE2-NEXT: pxor %xmm3, %xmm3
+; SSE2-NEXT: pcmpgtb %xmm1, %xmm3
+; SSE2-NEXT: movdqa %xmm3, %xmm4
+; SSE2-NEXT: pandn %xmm0, %xmm4
+; SSE2-NEXT: psrlw $4, %xmm0
+; SSE2-NEXT: pand %xmm3, %xmm0
+; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: por %xmm4, %xmm0
+; SSE2-NEXT: paddb %xmm1, %xmm1
+; SSE2-NEXT: pxor %xmm3, %xmm3
+; SSE2-NEXT: pcmpgtb %xmm1, %xmm3
+; SSE2-NEXT: movdqa %xmm3, %xmm4
+; SSE2-NEXT: pandn %xmm0, %xmm4
+; SSE2-NEXT: psrlw $2, %xmm0
+; SSE2-NEXT: pand %xmm3, %xmm0
+; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: por %xmm4, %xmm0
+; SSE2-NEXT: paddb %xmm1, %xmm1
+; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
+; SSE2-NEXT: movdqa %xmm2, %xmm1
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: psrlw $1, %xmm0
+; SSE2-NEXT: pand %xmm2, %xmm0
+; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: por %xmm1, %xmm0
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitextract_v16i4_var:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX2-NEXT: vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT: vpand %ymm3, %ymm1, %ymm1
+; AVX2-NEXT: vpackusdw %ymm2, %ymm1, %ymm1
+; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT: vpackuswb %xmm2, %xmm1, %xmm1
+; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,1,3]
+; AVX2-NEXT: vpsllw $5, %xmm1, %xmm1
+; AVX2-NEXT: vpaddb %xmm1, %xmm1, %xmm2
+; AVX2-NEXT: vpsrlw $4, %xmm0, %xmm3
+; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
+; AVX2-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0
+; AVX2-NEXT: vpsrlw $2, %xmm0, %xmm1
+; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vpsrlw $1, %xmm0, %xmm1
+; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT: vpaddb %xmm2, %xmm2, %xmm2
+; AVX2-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vzeroupper
+; AVX2-NEXT: retq
+ %result = bitextract <16 x i4>, <16 x b8> %src, <16 x i32> %off
+ ret <16 x i4> %result
+}
+
+define <4 x b32> @test_bitinsert_v4i8_const(<4 x b32> %base, <4 x i8> %val) {
+; SSE2-LABEL: test_bitinsert_v4i8_const:
+; SSE2: # %bb.0:
+; SSE2-NEXT: pxor %xmm2, %xmm2
+; SSE2-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; SSE2-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; SSE2-NEXT: pslld $8, %xmm1
+; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: por %xmm1, %xmm0
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitinsert_v4i8_const:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT: vpmovzxbd {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
+; AVX2-NEXT: vpslld $8, %xmm1, %xmm1
+; AVX2-NEXT: vpor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: retq
+ %result = bitinsert <4 x b32> %base, <4 x i8> %val, <4 x i32> splat (i32 8)
+ ret <4 x b32> %result
+}
+
+define <4 x b32> @test_bitinsert_v4i8_var(<4 x b32> %base, <4 x i8> %val, <4 x i32> %off) {
+; SSE2-LABEL: test_bitinsert_v4i8_var:
+; SSE2: # %bb.0:
+; SSE2-NEXT: pxor %xmm3, %xmm3
+; SSE2-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
+; SSE2-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
+; SSE2-NEXT: pslld $23, %xmm2
+; SSE2-NEXT: paddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2 # [1065353216,1065353216,1065353216,1065353216]
+; SSE2-NEXT: cvttps2dq %xmm2, %xmm2
+; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; SSE2-NEXT: pmuludq %xmm2, %xmm1
+; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm1[0,2,2,3]
+; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm2[1,1,3,3]
+; SSE2-NEXT: pmuludq %xmm5, %xmm3
+; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
+; SSE2-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
+; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255]
+; SSE2-NEXT: pmuludq %xmm3, %xmm2
+; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
+; SSE2-NEXT: pmuludq %xmm3, %xmm5
+; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm5[0,2,2,3]
+; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm4, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitinsert_v4i8_var:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpmovzxbd {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
+; AVX2-NEXT: vpsllvd %xmm2, %xmm1, %xmm1
+; AVX2-NEXT: vpbroadcastd {{.*#+}} xmm3 = [255,255,255,255]
+; AVX2-NEXT: vpsllvd %xmm2, %xmm3, %xmm2
+; AVX2-NEXT: vpandn %xmm0, %xmm2, %xmm0
+; AVX2-NEXT: vpor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: retq
+ %result = bitinsert <4 x b32> %base, <4 x i8> %val, <4 x i32> %off
+ ret <4 x b32> %result
+}
+
+define <2 x b64> @test_bitinsert_v2f32(<2 x b64> %base, <2 x float> %val) {
+; SSE2-LABEL: test_bitinsert_v2f32:
+; SSE2: # %bb.0:
+; SSE2-NEXT: xorps %xmm2, %xmm2
+; SSE2-NEXT: xorps %xmm3, %xmm3
+; SSE2-NEXT: movss {{.*#+}} xmm3 = xmm1[0],xmm3[1,2,3]
+; SSE2-NEXT: unpcklps {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; SSE2-NEXT: movsd {{.*#+}} xmm2 = xmm3[0],xmm2[1]
+; SSE2-NEXT: andpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: orpd %xmm2, %xmm0
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitinsert_v2f32:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2
+; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm2[0],xmm0[1,2],xmm2[3]
+; AVX2-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; AVX2-NEXT: vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT: vpor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: retq
+ %result = bitinsert <2 x b64> %base, <2 x float> %val, <2 x i32> <i32 0, i32 32>
+ ret <2 x b64> %result
+}
+
+define <2 x b64> @test_bitinsert_v2ptr(<2 x b64> %base, <2 x ptr> %val) {
+; SSE2-LABEL: test_bitinsert_v2ptr:
+; SSE2: # %bb.0:
+; SSE2-NEXT: movaps %xmm1, %xmm0
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitinsert_v2ptr:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vmovaps %xmm1, %xmm0
+; AVX2-NEXT: retq
+ %result = bitinsert <2 x b64> %base, <2 x ptr> %val, <2 x i32> zeroinitializer
+ ret <2 x b64> %result
+}
+
+define <2 x b128> @test_bitinsert_v2ptr_b128(<2 x b128> %base, <2 x ptr> %val) {
+; SSE2-LABEL: test_bitinsert_v2ptr_b128:
+; SSE2: # %bb.0:
+; SSE2-NEXT: movq %rdi, %rax
+; SSE2-NEXT: movq %xmm0, %rdi
+; SSE2-NEXT: movq %rdi, %r9
+; SSE2-NEXT: shlq $32, %r9
+; SSE2-NEXT: movl %esi, %esi
+; SSE2-NEXT: orq %r9, %rsi
+; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
+; SSE2-NEXT: movq %xmm0, %r9
+; SSE2-NEXT: movq %r9, %r10
+; SSE2-NEXT: shlq $32, %r10
+; SSE2-NEXT: movl %ecx, %ecx
+; SSE2-NEXT: orq %r10, %rcx
+; SSE2-NEXT: movabsq $-4294967296, %r10 # imm = 0xFFFFFFFF00000000
+; SSE2-NEXT: andq %r10, %r8
+; SSE2-NEXT: shrq $32, %r9
+; SSE2-NEXT: orq %r8, %r9
+; SSE2-NEXT: andq %r10, %rdx
+; SSE2-NEXT: shrq $32, %rdi
+; SSE2-NEXT: orq %rdx, %rdi
+; SSE2-NEXT: movq %rdi, 8(%rax)
+; SSE2-NEXT: movq %r9, 24(%rax)
+; SSE2-NEXT: movq %rcx, 16(%rax)
+; SSE2-NEXT: movq %rsi, (%rax)
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitinsert_v2ptr_b128:
+; AVX2: # %bb.0:
+; AVX2-NEXT: movq %rdi, %rax
+; AVX2-NEXT: shrq $32, %r8
+; AVX2-NEXT: shrq $32, %rdx
+; AVX2-NEXT: vmovq %xmm0, %rdi
+; AVX2-NEXT: movq %rdi, %r9
+; AVX2-NEXT: shrq $32, %r9
+; AVX2-NEXT: vmovd %edx, %xmm1
+; AVX2-NEXT: vpslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3]
+; AVX2-NEXT: vpextrq $1, %xmm1, %rdx
+; AVX2-NEXT: orq %r9, %rdx
+; AVX2-NEXT: vpextrq $1, %xmm0, %r9
+; AVX2-NEXT: movq %r9, %r10
+; AVX2-NEXT: shrq $32, %r10
+; AVX2-NEXT: vmovd %r8d, %xmm0
+; AVX2-NEXT: vpslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3]
+; AVX2-NEXT: vpextrq $1, %xmm0, %r8
+; AVX2-NEXT: orq %r10, %r8
+; AVX2-NEXT: shlq $32, %rdi
+; AVX2-NEXT: movl %esi, %esi
+; AVX2-NEXT: orq %rdi, %rsi
+; AVX2-NEXT: shlq $32, %r9
+; AVX2-NEXT: movl %ecx, %ecx
+; AVX2-NEXT: orq %r9, %rcx
+; AVX2-NEXT: movq %rcx, 16(%rax)
+; AVX2-NEXT: movq %rsi, (%rax)
+; AVX2-NEXT: movq %r8, 24(%rax)
+; AVX2-NEXT: movq %rdx, 8(%rax)
+; AVX2-NEXT: retq
+ %result = bitinsert <2 x b128> %base, <2 x ptr> %val, <2 x i32> splat (i32 32)
+ ret <2 x b128> %result
+}
+
+define <2 x b128> @test_bitinsert_v2ptr_addrspace3_const(<2 x b128> %base, <2 x ptr addrspace(3)> %val) {
+; SSE2-LABEL: test_bitinsert_v2ptr_addrspace3_const:
+; SSE2: # %bb.0:
+; SSE2-NEXT: movq %rdi, %rax
+; SSE2-NEXT: movq %xmm0, %rdi
+; SSE2-NEXT: movq %rdi, %r9
+; SSE2-NEXT: shlq $16, %r9
+; SSE2-NEXT: movzwl %si, %esi
+; SSE2-NEXT: orq %r9, %rsi
+; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
+; SSE2-NEXT: movq %xmm0, %r9
+; SSE2-NEXT: movq %r9, %r10
+; SSE2-NEXT: shlq $16, %r10
+; SSE2-NEXT: movzwl %cx, %ecx
+; SSE2-NEXT: orq %r10, %rcx
+; SSE2-NEXT: andq $-65536, %r8 # imm = 0xFFFF0000
+; SSE2-NEXT: shrq $48, %r9
+; SSE2-NEXT: orq %r8, %r9
+; SSE2-NEXT: andq $-65536, %rdx # imm = 0xFFFF0000
+; SSE2-NEXT: shrq $48, %rdi
+; SSE2-NEXT: orq %rdx, %rdi
+; SSE2-NEXT: movq %rdi, 8(%rax)
+; SSE2-NEXT: movq %r9, 24(%rax)
+; SSE2-NEXT: movq %rcx, 16(%rax)
+; SSE2-NEXT: movq %rsi, (%rax)
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitinsert_v2ptr_addrspace3_const:
+; AVX2: # %bb.0:
+; AVX2-NEXT: movq %rdi, %rax
+; AVX2-NEXT: movq %r8, %rdi
+; AVX2-NEXT: shrq $48, %rdi
+; AVX2-NEXT: movq %r8, %r9
+; AVX2-NEXT: shrq $32, %r9
+; AVX2-NEXT: shrl $16, %r8d
+; AVX2-NEXT: movq %rdx, %r10
+; AVX2-NEXT: shrq $48, %r10
+; AVX2-NEXT: movq %rdx, %r11
+; AVX2-NEXT: shrq $32, %r11
+; AVX2-NEXT: shrl $16, %edx
+; AVX2-NEXT: movzwl %si, %esi
+; AVX2-NEXT: vmovd %esi, %xmm1
+; AVX2-NEXT: vpinsrw $5, %edx, %xmm1, %xmm1
+; AVX2-NEXT: vpinsrw $6, %r11d, %xmm1, %xmm1
+; AVX2-NEXT: vpinsrw $7, %r10d, %xmm1, %xmm1
+; AVX2-NEXT: vpextrq $1, %xmm1, %r10
+; AVX2-NEXT: vmovq %xmm0, %r11
+; AVX2-NEXT: movq %r11, %rdx
+; AVX2-NEXT: shrq $48, %rdx
+; AVX2-NEXT: orq %r10, %rdx
+; AVX2-NEXT: movzwl %cx, %ecx
+; AVX2-NEXT: vmovd %ecx, %xmm1
+; AVX2-NEXT: vpinsrw $5, %r8d, %xmm1, %xmm1
+; AVX2-NEXT: vpinsrw $6, %r9d, %xmm1, %xmm1
+; AVX2-NEXT: vpinsrw $7, %edi, %xmm1, %xmm1
+; AVX2-NEXT: vpextrq $1, %xmm1, %rdi
+; AVX2-NEXT: vpextrq $1, %xmm0, %r8
+; AVX2-NEXT: movq %r8, %r9
+; AVX2-NEXT: shrq $48, %r9
+; AVX2-NEXT: orq %rdi, %r9
+; AVX2-NEXT: shlq $16, %r11
+; AVX2-NEXT: orq %r11, %rsi
+; AVX2-NEXT: shlq $16, %r8
+; AVX2-NEXT: orq %r8, %rcx
+; AVX2-NEXT: movq %rcx, 16(%rax)
+; AVX2-NEXT: movq %rsi, (%rax)
+; AVX2-NEXT: movq %r9, 24(%rax)
+; AVX2-NEXT: movq %rdx, 8(%rax)
+; AVX2-NEXT: retq
+ %result = bitinsert <2 x b128> %base, <2 x ptr addrspace(3)> %val, <2 x i32> splat (i32 16)
+ ret <2 x b128> %result
+}
+
+define <16 x b8> @test_bitinsert_v16i4_var(<16 x b8> %base, <16 x i4> %val, <16 x i32> %off) {
+; SSE2-LABEL: test_bitinsert_v16i4_var:
+; SSE2: # %bb.0:
+; SSE2-NEXT: movdqa {{.*#+}} xmm6 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; SSE2-NEXT: pand %xmm6, %xmm5
+; SSE2-NEXT: pand %xmm6, %xmm4
+; SSE2-NEXT: packuswb %xmm5, %xmm4
+; SSE2-NEXT: pand %xmm6, %xmm3
+; SSE2-NEXT: pand %xmm6, %xmm2
+; SSE2-NEXT: packuswb %xmm3, %xmm2
+; SSE2-NEXT: packuswb %xmm4, %xmm2
+; SSE2-NEXT: psllw $5, %xmm2
+; SSE2-NEXT: pxor %xmm3, %xmm3
+; SSE2-NEXT: pxor %xmm5, %xmm5
+; SSE2-NEXT: pcmpgtb %xmm2, %xmm5
+; SSE2-NEXT: movdqa {{.*#+}} xmm6 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE2-NEXT: pand %xmm1, %xmm6
+; SSE2-NEXT: movdqa %xmm5, %xmm4
+; SSE2-NEXT: pandn %xmm6, %xmm4
+; SSE2-NEXT: movdqa {{.*#+}} xmm8 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; SSE2-NEXT: movdqa %xmm5, %xmm6
+; SSE2-NEXT: pand %xmm8, %xmm6
+; SSE2-NEXT: psllw $4, %xmm1
+; SSE2-NEXT: pand %xmm6, %xmm1
+; SSE2-NEXT: por %xmm1, %xmm4
+; SSE2-NEXT: paddb %xmm2, %xmm2
+; SSE2-NEXT: pxor %xmm7, %xmm7
+; SSE2-NEXT: pcmpgtb %xmm2, %xmm7
+; SSE2-NEXT: movdqa %xmm7, %xmm1
+; SSE2-NEXT: pandn %xmm4, %xmm1
+; SSE2-NEXT: paddb %xmm4, %xmm4
+; SSE2-NEXT: paddb %xmm4, %xmm4
+; SSE2-NEXT: pand %xmm7, %xmm4
+; SSE2-NEXT: por %xmm1, %xmm4
+; SSE2-NEXT: paddb %xmm2, %xmm2
+; SSE2-NEXT: pcmpgtb %xmm2, %xmm3
+; SSE2-NEXT: movdqa %xmm3, %xmm1
+; SSE2-NEXT: pandn %xmm4, %xmm1
+; SSE2-NEXT: paddb %xmm4, %xmm4
+; SSE2-NEXT: pand %xmm3, %xmm4
+; SSE2-NEXT: por %xmm1, %xmm4
+; SSE2-NEXT: por %xmm8, %xmm5
+; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT: pxor %xmm5, %xmm1
+; SSE2-NEXT: por %xmm6, %xmm1
+; SSE2-NEXT: movdqa %xmm7, %xmm2
+; SSE2-NEXT: pandn %xmm1, %xmm2
+; SSE2-NEXT: paddb %xmm1, %xmm1
+; SSE2-NEXT: paddb %xmm1, %xmm1
+; SSE2-NEXT: pand %xmm7, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm3, %xmm2
+; SSE2-NEXT: pandn %xmm1, %xmm2
+; SSE2-NEXT: paddb %xmm1, %xmm1
+; SSE2-NEXT: pand %xmm3, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm4, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: retq
+;
+; AVX2-LABEL: test_bitinsert_v16i4_var:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX2-NEXT: vpand %ymm4, %ymm3, %ymm3
+; AVX2-NEXT: vpand %ymm4, %ymm2, %ymm2
+; AVX2-NEXT: vpackusdw %ymm3, %ymm2, %ymm2
+; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm3
+; AVX2-NEXT: vpackuswb %xmm3, %xmm2, %xmm2
+; AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,2,1,3]
+; AVX2-NEXT: vpsllw $5, %xmm2, %xmm3
+; AVX2-NEXT: vpaddb %xmm3, %xmm3, %xmm4
+; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT: vpsllw $4, %xmm1, %xmm5
+; AVX2-NEXT: vpblendvb %xmm3, %xmm5, %xmm1, %xmm1
+; AVX2-NEXT: vpaddb %xmm1, %xmm1, %xmm3
+; AVX2-NEXT: vpaddb %xmm3, %xmm3, %xmm3
+; AVX2-NEXT: vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
+; AVX2-NEXT: vpaddb %xmm1, %xmm1, %xmm3
+; AVX2-NEXT: vpaddb %xmm4, %xmm4, %xmm4
+; AVX2-NEXT: vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
+; AVX2-NEXT: vmovq {{.*#+}} xmm3 = [15,30,60,120,240,224,192,128,0,0,0,0,0,0,0,0]
+; AVX2-NEXT: vpshufb %xmm2, %xmm3, %xmm2
+; AVX2-NEXT: vpandn %xmm0, %xmm2, %xmm0
+; AVX2-NEXT: vpor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vzeroupper
+; AVX2-NEXT: retq
+ %result = bitinsert <16 x b8> %base, <16 x i4> %val, <16 x i32> %off
+ ret <16 x b8> %result
+}
>From c4837f8fcf5f168b2fb2d2080129b835e10735d0 Mon Sep 17 00:00:00 2001
From: Pedro Lobo <pedro.lobo at tecnico.ulisboa.pt>
Date: Tue, 10 Nov 2026 21:51:08 +0000
Subject: [PATCH 2/3] [SelectionDAG] Use x86-64 levels in vector
`bitinsert`/`bitextract` X86 test
---
.../X86/bitinsert-bitextract-vector.ll | 419 +++++++++++++++---
1 file changed, 361 insertions(+), 58 deletions(-)
diff --git a/llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll b/llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll
index 227f8c6d97a29..b8bf19a0ebde9 100644
--- a/llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll
+++ b/llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll
@@ -1,6 +1,8 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
-; RUN: llc -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s --check-prefix=SSE2
-; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 < %s | FileCheck %s --check-prefix=AVX2
+; RUN: llc -mtriple=x86_64-- -mcpu=x86-64 < %s | FileCheck %s --check-prefixes=SSE,SSE2
+; RUN: llc -mtriple=x86_64-- -mcpu=x86-64-v2 < %s | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: llc -mtriple=x86_64-- -mcpu=x86-64-v3 < %s | FileCheck %s --check-prefixes=AVX,AVX2
+; RUN: llc -mtriple=x86_64-- -mcpu=x86-64-v4 < %s | FileCheck %s --check-prefixes=AVX,AVX512
define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
; SSE2-LABEL: test_bitextract_v4i8_const:
@@ -11,10 +13,21 @@ define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
; SSE2-NEXT: packuswb %xmm0, %xmm0
; SSE2-NEXT: retq
;
+; SSE42-LABEL: test_bitextract_v4i8_const:
+; SSE42: # %bb.0:
+; SSE42-NEXT: pshufb {{.*#+}} xmm0 = xmm0[1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u]
+; SSE42-NEXT: retq
+;
; AVX2-LABEL: test_bitextract_v4i8_const:
; AVX2: # %bb.0:
; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u]
; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitextract_v4i8_const:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpsrld $8, %xmm0, %xmm0
+; AVX512-NEXT: vpmovdb %xmm0, %xmm0
+; AVX512-NEXT: retq
%result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
ret <4 x i8> %result
}
@@ -28,14 +41,14 @@ define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm1[0,1,1,1,4,5,6,7]
; SSE2-NEXT: movdqa %xmm0, %xmm2
; SSE2-NEXT: psrld %xmm4, %xmm2
-; SSE2-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm3 = xmm1[2,3,3,3,4,5,6,7]
-; SSE2-NEXT: movdqa %xmm0, %xmm4
-; SSE2-NEXT: psrld %xmm3, %xmm4
+; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm1[2,3,3,3,4,5,6,7]
+; SSE2-NEXT: movdqa %xmm0, %xmm5
+; SSE2-NEXT: psrld %xmm4, %xmm5
+; SSE2-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7]
; SSE2-NEXT: psrld %xmm1, %xmm0
-; SSE2-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1]
+; SSE2-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm5[1]
; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm0[0,3]
; SSE2-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE2-NEXT: packuswb %xmm2, %xmm2
@@ -43,11 +56,37 @@ define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
; SSE2-NEXT: movdqa %xmm2, %xmm0
; SSE2-NEXT: retq
;
+; SSE42-LABEL: test_bitextract_v4i8_var:
+; SSE42: # %bb.0:
+; SSE42-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7]
+; SSE42-NEXT: movdqa %xmm0, %xmm3
+; SSE42-NEXT: psrld %xmm2, %xmm3
+; SSE42-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
+; SSE42-NEXT: pshuflw {{.*#+}} xmm4 = xmm2[2,3,3,3,4,5,6,7]
+; SSE42-NEXT: movdqa %xmm0, %xmm5
+; SSE42-NEXT: psrld %xmm4, %xmm5
+; SSE42-NEXT: pblendw {{.*#+}} xmm5 = xmm3[0,1,2,3],xmm5[4,5,6,7]
+; SSE42-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7]
+; SSE42-NEXT: movdqa %xmm0, %xmm3
+; SSE42-NEXT: psrld %xmm1, %xmm3
+; SSE42-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,1,1,1,4,5,6,7]
+; SSE42-NEXT: psrld %xmm1, %xmm0
+; SSE42-NEXT: pblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7]
+; SSE42-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3],xmm0[4,5],xmm5[6,7]
+; SSE42-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; SSE42-NEXT: retq
+;
; AVX2-LABEL: test_bitextract_v4i8_var:
; AVX2: # %bb.0:
; AVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitextract_v4i8_var:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0
+; AVX512-NEXT: vpmovdb %xmm0, %xmm0
+; AVX512-NEXT: retq
%result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
ret <4 x i8> %result
}
@@ -61,23 +100,27 @@ define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2,2,3]
; SSE2-NEXT: retq
;
-; AVX2-LABEL: test_bitextract_v2f32:
-; AVX2: # %bb.0:
-; AVX2-NEXT: vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX2-NEXT: retq
+; SSE42-LABEL: test_bitextract_v2f32:
+; SSE42: # %bb.0:
+; SSE42-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,12,13,14,15,12,13,14,15],zero,zero,zero,zero
+; SSE42-NEXT: retq
+;
+; AVX-LABEL: test_bitextract_v2f32:
+; AVX: # %bb.0:
+; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,12,13,14,15,12,13,14,15],zero,zero,zero,zero
+; AVX-NEXT: retq
%result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
ret <2 x float> %result
}
define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
-; SSE2-LABEL: test_bitextract_v2ptr:
-; SSE2: # %bb.0:
-; SSE2-NEXT: retq
+; SSE-LABEL: test_bitextract_v2ptr:
+; SSE: # %bb.0:
+; SSE-NEXT: retq
;
-; AVX2-LABEL: test_bitextract_v2ptr:
-; AVX2: # %bb.0:
-; AVX2-NEXT: retq
+; AVX-LABEL: test_bitextract_v2ptr:
+; AVX: # %bb.0:
+; AVX-NEXT: retq
%result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
ret <2 x ptr> %result
}
@@ -95,23 +138,23 @@ define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
; SSE2-NEXT: packuswb %xmm3, %xmm1
; SSE2-NEXT: psllw $5, %xmm1
; SSE2-NEXT: pxor %xmm2, %xmm2
-; SSE2-NEXT: pxor %xmm3, %xmm3
-; SSE2-NEXT: pcmpgtb %xmm1, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm4
-; SSE2-NEXT: pandn %xmm0, %xmm4
+; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
+; SSE2-NEXT: movdqa %xmm2, %xmm3
+; SSE2-NEXT: pandn %xmm0, %xmm3
; SSE2-NEXT: psrlw $4, %xmm0
-; SSE2-NEXT: pand %xmm3, %xmm0
+; SSE2-NEXT: pand %xmm2, %xmm0
; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: por %xmm4, %xmm0
+; SSE2-NEXT: por %xmm3, %xmm0
; SSE2-NEXT: paddb %xmm1, %xmm1
-; SSE2-NEXT: pxor %xmm3, %xmm3
-; SSE2-NEXT: pcmpgtb %xmm1, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm4
-; SSE2-NEXT: pandn %xmm0, %xmm4
+; SSE2-NEXT: pxor %xmm2, %xmm2
+; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
+; SSE2-NEXT: movdqa %xmm2, %xmm3
+; SSE2-NEXT: pandn %xmm0, %xmm3
; SSE2-NEXT: psrlw $2, %xmm0
-; SSE2-NEXT: pand %xmm3, %xmm0
+; SSE2-NEXT: pand %xmm2, %xmm0
; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: por %xmm4, %xmm0
+; SSE2-NEXT: pxor %xmm2, %xmm2
+; SSE2-NEXT: por %xmm3, %xmm0
; SSE2-NEXT: paddb %xmm1, %xmm1
; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
; SSE2-NEXT: movdqa %xmm2, %xmm1
@@ -122,6 +165,38 @@ define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
; SSE2-NEXT: por %xmm1, %xmm0
; SSE2-NEXT: retq
;
+; SSE42-LABEL: test_bitextract_v16i4_var:
+; SSE42: # %bb.0:
+; SSE42-NEXT: movdqa %xmm0, %xmm5
+; SSE42-NEXT: movdqa {{.*#+}} xmm0 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; SSE42-NEXT: pand %xmm0, %xmm4
+; SSE42-NEXT: pand %xmm0, %xmm3
+; SSE42-NEXT: packusdw %xmm4, %xmm3
+; SSE42-NEXT: pand %xmm0, %xmm2
+; SSE42-NEXT: pand %xmm1, %xmm0
+; SSE42-NEXT: packusdw %xmm2, %xmm0
+; SSE42-NEXT: packuswb %xmm3, %xmm0
+; SSE42-NEXT: psllw $5, %xmm0
+; SSE42-NEXT: movdqa %xmm0, %xmm1
+; SSE42-NEXT: movdqa %xmm5, %xmm2
+; SSE42-NEXT: psrlw $4, %xmm2
+; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE42-NEXT: paddb %xmm0, %xmm1
+; SSE42-NEXT: pblendvb %xmm0, %xmm2, %xmm5
+; SSE42-NEXT: movdqa %xmm5, %xmm2
+; SSE42-NEXT: psrlw $2, %xmm2
+; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE42-NEXT: movdqa %xmm1, %xmm0
+; SSE42-NEXT: pblendvb %xmm0, %xmm2, %xmm5
+; SSE42-NEXT: movdqa %xmm5, %xmm2
+; SSE42-NEXT: psrlw $1, %xmm2
+; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE42-NEXT: paddb %xmm1, %xmm1
+; SSE42-NEXT: movdqa %xmm1, %xmm0
+; SSE42-NEXT: pblendvb %xmm0, %xmm2, %xmm5
+; SSE42-NEXT: movdqa %xmm5, %xmm0
+; SSE42-NEXT: retq
+;
; AVX2-LABEL: test_bitextract_v16i4_var:
; AVX2: # %bb.0:
; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
@@ -132,19 +207,29 @@ define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
; AVX2-NEXT: vpackuswb %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,1,3]
; AVX2-NEXT: vpsllw $5, %xmm1, %xmm1
-; AVX2-NEXT: vpaddb %xmm1, %xmm1, %xmm2
-; AVX2-NEXT: vpsrlw $4, %xmm0, %xmm3
-; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
-; AVX2-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0
+; AVX2-NEXT: vpsrlw $4, %xmm0, %xmm2
+; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
+; AVX2-NEXT: vpaddb %xmm1, %xmm1, %xmm3
+; AVX2-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
; AVX2-NEXT: vpsrlw $2, %xmm0, %xmm1
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX2-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpsrlw $1, %xmm0, %xmm1
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX2-NEXT: vpaddb %xmm2, %xmm2, %xmm2
+; AVX2-NEXT: vpaddb %xmm3, %xmm3, %xmm2
; AVX2-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitextract_v16i4_var:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpmovdw %zmm1, %ymm1
+; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm1
+; AVX512-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0
+; AVX512-NEXT: vpmovwb %ymm0, %xmm0
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%result = bitextract <16 x i4>, <16 x b8> %src, <16 x i32> %off
ret <16 x i4> %result
}
@@ -160,13 +245,26 @@ define <4 x b32> @test_bitinsert_v4i8_const(<4 x b32> %base, <4 x i8> %val) {
; SSE2-NEXT: por %xmm1, %xmm0
; SSE2-NEXT: retq
;
+; SSE42-LABEL: test_bitinsert_v4i8_const:
+; SSE42: # %bb.0:
+; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE42-NEXT: pmovzxbd {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
+; SSE42-NEXT: pslld $8, %xmm1
+; SSE42-NEXT: por %xmm1, %xmm0
+; SSE42-NEXT: retq
+;
; AVX2-LABEL: test_bitinsert_v4i8_const:
; AVX2: # %bb.0:
+; AVX2-NEXT: vpshufb {{.*#+}} xmm1 = zero,xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-NEXT: vpmovzxbd {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
-; AVX2-NEXT: vpslld $8, %xmm1, %xmm1
; AVX2-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitinsert_v4i8_const:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpshufb {{.*#+}} xmm1 = zero,xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero
+; AVX512-NEXT: vpternlogd {{.*#+}} xmm0 = (xmm0 & m32bcst) | xmm1
+; AVX512-NEXT: retq
%result = bitinsert <4 x b32> %base, <4 x i8> %val, <4 x i32> splat (i32 8)
ret <4 x b32> %result
}
@@ -198,6 +296,21 @@ define <4 x b32> @test_bitinsert_v4i8_var(<4 x b32> %base, <4 x i8> %val, <4 x i
; SSE2-NEXT: movdqa %xmm1, %xmm0
; SSE2-NEXT: retq
;
+; SSE42-LABEL: test_bitinsert_v4i8_var:
+; SSE42: # %bb.0:
+; SSE42-NEXT: pmovzxbd {{.*#+}} xmm3 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
+; SSE42-NEXT: pslld $23, %xmm2
+; SSE42-NEXT: paddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2 # [1065353216,1065353216,1065353216,1065353216]
+; SSE42-NEXT: cvttps2dq %xmm2, %xmm2
+; SSE42-NEXT: pmulld %xmm2, %xmm3
+; SSE42-NEXT: movdqa %xmm2, %xmm1
+; SSE42-NEXT: pslld $8, %xmm1
+; SSE42-NEXT: psubd %xmm2, %xmm1
+; SSE42-NEXT: pandn %xmm0, %xmm1
+; SSE42-NEXT: por %xmm3, %xmm1
+; SSE42-NEXT: movdqa %xmm1, %xmm0
+; SSE42-NEXT: retq
+;
; AVX2-LABEL: test_bitinsert_v4i8_var:
; AVX2: # %bb.0:
; AVX2-NEXT: vpmovzxbd {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
@@ -207,6 +320,15 @@ define <4 x b32> @test_bitinsert_v4i8_var(<4 x b32> %base, <4 x i8> %val, <4 x i
; AVX2-NEXT: vpandn %xmm0, %xmm2, %xmm0
; AVX2-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitinsert_v4i8_var:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpbroadcastd {{.*#+}} xmm3 = [255,255,255,255]
+; AVX512-NEXT: vpsllvd %xmm2, %xmm3, %xmm3
+; AVX512-NEXT: vpmovzxbd {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
+; AVX512-NEXT: vpsllvd %xmm2, %xmm1, %xmm1
+; AVX512-NEXT: vpternlogd {{.*#+}} xmm0 = (xmm0 & ~xmm3) | xmm1
+; AVX512-NEXT: retq
%result = bitinsert <4 x b32> %base, <4 x i8> %val, <4 x i32> %off
ret <4 x b32> %result
}
@@ -223,28 +345,34 @@ define <2 x b64> @test_bitinsert_v2f32(<2 x b64> %base, <2 x float> %val) {
; SSE2-NEXT: orpd %xmm2, %xmm0
; SSE2-NEXT: retq
;
-; AVX2-LABEL: test_bitinsert_v2f32:
-; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm2[0],xmm0[1,2],xmm2[3]
-; AVX2-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; AVX2-NEXT: vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX2-NEXT: vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: retq
+; SSE42-LABEL: test_bitinsert_v2f32:
+; SSE42: # %bb.0:
+; SSE42-NEXT: movddup {{.*#+}} xmm1 = xmm1[0,0]
+; SSE42-NEXT: blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2],xmm1[3]
+; SSE42-NEXT: retq
+;
+; AVX-LABEL: test_bitinsert_v2f32:
+; AVX: # %bb.0:
+; AVX-NEXT: vpxor %xmm2, %xmm2, %xmm2
+; AVX-NEXT: vpblendd {{.*#+}} xmm0 = xmm2[0],xmm0[1,2],xmm2[3]
+; AVX-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; AVX-NEXT: vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX-NEXT: vpor %xmm1, %xmm0, %xmm0
+; AVX-NEXT: retq
%result = bitinsert <2 x b64> %base, <2 x float> %val, <2 x i32> <i32 0, i32 32>
ret <2 x b64> %result
}
define <2 x b64> @test_bitinsert_v2ptr(<2 x b64> %base, <2 x ptr> %val) {
-; SSE2-LABEL: test_bitinsert_v2ptr:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movaps %xmm1, %xmm0
-; SSE2-NEXT: retq
+; SSE-LABEL: test_bitinsert_v2ptr:
+; SSE: # %bb.0:
+; SSE-NEXT: movaps %xmm1, %xmm0
+; SSE-NEXT: retq
;
-; AVX2-LABEL: test_bitinsert_v2ptr:
-; AVX2: # %bb.0:
-; AVX2-NEXT: vmovaps %xmm1, %xmm0
-; AVX2-NEXT: retq
+; AVX-LABEL: test_bitinsert_v2ptr:
+; AVX: # %bb.0:
+; AVX-NEXT: vmovaps %xmm1, %xmm0
+; AVX-NEXT: retq
%result = bitinsert <2 x b64> %base, <2 x ptr> %val, <2 x i32> zeroinitializer
ret <2 x b64> %result
}
@@ -277,6 +405,32 @@ define <2 x b128> @test_bitinsert_v2ptr_b128(<2 x b128> %base, <2 x ptr> %val) {
; SSE2-NEXT: movq %rsi, (%rax)
; SSE2-NEXT: retq
;
+; SSE42-LABEL: test_bitinsert_v2ptr_b128:
+; SSE42: # %bb.0:
+; SSE42-NEXT: movq %rdi, %rax
+; SSE42-NEXT: movq %xmm0, %rdi
+; SSE42-NEXT: movq %rdi, %r9
+; SSE42-NEXT: shlq $32, %r9
+; SSE42-NEXT: movl %esi, %esi
+; SSE42-NEXT: orq %r9, %rsi
+; SSE42-NEXT: pextrq $1, %xmm0, %r9
+; SSE42-NEXT: movq %r9, %r10
+; SSE42-NEXT: shlq $32, %r10
+; SSE42-NEXT: movl %ecx, %ecx
+; SSE42-NEXT: orq %r10, %rcx
+; SSE42-NEXT: movabsq $-4294967296, %r10 # imm = 0xFFFFFFFF00000000
+; SSE42-NEXT: andq %r10, %rdx
+; SSE42-NEXT: shrq $32, %rdi
+; SSE42-NEXT: orq %rdx, %rdi
+; SSE42-NEXT: andq %r10, %r8
+; SSE42-NEXT: shrq $32, %r9
+; SSE42-NEXT: orq %r8, %r9
+; SSE42-NEXT: movq %r9, 24(%rax)
+; SSE42-NEXT: movq %rdi, 8(%rax)
+; SSE42-NEXT: movq %rcx, 16(%rax)
+; SSE42-NEXT: movq %rsi, (%rax)
+; SSE42-NEXT: retq
+;
; AVX2-LABEL: test_bitinsert_v2ptr_b128:
; AVX2: # %bb.0:
; AVX2-NEXT: movq %rdi, %rax
@@ -284,17 +438,17 @@ define <2 x b128> @test_bitinsert_v2ptr_b128(<2 x b128> %base, <2 x ptr> %val) {
; AVX2-NEXT: shrq $32, %rdx
; AVX2-NEXT: vmovq %xmm0, %rdi
; AVX2-NEXT: movq %rdi, %r9
-; AVX2-NEXT: shrq $32, %r9
; AVX2-NEXT: vmovd %edx, %xmm1
; AVX2-NEXT: vpslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3]
; AVX2-NEXT: vpextrq $1, %xmm1, %rdx
+; AVX2-NEXT: shrq $32, %r9
; AVX2-NEXT: orq %r9, %rdx
; AVX2-NEXT: vpextrq $1, %xmm0, %r9
; AVX2-NEXT: movq %r9, %r10
-; AVX2-NEXT: shrq $32, %r10
; AVX2-NEXT: vmovd %r8d, %xmm0
; AVX2-NEXT: vpslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3]
; AVX2-NEXT: vpextrq $1, %xmm0, %r8
+; AVX2-NEXT: shrq $32, %r10
; AVX2-NEXT: orq %r10, %r8
; AVX2-NEXT: shlq $32, %rdi
; AVX2-NEXT: movl %esi, %esi
@@ -307,6 +461,37 @@ define <2 x b128> @test_bitinsert_v2ptr_b128(<2 x b128> %base, <2 x ptr> %val) {
; AVX2-NEXT: movq %r8, 24(%rax)
; AVX2-NEXT: movq %rdx, 8(%rax)
; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitinsert_v2ptr_b128:
+; AVX512: # %bb.0:
+; AVX512-NEXT: movq %rdi, %rax
+; AVX512-NEXT: shrq $32, %r8
+; AVX512-NEXT: shrq $32, %rdx
+; AVX512-NEXT: vmovq %xmm0, %rdi
+; AVX512-NEXT: movq %rdi, %r9
+; AVX512-NEXT: shrq $32, %r9
+; AVX512-NEXT: vmovd %edx, %xmm1
+; AVX512-NEXT: vpslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3]
+; AVX512-NEXT: vpextrq $1, %xmm1, %rdx
+; AVX512-NEXT: orq %r9, %rdx
+; AVX512-NEXT: vpextrq $1, %xmm0, %r9
+; AVX512-NEXT: movq %r9, %r10
+; AVX512-NEXT: vmovd %r8d, %xmm0
+; AVX512-NEXT: vpslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3]
+; AVX512-NEXT: vpextrq $1, %xmm0, %r8
+; AVX512-NEXT: shrq $32, %r10
+; AVX512-NEXT: orq %r10, %r8
+; AVX512-NEXT: shlq $32, %rdi
+; AVX512-NEXT: movl %esi, %esi
+; AVX512-NEXT: orq %rdi, %rsi
+; AVX512-NEXT: shlq $32, %r9
+; AVX512-NEXT: movl %ecx, %ecx
+; AVX512-NEXT: orq %r9, %rcx
+; AVX512-NEXT: movq %rcx, 16(%rax)
+; AVX512-NEXT: movq %rsi, (%rax)
+; AVX512-NEXT: movq %r8, 24(%rax)
+; AVX512-NEXT: movq %rdx, 8(%rax)
+; AVX512-NEXT: retq
%result = bitinsert <2 x b128> %base, <2 x ptr> %val, <2 x i32> splat (i32 32)
ret <2 x b128> %result
}
@@ -338,6 +523,31 @@ define <2 x b128> @test_bitinsert_v2ptr_addrspace3_const(<2 x b128> %base, <2 x
; SSE2-NEXT: movq %rsi, (%rax)
; SSE2-NEXT: retq
;
+; SSE42-LABEL: test_bitinsert_v2ptr_addrspace3_const:
+; SSE42: # %bb.0:
+; SSE42-NEXT: movq %rdi, %rax
+; SSE42-NEXT: movq %xmm0, %rdi
+; SSE42-NEXT: movq %rdi, %r9
+; SSE42-NEXT: shlq $16, %r9
+; SSE42-NEXT: movzwl %si, %esi
+; SSE42-NEXT: pextrq $1, %xmm0, %r10
+; SSE42-NEXT: orq %r9, %rsi
+; SSE42-NEXT: movq %r10, %r9
+; SSE42-NEXT: shlq $16, %r9
+; SSE42-NEXT: movzwl %cx, %ecx
+; SSE42-NEXT: orq %r9, %rcx
+; SSE42-NEXT: andq $-65536, %rdx # imm = 0xFFFF0000
+; SSE42-NEXT: shrq $48, %rdi
+; SSE42-NEXT: orq %rdx, %rdi
+; SSE42-NEXT: andq $-65536, %r8 # imm = 0xFFFF0000
+; SSE42-NEXT: shrq $48, %r10
+; SSE42-NEXT: orq %r8, %r10
+; SSE42-NEXT: movq %r10, 24(%rax)
+; SSE42-NEXT: movq %rdi, 8(%rax)
+; SSE42-NEXT: movq %rcx, 16(%rax)
+; SSE42-NEXT: movq %rsi, (%rax)
+; SSE42-NEXT: retq
+;
; AVX2-LABEL: test_bitinsert_v2ptr_addrspace3_const:
; AVX2: # %bb.0:
; AVX2-NEXT: movq %rdi, %rax
@@ -345,7 +555,6 @@ define <2 x b128> @test_bitinsert_v2ptr_addrspace3_const(<2 x b128> %base, <2 x
; AVX2-NEXT: shrq $48, %rdi
; AVX2-NEXT: movq %r8, %r9
; AVX2-NEXT: shrq $32, %r9
-; AVX2-NEXT: shrl $16, %r8d
; AVX2-NEXT: movq %rdx, %r10
; AVX2-NEXT: shrq $48, %r10
; AVX2-NEXT: movq %rdx, %r11
@@ -356,6 +565,7 @@ define <2 x b128> @test_bitinsert_v2ptr_addrspace3_const(<2 x b128> %base, <2 x
; AVX2-NEXT: vpinsrw $5, %edx, %xmm1, %xmm1
; AVX2-NEXT: vpinsrw $6, %r11d, %xmm1, %xmm1
; AVX2-NEXT: vpinsrw $7, %r10d, %xmm1, %xmm1
+; AVX2-NEXT: shrl $16, %r8d
; AVX2-NEXT: vpextrq $1, %xmm1, %r10
; AVX2-NEXT: vmovq %xmm0, %r11
; AVX2-NEXT: movq %r11, %rdx
@@ -380,6 +590,49 @@ define <2 x b128> @test_bitinsert_v2ptr_addrspace3_const(<2 x b128> %base, <2 x
; AVX2-NEXT: movq %r9, 24(%rax)
; AVX2-NEXT: movq %rdx, 8(%rax)
; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitinsert_v2ptr_addrspace3_const:
+; AVX512: # %bb.0:
+; AVX512-NEXT: movq %rdi, %rax
+; AVX512-NEXT: movq %r8, %rdi
+; AVX512-NEXT: shrq $48, %rdi
+; AVX512-NEXT: movq %r8, %r9
+; AVX512-NEXT: shrq $32, %r9
+; AVX512-NEXT: shrl $16, %r8d
+; AVX512-NEXT: movq %rdx, %r10
+; AVX512-NEXT: movq %rdx, %r11
+; AVX512-NEXT: shrq $32, %r11
+; AVX512-NEXT: shrl $16, %edx
+; AVX512-NEXT: movzwl %si, %esi
+; AVX512-NEXT: vmovd %esi, %xmm1
+; AVX512-NEXT: vpinsrw $5, %edx, %xmm1, %xmm1
+; AVX512-NEXT: vpinsrw $6, %r11d, %xmm1, %xmm1
+; AVX512-NEXT: shrq $48, %r10
+; AVX512-NEXT: vpinsrw $7, %r10d, %xmm1, %xmm1
+; AVX512-NEXT: vpextrq $1, %xmm1, %r10
+; AVX512-NEXT: vmovq %xmm0, %r11
+; AVX512-NEXT: movq %r11, %rdx
+; AVX512-NEXT: shrq $48, %rdx
+; AVX512-NEXT: orq %r10, %rdx
+; AVX512-NEXT: movzwl %cx, %ecx
+; AVX512-NEXT: vmovd %ecx, %xmm1
+; AVX512-NEXT: vpinsrw $5, %r8d, %xmm1, %xmm1
+; AVX512-NEXT: vpinsrw $6, %r9d, %xmm1, %xmm1
+; AVX512-NEXT: vpinsrw $7, %edi, %xmm1, %xmm1
+; AVX512-NEXT: vpextrq $1, %xmm1, %rdi
+; AVX512-NEXT: vpextrq $1, %xmm0, %r8
+; AVX512-NEXT: movq %r8, %r9
+; AVX512-NEXT: shrq $48, %r9
+; AVX512-NEXT: orq %rdi, %r9
+; AVX512-NEXT: shlq $16, %r11
+; AVX512-NEXT: orq %r11, %rsi
+; AVX512-NEXT: shlq $16, %r8
+; AVX512-NEXT: orq %r8, %rcx
+; AVX512-NEXT: movq %rcx, 16(%rax)
+; AVX512-NEXT: movq %rsi, (%rax)
+; AVX512-NEXT: movq %r9, 24(%rax)
+; AVX512-NEXT: movq %rdx, 8(%rax)
+; AVX512-NEXT: retq
%result = bitinsert <2 x b128> %base, <2 x ptr addrspace(3)> %val, <2 x i32> splat (i32 16)
ret <2 x b128> %result
}
@@ -445,6 +698,41 @@ define <16 x b8> @test_bitinsert_v16i4_var(<16 x b8> %base, <16 x i4> %val, <16
; SSE2-NEXT: movdqa %xmm1, %xmm0
; SSE2-NEXT: retq
;
+; SSE42-LABEL: test_bitinsert_v16i4_var:
+; SSE42: # %bb.0:
+; SSE42-NEXT: movdqa %xmm0, %xmm6
+; SSE42-NEXT: movdqa {{.*#+}} xmm0 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; SSE42-NEXT: pand %xmm0, %xmm5
+; SSE42-NEXT: pand %xmm0, %xmm4
+; SSE42-NEXT: packusdw %xmm5, %xmm4
+; SSE42-NEXT: pand %xmm0, %xmm3
+; SSE42-NEXT: pand %xmm0, %xmm2
+; SSE42-NEXT: packusdw %xmm3, %xmm2
+; SSE42-NEXT: packuswb %xmm4, %xmm2
+; SSE42-NEXT: movdqa %xmm2, %xmm0
+; SSE42-NEXT: psllw $5, %xmm0
+; SSE42-NEXT: movdqa %xmm0, %xmm3
+; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE42-NEXT: paddb %xmm0, %xmm3
+; SSE42-NEXT: movdqa %xmm1, %xmm4
+; SSE42-NEXT: psllw $4, %xmm4
+; SSE42-NEXT: pblendvb %xmm0, %xmm4, %xmm1
+; SSE42-NEXT: movdqa %xmm1, %xmm4
+; SSE42-NEXT: paddb %xmm1, %xmm4
+; SSE42-NEXT: paddb %xmm4, %xmm4
+; SSE42-NEXT: movdqa %xmm3, %xmm0
+; SSE42-NEXT: pblendvb %xmm0, %xmm4, %xmm1
+; SSE42-NEXT: movdqa %xmm1, %xmm4
+; SSE42-NEXT: paddb %xmm1, %xmm4
+; SSE42-NEXT: paddb %xmm3, %xmm3
+; SSE42-NEXT: movdqa %xmm3, %xmm0
+; SSE42-NEXT: pblendvb %xmm0, %xmm4, %xmm1
+; SSE42-NEXT: movq {{.*#+}} xmm0 = [15,30,60,120,240,224,192,128,0,0,0,0,0,0,0,0]
+; SSE42-NEXT: pshufb %xmm2, %xmm0
+; SSE42-NEXT: pandn %xmm6, %xmm0
+; SSE42-NEXT: por %xmm1, %xmm0
+; SSE42-NEXT: retq
+;
; AVX2-LABEL: test_bitinsert_v16i4_var:
; AVX2: # %bb.0:
; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
@@ -455,8 +743,8 @@ define <16 x b8> @test_bitinsert_v16i4_var(<16 x b8> %base, <16 x i4> %val, <16
; AVX2-NEXT: vpackuswb %xmm3, %xmm2, %xmm2
; AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,2,1,3]
; AVX2-NEXT: vpsllw $5, %xmm2, %xmm3
-; AVX2-NEXT: vpaddb %xmm3, %xmm3, %xmm4
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT: vpaddb %xmm3, %xmm3, %xmm4
; AVX2-NEXT: vpsllw $4, %xmm1, %xmm5
; AVX2-NEXT: vpblendvb %xmm3, %xmm5, %xmm1, %xmm1
; AVX2-NEXT: vpaddb %xmm1, %xmm1, %xmm3
@@ -471,6 +759,21 @@ define <16 x b8> @test_bitinsert_v16i4_var(<16 x b8> %base, <16 x i4> %val, <16
; AVX2-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitinsert_v16i4_var:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpmovdb %zmm2, %xmm3
+; AVX512-NEXT: vmovq {{.*#+}} xmm4 = [15,30,60,120,240,224,192,128,0,0,0,0,0,0,0,0]
+; AVX512-NEXT: vpmovdw %zmm2, %ymm2
+; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm2, %ymm2
+; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm1, %xmm1
+; AVX512-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512-NEXT: vpsllvw %ymm2, %ymm1, %ymm1
+; AVX512-NEXT: vpmovwb %ymm1, %xmm1
+; AVX512-NEXT: vpternlogq {{.*#+}} xmm0 = (xmm0 & ~xmm3) | xmm1
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%result = bitinsert <16 x b8> %base, <16 x i4> %val, <16 x i32> %off
ret <16 x b8> %result
}
>From b7bb0c9410e9f564792eea2c6d3e860e71b81336 Mon Sep 17 00:00:00 2001
From: Pedro Lobo <pedro.lobo at tecnico.ulisboa.pt>
Date: Thu, 12 Nov 2026 12:37:32 +0000
Subject: [PATCH 3/3] [SelectionDAG] Split vector `bitinsert` and `bitextract`
tests
---
.../test/CodeGen/AArch64/bitextract-vector.ll | 55 ++++
...textract-vector.ll => bitinsert-vector.ll} | 61 +----
llvm/test/CodeGen/AArch64/sve-bitextract.ll | 12 +
...tinsert-bitextract.ll => sve-bitinsert.ll} | 10 -
llvm/test/CodeGen/AMDGPU/bitextract-vector.ll | 188 ++++++++++++++
...textract-vector.ll => bitinsert-vector.ll} | 184 --------------
llvm/test/CodeGen/RISCV/rvv/bitextract.ll | 17 ++
.../{bitinsert-bitextract.ll => bitinsert.ll} | 14 --
.../RISCV/rvv/fixed-vectors-bitextract.ll | 70 ++++++
...textract.ll => fixed-vectors-bitinsert.ll} | 67 -----
llvm/test/CodeGen/X86/bitextract-vector.ll | 235 ++++++++++++++++++
...textract-vector.ll => bitinsert-vector.ll} | 230 -----------------
12 files changed, 581 insertions(+), 562 deletions(-)
create mode 100644 llvm/test/CodeGen/AArch64/bitextract-vector.ll
rename llvm/test/CodeGen/AArch64/{bitinsert-bitextract-vector.ll => bitinsert-vector.ll} (61%)
create mode 100644 llvm/test/CodeGen/AArch64/sve-bitextract.ll
rename llvm/test/CodeGen/AArch64/{sve-bitinsert-bitextract.ll => sve-bitinsert.ll} (66%)
create mode 100644 llvm/test/CodeGen/AMDGPU/bitextract-vector.ll
rename llvm/test/CodeGen/AMDGPU/{bitinsert-bitextract-vector.ll => bitinsert-vector.ll} (80%)
create mode 100644 llvm/test/CodeGen/RISCV/rvv/bitextract.ll
rename llvm/test/CodeGen/RISCV/rvv/{bitinsert-bitextract.ll => bitinsert.ll} (60%)
create mode 100644 llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitextract.ll
rename llvm/test/CodeGen/RISCV/rvv/{fixed-vectors-bitinsert-bitextract.ll => fixed-vectors-bitinsert.ll} (72%)
create mode 100644 llvm/test/CodeGen/X86/bitextract-vector.ll
rename llvm/test/CodeGen/X86/{bitinsert-bitextract-vector.ll => bitinsert-vector.ll} (70%)
diff --git a/llvm/test/CodeGen/AArch64/bitextract-vector.ll b/llvm/test/CodeGen/AArch64/bitextract-vector.ll
new file mode 100644
index 0000000000000..e43dd11e7bebd
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/bitextract-vector.ll
@@ -0,0 +1,55 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=aarch64 < %s | FileCheck %s
+
+define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
+; CHECK-LABEL: test_bitextract_v4i8_const:
+; CHECK: // %bb.0:
+; CHECK-NEXT: shrn v0.4h, v0.4s, #8
+; CHECK-NEXT: ret
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
+ ret <4 x i8> %result
+}
+
+define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
+; CHECK-LABEL: test_bitextract_v4i8_var:
+; CHECK: // %bb.0:
+; CHECK-NEXT: neg v1.4s, v1.4s
+; CHECK-NEXT: ushl v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: xtn v0.4h, v0.4s
+; CHECK-NEXT: ret
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
+ ret <4 x i8> %result
+}
+
+define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
+; CHECK-LABEL: test_bitextract_v2f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: adrp x8, .LCPI2_0
+; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI2_0]
+; CHECK-NEXT: ushl v0.2d, v0.2d, v1.2d
+; CHECK-NEXT: xtn v0.2s, v0.2d
+; CHECK-NEXT: ret
+ %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
+ ret <2 x float> %result
+}
+
+define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
+; CHECK-LABEL: test_bitextract_v2ptr:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ret
+ %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
+ ret <2 x ptr> %result
+}
+
+define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
+; CHECK-LABEL: test_bitextract_v16i4_var:
+; CHECK: // %bb.0:
+; CHECK-NEXT: uzp1 v3.8h, v3.8h, v4.8h
+; CHECK-NEXT: uzp1 v1.8h, v1.8h, v2.8h
+; CHECK-NEXT: uzp1 v1.16b, v1.16b, v3.16b
+; CHECK-NEXT: neg v1.16b, v1.16b
+; CHECK-NEXT: ushl v0.16b, v0.16b, v1.16b
+; CHECK-NEXT: ret
+ %result = bitextract <16 x i4>, <16 x b8> %src, <16 x i32> %off
+ ret <16 x i4> %result
+}
diff --git a/llvm/test/CodeGen/AArch64/bitinsert-bitextract-vector.ll b/llvm/test/CodeGen/AArch64/bitinsert-vector.ll
similarity index 61%
rename from llvm/test/CodeGen/AArch64/bitinsert-bitextract-vector.ll
rename to llvm/test/CodeGen/AArch64/bitinsert-vector.ll
index ade116ef126b6..0d6f3a9fc18e9 100644
--- a/llvm/test/CodeGen/AArch64/bitinsert-bitextract-vector.ll
+++ b/llvm/test/CodeGen/AArch64/bitinsert-vector.ll
@@ -1,59 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc -mtriple=aarch64 < %s | FileCheck %s
-define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
-; CHECK-LABEL: test_bitextract_v4i8_const:
-; CHECK: // %bb.0:
-; CHECK-NEXT: shrn v0.4h, v0.4s, #8
-; CHECK-NEXT: ret
- %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
- ret <4 x i8> %result
-}
-
-define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
-; CHECK-LABEL: test_bitextract_v4i8_var:
-; CHECK: // %bb.0:
-; CHECK-NEXT: neg v1.4s, v1.4s
-; CHECK-NEXT: ushl v0.4s, v0.4s, v1.4s
-; CHECK-NEXT: xtn v0.4h, v0.4s
-; CHECK-NEXT: ret
- %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
- ret <4 x i8> %result
-}
-
-define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
-; CHECK-LABEL: test_bitextract_v2f32:
-; CHECK: // %bb.0:
-; CHECK-NEXT: adrp x8, .LCPI2_0
-; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI2_0]
-; CHECK-NEXT: ushl v0.2d, v0.2d, v1.2d
-; CHECK-NEXT: xtn v0.2s, v0.2d
-; CHECK-NEXT: ret
- %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
- ret <2 x float> %result
-}
-
-define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
-; CHECK-LABEL: test_bitextract_v2ptr:
-; CHECK: // %bb.0:
-; CHECK-NEXT: ret
- %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
- ret <2 x ptr> %result
-}
-
-define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
-; CHECK-LABEL: test_bitextract_v16i4_var:
-; CHECK: // %bb.0:
-; CHECK-NEXT: uzp1 v3.8h, v3.8h, v4.8h
-; CHECK-NEXT: uzp1 v1.8h, v1.8h, v2.8h
-; CHECK-NEXT: uzp1 v1.16b, v1.16b, v3.16b
-; CHECK-NEXT: neg v1.16b, v1.16b
-; CHECK-NEXT: ushl v0.16b, v0.16b, v1.16b
-; CHECK-NEXT: ret
- %result = bitextract <16 x i4>, <16 x b8> %src, <16 x i32> %off
- ret <16 x i4> %result
-}
-
define <4 x b32> @test_bitinsert_v4i8_const(<4 x b32> %base, <4 x i8> %val) {
; CHECK-LABEL: test_bitinsert_v4i8_const:
; CHECK: // %bb.0:
@@ -84,11 +31,11 @@ define <4 x b32> @test_bitinsert_v4i8_var(<4 x b32> %base, <4 x i8> %val, <4 x i
define <2 x b64> @test_bitinsert_v2f32(<2 x b64> %base, <2 x float> %val) {
; CHECK-LABEL: test_bitinsert_v2f32:
; CHECK: // %bb.0:
-; CHECK-NEXT: adrp x8, .LCPI7_0
+; CHECK-NEXT: adrp x8, .LCPI2_0
; CHECK-NEXT: ushll v1.2d, v1.2s, #0
-; CHECK-NEXT: adrp x9, .LCPI7_1
-; CHECK-NEXT: ldr q2, [x8, :lo12:.LCPI7_0]
-; CHECK-NEXT: ldr q3, [x9, :lo12:.LCPI7_1]
+; CHECK-NEXT: adrp x9, .LCPI2_1
+; CHECK-NEXT: ldr q2, [x8, :lo12:.LCPI2_0]
+; CHECK-NEXT: ldr q3, [x9, :lo12:.LCPI2_1]
; CHECK-NEXT: ushl v1.2d, v1.2d, v2.2d
; CHECK-NEXT: and v0.16b, v0.16b, v3.16b
; CHECK-NEXT: orr v0.16b, v0.16b, v1.16b
diff --git a/llvm/test/CodeGen/AArch64/sve-bitextract.ll b/llvm/test/CodeGen/AArch64/sve-bitextract.ll
new file mode 100644
index 0000000000000..ea208f621e9ea
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve-bitextract.ll
@@ -0,0 +1,12 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=aarch64 -mattr=+sve < %s | FileCheck %s
+
+define <vscale x 4 x i8> @test_bitextract_nxv4i8(<vscale x 4 x b32> %src, <vscale x 4 x i32> %off) {
+; CHECK-LABEL: test_bitextract_nxv4i8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: lsr z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT: ret
+ %result = bitextract <vscale x 4 x i8>, <vscale x 4 x b32> %src, <vscale x 4 x i32> %off
+ ret <vscale x 4 x i8> %result
+}
diff --git a/llvm/test/CodeGen/AArch64/sve-bitinsert-bitextract.ll b/llvm/test/CodeGen/AArch64/sve-bitinsert.ll
similarity index 66%
rename from llvm/test/CodeGen/AArch64/sve-bitinsert-bitextract.ll
rename to llvm/test/CodeGen/AArch64/sve-bitinsert.ll
index 4b3614708cbc2..af37e1d21e1d9 100644
--- a/llvm/test/CodeGen/AArch64/sve-bitinsert-bitextract.ll
+++ b/llvm/test/CodeGen/AArch64/sve-bitinsert.ll
@@ -1,16 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc -mtriple=aarch64 -mattr=+sve < %s | FileCheck %s
-define <vscale x 4 x i8> @test_bitextract_nxv4i8(<vscale x 4 x b32> %src, <vscale x 4 x i32> %off) {
-; CHECK-LABEL: test_bitextract_nxv4i8:
-; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p0.s
-; CHECK-NEXT: lsr z0.s, p0/m, z0.s, z1.s
-; CHECK-NEXT: ret
- %result = bitextract <vscale x 4 x i8>, <vscale x 4 x b32> %src, <vscale x 4 x i32> %off
- ret <vscale x 4 x i8> %result
-}
-
define <vscale x 4 x b32> @test_bitinsert_nxv4i8(<vscale x 4 x b32> %base, <vscale x 4 x i8> %val, <vscale x 4 x i32> %off) {
; CHECK-LABEL: test_bitinsert_nxv4i8:
; CHECK: // %bb.0:
diff --git a/llvm/test/CodeGen/AMDGPU/bitextract-vector.ll b/llvm/test/CodeGen/AMDGPU/bitextract-vector.ll
new file mode 100644
index 0000000000000..ad53487025699
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/bitextract-vector.ll
@@ -0,0 +1,188 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=amdgpu9.00-amd-amdhsa < %s | FileCheck -check-prefix=GFX9 %s
+; RUN: llc -mtriple=amdgpu12.00-amd-amdhsa -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-TRUE16 %s
+; RUN: llc -mtriple=amdgpu12.00-amd-amdhsa -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-FAKE16 %s
+
+define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
+; GFX9-LABEL: test_bitextract_v4i8_const:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v4, 8
+; GFX9-NEXT: v_lshrrev_b32_sdwa v3, v4, v3 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_lshrrev_b32_sdwa v1, v4, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_or_b32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX9-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX9-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX9-NEXT: v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX9-NEXT: v_lshrrev_b32_e32 v3, 24, v3
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-TRUE16-LABEL: test_bitextract_v4i8_const:
+; GFX12-TRUE16: ; %bb.0:
+; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, 8, v3
+; GFX12-TRUE16-NEXT: v_lshrrev_b16 v0.h, 8, v2.l
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX12-TRUE16-NEXT: v_lshrrev_b16 v0.l, 8, v0.l
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v2.l, 8, v3.l
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v1.l, 8, v1.l
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_or_b16 v2.l, v0.h, v2.l
+; GFX12-TRUE16-NEXT: v_or_b16 v0.l, v0.l, v1.l
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.h, v2.l
+; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, 24, v3
+; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_bitextract_v4i8_const:
+; GFX12-FAKE16: ; %bb.0:
+; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 8, v3
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX12-FAKE16-NEXT: v_lshrrev_b16 v2, 8, v2
+; GFX12-FAKE16-NEXT: v_lshrrev_b16 v0, 8, v0
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v3, 8, v3
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v1, 8, v1
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v2, v2, v3
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v1
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v2
+; GFX12-FAKE16-NEXT: v_lshl_or_b32 v3, v2, 16, v3
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 8, v3
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 24, v4
+; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
+ ret <4 x i8> %result
+}
+
+define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
+; GFX9-LABEL: test_bitextract_v4i8_var:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_lshrrev_b32_sdwa v3, v7, v3 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_lshrrev_b32_e32 v2, v6, v2
+; GFX9-NEXT: v_lshrrev_b32_sdwa v1, v5, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT: v_lshrrev_b32_e32 v0, v4, v0
+; GFX9-NEXT: v_or_b32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX9-NEXT: v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX9-NEXT: v_lshrrev_b32_e32 v3, 24, v3
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-TRUE16-LABEL: test_bitextract_v4i8_var:
+; GFX12-TRUE16: ; %bb.0:
+; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v7, v3
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v6, v6, v2
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, v5, v1
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v0, v4, v0
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v2.l, 8, v3.l
+; GFX12-TRUE16-NEXT: v_and_b16 v2.h, v6.l, 0xff
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-TRUE16-NEXT: v_lshlrev_b16 v1.l, 8, v1.l
+; GFX12-TRUE16-NEXT: v_and_b16 v0.l, v0.l, 0xff
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_or_b16 v2.l, v2.h, v2.l
+; GFX12-TRUE16-NEXT: v_or_b16 v0.l, v0.l, v1.l
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.h, v2.l
+; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, 24, v3
+; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_bitextract_v4i8_var:
+; GFX12-FAKE16: ; %bb.0:
+; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v7, v3
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v2, v6, v2
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, v5, v1
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v0, v4, v0
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v3, 8, v3
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v2, 0xff, v2
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-FAKE16-NEXT: v_lshlrev_b16 v1, 8, v1
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 0xff, v0
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v2, v2, v3
+; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v1
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-FAKE16-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v2
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX12-FAKE16-NEXT: v_lshl_or_b32 v3, v2, 16, v3
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 8, v3
+; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3)
+; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 24, v4
+; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
+ ret <4 x i8> %result
+}
+
+define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
+; GFX9-LABEL: test_bitextract_v2f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v1, v3
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: test_bitextract_v2f32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mov_b32_e32 v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
+ ret <2 x float> %result
+}
+
+define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
+; GFX9-LABEL: test_bitextract_v2ptr:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: test_bitextract_v2ptr:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
+ ret <2 x ptr> %result
+}
diff --git a/llvm/test/CodeGen/AMDGPU/bitinsert-bitextract-vector.ll b/llvm/test/CodeGen/AMDGPU/bitinsert-vector.ll
similarity index 80%
rename from llvm/test/CodeGen/AMDGPU/bitinsert-bitextract-vector.ll
rename to llvm/test/CodeGen/AMDGPU/bitinsert-vector.ll
index 9bcbddacf7a1d..4e7472b7cc297 100644
--- a/llvm/test/CodeGen/AMDGPU/bitinsert-bitextract-vector.ll
+++ b/llvm/test/CodeGen/AMDGPU/bitinsert-vector.ll
@@ -3,190 +3,6 @@
; RUN: llc -mtriple=amdgpu12.00-amd-amdhsa -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-TRUE16 %s
; RUN: llc -mtriple=amdgpu12.00-amd-amdhsa -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-FAKE16 %s
-define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
-; GFX9-LABEL: test_bitextract_v4i8_const:
-; GFX9: ; %bb.0:
-; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT: v_mov_b32_e32 v4, 8
-; GFX9-NEXT: v_lshrrev_b32_sdwa v3, v4, v3 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GFX9-NEXT: v_lshrrev_b32_sdwa v1, v4, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GFX9-NEXT: v_or_b32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
-; GFX9-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
-; GFX9-NEXT: v_lshlrev_b32_e32 v3, 16, v2
-; GFX9-NEXT: v_lshl_or_b32 v1, v2, 16, v1
-; GFX9-NEXT: v_lshrrev_b32_e32 v1, 8, v1
-; GFX9-NEXT: v_lshrrev_b32_e32 v3, 24, v3
-; GFX9-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-TRUE16-LABEL: test_bitextract_v4i8_const:
-; GFX12-TRUE16: ; %bb.0:
-; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
-; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
-; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
-; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
-; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, 8, v3
-; GFX12-TRUE16-NEXT: v_lshrrev_b16 v0.h, 8, v2.l
-; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
-; GFX12-TRUE16-NEXT: v_lshrrev_b16 v0.l, 8, v0.l
-; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX12-TRUE16-NEXT: v_lshlrev_b16 v2.l, 8, v3.l
-; GFX12-TRUE16-NEXT: v_lshlrev_b16 v1.l, 8, v1.l
-; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX12-TRUE16-NEXT: v_or_b16 v2.l, v0.h, v2.l
-; GFX12-TRUE16-NEXT: v_or_b16 v0.l, v0.l, v1.l
-; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.h, v2.l
-; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v2
-; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
-; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, 24, v3
-; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-FAKE16-LABEL: test_bitextract_v4i8_const:
-; GFX12-FAKE16: ; %bb.0:
-; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
-; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
-; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
-; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
-; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 8, v3
-; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
-; GFX12-FAKE16-NEXT: v_lshrrev_b16 v2, 8, v2
-; GFX12-FAKE16-NEXT: v_lshrrev_b16 v0, 8, v0
-; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX12-FAKE16-NEXT: v_lshlrev_b16 v3, 8, v3
-; GFX12-FAKE16-NEXT: v_lshlrev_b16 v1, 8, v1
-; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX12-FAKE16-NEXT: v_or_b32_e32 v2, v2, v3
-; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v1
-; GFX12-FAKE16-NEXT: v_or_b32_e32 v0, v0, v1
-; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v2
-; GFX12-FAKE16-NEXT: v_lshl_or_b32 v3, v2, 16, v3
-; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 8, v3
-; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 24, v4
-; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
- %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
- ret <4 x i8> %result
-}
-
-define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
-; GFX9-LABEL: test_bitextract_v4i8_var:
-; GFX9: ; %bb.0:
-; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT: v_lshrrev_b32_sdwa v3, v7, v3 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GFX9-NEXT: v_lshrrev_b32_e32 v2, v6, v2
-; GFX9-NEXT: v_lshrrev_b32_sdwa v1, v5, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GFX9-NEXT: v_lshrrev_b32_e32 v0, v4, v0
-; GFX9-NEXT: v_or_b32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
-; GFX9-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
-; GFX9-NEXT: v_lshlrev_b32_e32 v3, 16, v2
-; GFX9-NEXT: v_lshl_or_b32 v1, v2, 16, v1
-; GFX9-NEXT: v_lshrrev_b32_e32 v1, 8, v1
-; GFX9-NEXT: v_lshrrev_b32_e32 v3, 24, v3
-; GFX9-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-TRUE16-LABEL: test_bitextract_v4i8_var:
-; GFX12-TRUE16: ; %bb.0:
-; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
-; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
-; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
-; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
-; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v7, v3
-; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v6, v6, v2
-; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, v5, v1
-; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v0, v4, v0
-; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX12-TRUE16-NEXT: v_lshlrev_b16 v2.l, 8, v3.l
-; GFX12-TRUE16-NEXT: v_and_b16 v2.h, v6.l, 0xff
-; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX12-TRUE16-NEXT: v_lshlrev_b16 v1.l, 8, v1.l
-; GFX12-TRUE16-NEXT: v_and_b16 v0.l, v0.l, 0xff
-; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX12-TRUE16-NEXT: v_or_b16 v2.l, v2.h, v2.l
-; GFX12-TRUE16-NEXT: v_or_b16 v0.l, v0.l, v1.l
-; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.h, v2.l
-; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v2
-; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
-; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, 24, v3
-; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-FAKE16-LABEL: test_bitextract_v4i8_var:
-; GFX12-FAKE16: ; %bb.0:
-; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
-; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
-; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
-; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
-; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v7, v3
-; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v2, v6, v2
-; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, v5, v1
-; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v0, v4, v0
-; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX12-FAKE16-NEXT: v_lshlrev_b16 v3, 8, v3
-; GFX12-FAKE16-NEXT: v_and_b32_e32 v2, 0xff, v2
-; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX12-FAKE16-NEXT: v_lshlrev_b16 v1, 8, v1
-; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 0xff, v0
-; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX12-FAKE16-NEXT: v_or_b32_e32 v2, v2, v3
-; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v1
-; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX12-FAKE16-NEXT: v_or_b32_e32 v0, v0, v1
-; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v2
-; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX12-FAKE16-NEXT: v_lshl_or_b32 v3, v2, 16, v3
-; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 8, v3
-; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3)
-; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 24, v4
-; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
- %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
- ret <4 x i8> %result
-}
-
-define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
-; GFX9-LABEL: test_bitextract_v2f32:
-; GFX9: ; %bb.0:
-; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT: v_mov_b32_e32 v1, v3
-; GFX9-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-LABEL: test_bitextract_v2f32:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: v_mov_b32_e32 v1, v3
-; GFX12-NEXT: s_setpc_b64 s[30:31]
- %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
- ret <2 x float> %result
-}
-
-define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
-; GFX9-LABEL: test_bitextract_v2ptr:
-; GFX9: ; %bb.0:
-; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-LABEL: test_bitextract_v2ptr:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: s_setpc_b64 s[30:31]
- %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
- ret <2 x ptr> %result
-}
-
define <4 x b32> @test_bitinsert_v4i8_const(<4 x b32> %base, <4 x i8> %val) {
; GFX9-LABEL: test_bitinsert_v4i8_const:
; GFX9: ; %bb.0:
diff --git a/llvm/test/CodeGen/RISCV/rvv/bitextract.ll b/llvm/test/CodeGen/RISCV/rvv/bitextract.ll
new file mode 100644
index 0000000000000..2f690168171be
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/bitextract.ll
@@ -0,0 +1,17 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=riscv32 -mattr=+v < %s | FileCheck %s
+; RUN: llc -mtriple=riscv64 -mattr=+v < %s | FileCheck %s
+
+define <vscale x 4 x i8> @test_bitextract_nxv4i8(<vscale x 4 x b32> %src, <vscale x 4 x i32> %off) {
+; CHECK-LABEL: test_bitextract_nxv4i8:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
+; CHECK-NEXT: vsrl.vv v8, v8, v10
+; CHECK-NEXT: vsetvli zero, zero, e16, m1, ta, ma
+; CHECK-NEXT: vnsrl.wi v10, v8, 0
+; CHECK-NEXT: vsetvli zero, zero, e8, mf2, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v10, 0
+; CHECK-NEXT: ret
+ %result = bitextract <vscale x 4 x i8>, <vscale x 4 x b32> %src, <vscale x 4 x i32> %off
+ ret <vscale x 4 x i8> %result
+}
diff --git a/llvm/test/CodeGen/RISCV/rvv/bitinsert-bitextract.ll b/llvm/test/CodeGen/RISCV/rvv/bitinsert.ll
similarity index 60%
rename from llvm/test/CodeGen/RISCV/rvv/bitinsert-bitextract.ll
rename to llvm/test/CodeGen/RISCV/rvv/bitinsert.ll
index 7ed246cfafaf1..8d9dfe9c0728c 100644
--- a/llvm/test/CodeGen/RISCV/rvv/bitinsert-bitextract.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/bitinsert.ll
@@ -2,20 +2,6 @@
; RUN: llc -mtriple=riscv32 -mattr=+v < %s | FileCheck %s
; RUN: llc -mtriple=riscv64 -mattr=+v < %s | FileCheck %s
-define <vscale x 4 x i8> @test_bitextract_nxv4i8(<vscale x 4 x b32> %src, <vscale x 4 x i32> %off) {
-; CHECK-LABEL: test_bitextract_nxv4i8:
-; CHECK: # %bb.0:
-; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
-; CHECK-NEXT: vsrl.vv v8, v8, v10
-; CHECK-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; CHECK-NEXT: vnsrl.wi v10, v8, 0
-; CHECK-NEXT: vsetvli zero, zero, e8, mf2, ta, ma
-; CHECK-NEXT: vnsrl.wi v8, v10, 0
-; CHECK-NEXT: ret
- %result = bitextract <vscale x 4 x i8>, <vscale x 4 x b32> %src, <vscale x 4 x i32> %off
- ret <vscale x 4 x i8> %result
-}
-
define <vscale x 4 x b32> @test_bitinsert_nxv4i8(<vscale x 4 x b32> %base, <vscale x 4 x i8> %val, <vscale x 4 x i32> %off) {
; CHECK-LABEL: test_bitinsert_nxv4i8:
; CHECK: # %bb.0:
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitextract.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitextract.ll
new file mode 100644
index 0000000000000..006d79d0f9e31
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitextract.ll
@@ -0,0 +1,70 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=riscv32 -mattr=+v < %s | FileCheck %s --check-prefixes=CHECK,RV32
+; RUN: llc -mtriple=riscv64 -mattr=+v < %s | FileCheck %s --check-prefixes=CHECK,RV64
+
+define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
+; CHECK-LABEL: test_bitextract_v4i8_const:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v8, 8
+; CHECK-NEXT: vsetvli zero, zero, e8, mf4, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v8, 0
+; CHECK-NEXT: ret
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
+ ret <4 x i8> %result
+}
+
+define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
+; CHECK-LABEL: test_bitextract_v4i8_var:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-NEXT: vsrl.vv v8, v8, v9
+; CHECK-NEXT: vsetvli zero, zero, e16, mf2, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v8, 0
+; CHECK-NEXT: vsetvli zero, zero, e8, mf4, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v8, 0
+; CHECK-NEXT: ret
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
+ ret <4 x i8> %result
+}
+
+define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
+; CHECK-LABEL: test_bitextract_v2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e64, m1, ta, ma
+; CHECK-NEXT: vid.v v9
+; CHECK-NEXT: vsll.vi v9, v9, 5
+; CHECK-NEXT: vsrl.vv v8, v8, v9
+; CHECK-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
+; CHECK-NEXT: vnsrl.wi v8, v8, 0
+; CHECK-NEXT: ret
+ %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
+ ret <2 x float> %result
+}
+
+define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
+; RV32-LABEL: test_bitextract_v2ptr:
+; RV32: # %bb.0:
+; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV32-NEXT: vnsrl.wi v8, v8, 0
+; RV32-NEXT: ret
+;
+; RV64-LABEL: test_bitextract_v2ptr:
+; RV64: # %bb.0:
+; RV64-NEXT: ret
+ %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
+ ret <2 x ptr> %result
+}
+
+define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
+; CHECK-LABEL: test_bitextract_v16i4_var:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, ma
+; CHECK-NEXT: vnsrl.wi v10, v12, 0
+; CHECK-NEXT: vsetvli zero, zero, e8, m1, ta, ma
+; CHECK-NEXT: vnsrl.wi v9, v10, 0
+; CHECK-NEXT: vsrl.vv v8, v8, v9
+; CHECK-NEXT: ret
+ %result = bitextract <16 x i4>, <16 x b8> %src, <16 x i32> %off
+ ret <16 x i4> %result
+}
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitinsert-bitextract.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitinsert.ll
similarity index 72%
rename from llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitinsert-bitextract.ll
rename to llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitinsert.ll
index 7b38d23cf3394..c0367d92ce449 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitinsert-bitextract.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-bitinsert.ll
@@ -2,73 +2,6 @@
; RUN: llc -mtriple=riscv32 -mattr=+v < %s | FileCheck %s --check-prefixes=CHECK,RV32
; RUN: llc -mtriple=riscv64 -mattr=+v < %s | FileCheck %s --check-prefixes=CHECK,RV64
-define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
-; CHECK-LABEL: test_bitextract_v4i8_const:
-; CHECK: # %bb.0:
-; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, ma
-; CHECK-NEXT: vnsrl.wi v8, v8, 8
-; CHECK-NEXT: vsetvli zero, zero, e8, mf4, ta, ma
-; CHECK-NEXT: vnsrl.wi v8, v8, 0
-; CHECK-NEXT: ret
- %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
- ret <4 x i8> %result
-}
-
-define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
-; CHECK-LABEL: test_bitextract_v4i8_var:
-; CHECK: # %bb.0:
-; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, ma
-; CHECK-NEXT: vsrl.vv v8, v8, v9
-; CHECK-NEXT: vsetvli zero, zero, e16, mf2, ta, ma
-; CHECK-NEXT: vnsrl.wi v8, v8, 0
-; CHECK-NEXT: vsetvli zero, zero, e8, mf4, ta, ma
-; CHECK-NEXT: vnsrl.wi v8, v8, 0
-; CHECK-NEXT: ret
- %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
- ret <4 x i8> %result
-}
-
-define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
-; CHECK-LABEL: test_bitextract_v2f32:
-; CHECK: # %bb.0:
-; CHECK-NEXT: vsetivli zero, 2, e64, m1, ta, ma
-; CHECK-NEXT: vid.v v9
-; CHECK-NEXT: vsll.vi v9, v9, 5
-; CHECK-NEXT: vsrl.vv v8, v8, v9
-; CHECK-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
-; CHECK-NEXT: vnsrl.wi v8, v8, 0
-; CHECK-NEXT: ret
- %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
- ret <2 x float> %result
-}
-
-define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
-; RV32-LABEL: test_bitextract_v2ptr:
-; RV32: # %bb.0:
-; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
-; RV32-NEXT: vnsrl.wi v8, v8, 0
-; RV32-NEXT: ret
-;
-; RV64-LABEL: test_bitextract_v2ptr:
-; RV64: # %bb.0:
-; RV64-NEXT: ret
- %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
- ret <2 x ptr> %result
-}
-
-define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
-; CHECK-LABEL: test_bitextract_v16i4_var:
-; CHECK: # %bb.0:
-; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, ma
-; CHECK-NEXT: vnsrl.wi v10, v12, 0
-; CHECK-NEXT: vsetvli zero, zero, e8, m1, ta, ma
-; CHECK-NEXT: vnsrl.wi v9, v10, 0
-; CHECK-NEXT: vsrl.vv v8, v8, v9
-; CHECK-NEXT: ret
- %result = bitextract <16 x i4>, <16 x b8> %src, <16 x i32> %off
- ret <16 x i4> %result
-}
-
define <4 x b32> @test_bitinsert_v4i8_const(<4 x b32> %base, <4 x i8> %val) {
; CHECK-LABEL: test_bitinsert_v4i8_const:
; CHECK: # %bb.0:
diff --git a/llvm/test/CodeGen/X86/bitextract-vector.ll b/llvm/test/CodeGen/X86/bitextract-vector.ll
new file mode 100644
index 0000000000000..436209cde2824
--- /dev/null
+++ b/llvm/test/CodeGen/X86/bitextract-vector.ll
@@ -0,0 +1,235 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=x86_64-- -mcpu=x86-64 < %s | FileCheck %s --check-prefixes=SSE,SSE2
+; RUN: llc -mtriple=x86_64-- -mcpu=x86-64-v2 < %s | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: llc -mtriple=x86_64-- -mcpu=x86-64-v3 < %s | FileCheck %s --check-prefixes=AVX,AVX2
+; RUN: llc -mtriple=x86_64-- -mcpu=x86-64-v4 < %s | FileCheck %s --check-prefixes=AVX,AVX512
+
+define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
+; SSE2-LABEL: test_bitextract_v4i8_const:
+; SSE2: # %bb.0:
+; SSE2-NEXT: psrld $8, %xmm0
+; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: packuswb %xmm0, %xmm0
+; SSE2-NEXT: packuswb %xmm0, %xmm0
+; SSE2-NEXT: retq
+;
+; SSE42-LABEL: test_bitextract_v4i8_const:
+; SSE42: # %bb.0:
+; SSE42-NEXT: pshufb {{.*#+}} xmm0 = xmm0[1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u]
+; SSE42-NEXT: retq
+;
+; AVX2-LABEL: test_bitextract_v4i8_const:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitextract_v4i8_const:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpsrld $8, %xmm0, %xmm0
+; AVX512-NEXT: vpmovdb %xmm0, %xmm0
+; AVX512-NEXT: retq
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
+ ret <4 x i8> %result
+}
+
+define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
+; SSE2-LABEL: test_bitextract_v4i8_var:
+; SSE2: # %bb.0:
+; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7]
+; SSE2-NEXT: movdqa %xmm0, %xmm3
+; SSE2-NEXT: psrld %xmm2, %xmm3
+; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm1[0,1,1,1,4,5,6,7]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: psrld %xmm4, %xmm2
+; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
+; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm1[2,3,3,3,4,5,6,7]
+; SSE2-NEXT: movdqa %xmm0, %xmm5
+; SSE2-NEXT: psrld %xmm4, %xmm5
+; SSE2-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7]
+; SSE2-NEXT: psrld %xmm1, %xmm0
+; SSE2-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm5[1]
+; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm0[0,3]
+; SSE2-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: packuswb %xmm2, %xmm2
+; SSE2-NEXT: packuswb %xmm2, %xmm2
+; SSE2-NEXT: movdqa %xmm2, %xmm0
+; SSE2-NEXT: retq
+;
+; SSE42-LABEL: test_bitextract_v4i8_var:
+; SSE42: # %bb.0:
+; SSE42-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7]
+; SSE42-NEXT: movdqa %xmm0, %xmm3
+; SSE42-NEXT: psrld %xmm2, %xmm3
+; SSE42-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
+; SSE42-NEXT: pshuflw {{.*#+}} xmm4 = xmm2[2,3,3,3,4,5,6,7]
+; SSE42-NEXT: movdqa %xmm0, %xmm5
+; SSE42-NEXT: psrld %xmm4, %xmm5
+; SSE42-NEXT: pblendw {{.*#+}} xmm5 = xmm3[0,1,2,3],xmm5[4,5,6,7]
+; SSE42-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7]
+; SSE42-NEXT: movdqa %xmm0, %xmm3
+; SSE42-NEXT: psrld %xmm1, %xmm3
+; SSE42-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,1,1,1,4,5,6,7]
+; SSE42-NEXT: psrld %xmm1, %xmm0
+; SSE42-NEXT: pblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7]
+; SSE42-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3],xmm0[4,5],xmm5[6,7]
+; SSE42-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; SSE42-NEXT: retq
+;
+; AVX2-LABEL: test_bitextract_v4i8_var:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitextract_v4i8_var:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0
+; AVX512-NEXT: vpmovdb %xmm0, %xmm0
+; AVX512-NEXT: retq
+ %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
+ ret <4 x i8> %result
+}
+
+define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
+; SSE2-LABEL: test_bitextract_v2f32:
+; SSE2: # %bb.0:
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlq $32, %xmm1
+; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
+; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-NEXT: retq
+;
+; SSE42-LABEL: test_bitextract_v2f32:
+; SSE42: # %bb.0:
+; SSE42-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,12,13,14,15,12,13,14,15],zero,zero,zero,zero
+; SSE42-NEXT: retq
+;
+; AVX-LABEL: test_bitextract_v2f32:
+; AVX: # %bb.0:
+; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,12,13,14,15,12,13,14,15],zero,zero,zero,zero
+; AVX-NEXT: retq
+ %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
+ ret <2 x float> %result
+}
+
+define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
+; SSE-LABEL: test_bitextract_v2ptr:
+; SSE: # %bb.0:
+; SSE-NEXT: retq
+;
+; AVX-LABEL: test_bitextract_v2ptr:
+; AVX: # %bb.0:
+; AVX-NEXT: retq
+ %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
+ ret <2 x ptr> %result
+}
+
+define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
+; SSE2-LABEL: test_bitextract_v16i4_var:
+; SSE2: # %bb.0:
+; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; SSE2-NEXT: pand %xmm5, %xmm4
+; SSE2-NEXT: pand %xmm5, %xmm3
+; SSE2-NEXT: packuswb %xmm4, %xmm3
+; SSE2-NEXT: pand %xmm5, %xmm2
+; SSE2-NEXT: pand %xmm5, %xmm1
+; SSE2-NEXT: packuswb %xmm2, %xmm1
+; SSE2-NEXT: packuswb %xmm3, %xmm1
+; SSE2-NEXT: psllw $5, %xmm1
+; SSE2-NEXT: pxor %xmm2, %xmm2
+; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
+; SSE2-NEXT: movdqa %xmm2, %xmm3
+; SSE2-NEXT: pandn %xmm0, %xmm3
+; SSE2-NEXT: psrlw $4, %xmm0
+; SSE2-NEXT: pand %xmm2, %xmm0
+; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: por %xmm3, %xmm0
+; SSE2-NEXT: paddb %xmm1, %xmm1
+; SSE2-NEXT: pxor %xmm2, %xmm2
+; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
+; SSE2-NEXT: movdqa %xmm2, %xmm3
+; SSE2-NEXT: pandn %xmm0, %xmm3
+; SSE2-NEXT: psrlw $2, %xmm0
+; SSE2-NEXT: pand %xmm2, %xmm0
+; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: pxor %xmm2, %xmm2
+; SSE2-NEXT: por %xmm3, %xmm0
+; SSE2-NEXT: paddb %xmm1, %xmm1
+; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
+; SSE2-NEXT: movdqa %xmm2, %xmm1
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: psrlw $1, %xmm0
+; SSE2-NEXT: pand %xmm2, %xmm0
+; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: por %xmm1, %xmm0
+; SSE2-NEXT: retq
+;
+; SSE42-LABEL: test_bitextract_v16i4_var:
+; SSE42: # %bb.0:
+; SSE42-NEXT: movdqa %xmm0, %xmm5
+; SSE42-NEXT: movdqa {{.*#+}} xmm0 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; SSE42-NEXT: pand %xmm0, %xmm4
+; SSE42-NEXT: pand %xmm0, %xmm3
+; SSE42-NEXT: packusdw %xmm4, %xmm3
+; SSE42-NEXT: pand %xmm0, %xmm2
+; SSE42-NEXT: pand %xmm1, %xmm0
+; SSE42-NEXT: packusdw %xmm2, %xmm0
+; SSE42-NEXT: packuswb %xmm3, %xmm0
+; SSE42-NEXT: psllw $5, %xmm0
+; SSE42-NEXT: movdqa %xmm0, %xmm1
+; SSE42-NEXT: movdqa %xmm5, %xmm2
+; SSE42-NEXT: psrlw $4, %xmm2
+; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE42-NEXT: paddb %xmm0, %xmm1
+; SSE42-NEXT: pblendvb %xmm0, %xmm2, %xmm5
+; SSE42-NEXT: movdqa %xmm5, %xmm2
+; SSE42-NEXT: psrlw $2, %xmm2
+; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE42-NEXT: movdqa %xmm1, %xmm0
+; SSE42-NEXT: pblendvb %xmm0, %xmm2, %xmm5
+; SSE42-NEXT: movdqa %xmm5, %xmm2
+; SSE42-NEXT: psrlw $1, %xmm2
+; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE42-NEXT: paddb %xmm1, %xmm1
+; SSE42-NEXT: movdqa %xmm1, %xmm0
+; SSE42-NEXT: pblendvb %xmm0, %xmm2, %xmm5
+; SSE42-NEXT: movdqa %xmm5, %xmm0
+; SSE42-NEXT: retq
+;
+; AVX2-LABEL: test_bitextract_v16i4_var:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX2-NEXT: vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT: vpand %ymm3, %ymm1, %ymm1
+; AVX2-NEXT: vpackusdw %ymm2, %ymm1, %ymm1
+; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT: vpackuswb %xmm2, %xmm1, %xmm1
+; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,1,3]
+; AVX2-NEXT: vpsllw $5, %xmm1, %xmm1
+; AVX2-NEXT: vpsrlw $4, %xmm0, %xmm2
+; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
+; AVX2-NEXT: vpaddb %xmm1, %xmm1, %xmm3
+; AVX2-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
+; AVX2-NEXT: vpsrlw $2, %xmm0, %xmm1
+; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vpsrlw $1, %xmm0, %xmm1
+; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT: vpaddb %xmm3, %xmm3, %xmm2
+; AVX2-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vzeroupper
+; AVX2-NEXT: retq
+;
+; AVX512-LABEL: test_bitextract_v16i4_var:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vpmovdw %zmm1, %ymm1
+; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm1
+; AVX512-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0
+; AVX512-NEXT: vpmovwb %ymm0, %xmm0
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
+ %result = bitextract <16 x i4>, <16 x b8> %src, <16 x i32> %off
+ ret <16 x i4> %result
+}
diff --git a/llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll b/llvm/test/CodeGen/X86/bitinsert-vector.ll
similarity index 70%
rename from llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll
rename to llvm/test/CodeGen/X86/bitinsert-vector.ll
index b8bf19a0ebde9..e1513888f81ee 100644
--- a/llvm/test/CodeGen/X86/bitinsert-bitextract-vector.ll
+++ b/llvm/test/CodeGen/X86/bitinsert-vector.ll
@@ -4,236 +4,6 @@
; RUN: llc -mtriple=x86_64-- -mcpu=x86-64-v3 < %s | FileCheck %s --check-prefixes=AVX,AVX2
; RUN: llc -mtriple=x86_64-- -mcpu=x86-64-v4 < %s | FileCheck %s --check-prefixes=AVX,AVX512
-define <4 x i8> @test_bitextract_v4i8_const(<4 x b32> %src) {
-; SSE2-LABEL: test_bitextract_v4i8_const:
-; SSE2: # %bb.0:
-; SSE2-NEXT: psrld $8, %xmm0
-; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: packuswb %xmm0, %xmm0
-; SSE2-NEXT: packuswb %xmm0, %xmm0
-; SSE2-NEXT: retq
-;
-; SSE42-LABEL: test_bitextract_v4i8_const:
-; SSE42: # %bb.0:
-; SSE42-NEXT: pshufb {{.*#+}} xmm0 = xmm0[1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u]
-; SSE42-NEXT: retq
-;
-; AVX2-LABEL: test_bitextract_v4i8_const:
-; AVX2: # %bb.0:
-; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u]
-; AVX2-NEXT: retq
-;
-; AVX512-LABEL: test_bitextract_v4i8_const:
-; AVX512: # %bb.0:
-; AVX512-NEXT: vpsrld $8, %xmm0, %xmm0
-; AVX512-NEXT: vpmovdb %xmm0, %xmm0
-; AVX512-NEXT: retq
- %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> splat (i32 8)
- ret <4 x i8> %result
-}
-
-define <4 x i8> @test_bitextract_v4i8_var(<4 x b32> %src, <4 x i32> %off) {
-; SSE2-LABEL: test_bitextract_v4i8_var:
-; SSE2: # %bb.0:
-; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7]
-; SSE2-NEXT: movdqa %xmm0, %xmm3
-; SSE2-NEXT: psrld %xmm2, %xmm3
-; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm1[0,1,1,1,4,5,6,7]
-; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: psrld %xmm4, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm1[2,3,3,3,4,5,6,7]
-; SSE2-NEXT: movdqa %xmm0, %xmm5
-; SSE2-NEXT: psrld %xmm4, %xmm5
-; SSE2-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7]
-; SSE2-NEXT: psrld %xmm1, %xmm0
-; SSE2-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm5[1]
-; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm0[0,3]
-; SSE2-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
-; SSE2-NEXT: packuswb %xmm2, %xmm2
-; SSE2-NEXT: packuswb %xmm2, %xmm2
-; SSE2-NEXT: movdqa %xmm2, %xmm0
-; SSE2-NEXT: retq
-;
-; SSE42-LABEL: test_bitextract_v4i8_var:
-; SSE42: # %bb.0:
-; SSE42-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7]
-; SSE42-NEXT: movdqa %xmm0, %xmm3
-; SSE42-NEXT: psrld %xmm2, %xmm3
-; SSE42-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
-; SSE42-NEXT: pshuflw {{.*#+}} xmm4 = xmm2[2,3,3,3,4,5,6,7]
-; SSE42-NEXT: movdqa %xmm0, %xmm5
-; SSE42-NEXT: psrld %xmm4, %xmm5
-; SSE42-NEXT: pblendw {{.*#+}} xmm5 = xmm3[0,1,2,3],xmm5[4,5,6,7]
-; SSE42-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7]
-; SSE42-NEXT: movdqa %xmm0, %xmm3
-; SSE42-NEXT: psrld %xmm1, %xmm3
-; SSE42-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,1,1,1,4,5,6,7]
-; SSE42-NEXT: psrld %xmm1, %xmm0
-; SSE42-NEXT: pblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7]
-; SSE42-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3],xmm0[4,5],xmm5[6,7]
-; SSE42-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
-; SSE42-NEXT: retq
-;
-; AVX2-LABEL: test_bitextract_v4i8_var:
-; AVX2: # %bb.0:
-; AVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
-; AVX2-NEXT: retq
-;
-; AVX512-LABEL: test_bitextract_v4i8_var:
-; AVX512: # %bb.0:
-; AVX512-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0
-; AVX512-NEXT: vpmovdb %xmm0, %xmm0
-; AVX512-NEXT: retq
- %result = bitextract <4 x i8>, <4 x b32> %src, <4 x i32> %off
- ret <4 x i8> %result
-}
-
-define <2 x float> @test_bitextract_v2f32(<2 x b64> %src) {
-; SSE2-LABEL: test_bitextract_v2f32:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa %xmm0, %xmm1
-; SSE2-NEXT: psrlq $32, %xmm1
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT: retq
-;
-; SSE42-LABEL: test_bitextract_v2f32:
-; SSE42: # %bb.0:
-; SSE42-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,12,13,14,15,12,13,14,15],zero,zero,zero,zero
-; SSE42-NEXT: retq
-;
-; AVX-LABEL: test_bitextract_v2f32:
-; AVX: # %bb.0:
-; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,12,13,14,15,12,13,14,15],zero,zero,zero,zero
-; AVX-NEXT: retq
- %result = bitextract <2 x float>, <2 x b64> %src, <2 x i32> <i32 0, i32 32>
- ret <2 x float> %result
-}
-
-define <2 x ptr> @test_bitextract_v2ptr(<2 x b64> %src) {
-; SSE-LABEL: test_bitextract_v2ptr:
-; SSE: # %bb.0:
-; SSE-NEXT: retq
-;
-; AVX-LABEL: test_bitextract_v2ptr:
-; AVX: # %bb.0:
-; AVX-NEXT: retq
- %result = bitextract <2 x ptr>, <2 x b64> %src, <2 x i32> zeroinitializer
- ret <2 x ptr> %result
-}
-
-define <16 x i4> @test_bitextract_v16i4_var(<16 x b8> %src, <16 x i32> %off) {
-; SSE2-LABEL: test_bitextract_v16i4_var:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; SSE2-NEXT: pand %xmm5, %xmm4
-; SSE2-NEXT: pand %xmm5, %xmm3
-; SSE2-NEXT: packuswb %xmm4, %xmm3
-; SSE2-NEXT: pand %xmm5, %xmm2
-; SSE2-NEXT: pand %xmm5, %xmm1
-; SSE2-NEXT: packuswb %xmm2, %xmm1
-; SSE2-NEXT: packuswb %xmm3, %xmm1
-; SSE2-NEXT: psllw $5, %xmm1
-; SSE2-NEXT: pxor %xmm2, %xmm2
-; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
-; SSE2-NEXT: movdqa %xmm2, %xmm3
-; SSE2-NEXT: pandn %xmm0, %xmm3
-; SSE2-NEXT: psrlw $4, %xmm0
-; SSE2-NEXT: pand %xmm2, %xmm0
-; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: por %xmm3, %xmm0
-; SSE2-NEXT: paddb %xmm1, %xmm1
-; SSE2-NEXT: pxor %xmm2, %xmm2
-; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
-; SSE2-NEXT: movdqa %xmm2, %xmm3
-; SSE2-NEXT: pandn %xmm0, %xmm3
-; SSE2-NEXT: psrlw $2, %xmm0
-; SSE2-NEXT: pand %xmm2, %xmm0
-; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: pxor %xmm2, %xmm2
-; SSE2-NEXT: por %xmm3, %xmm0
-; SSE2-NEXT: paddb %xmm1, %xmm1
-; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
-; SSE2-NEXT: movdqa %xmm2, %xmm1
-; SSE2-NEXT: pandn %xmm0, %xmm1
-; SSE2-NEXT: psrlw $1, %xmm0
-; SSE2-NEXT: pand %xmm2, %xmm0
-; SSE2-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: por %xmm1, %xmm0
-; SSE2-NEXT: retq
-;
-; SSE42-LABEL: test_bitextract_v16i4_var:
-; SSE42: # %bb.0:
-; SSE42-NEXT: movdqa %xmm0, %xmm5
-; SSE42-NEXT: movdqa {{.*#+}} xmm0 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; SSE42-NEXT: pand %xmm0, %xmm4
-; SSE42-NEXT: pand %xmm0, %xmm3
-; SSE42-NEXT: packusdw %xmm4, %xmm3
-; SSE42-NEXT: pand %xmm0, %xmm2
-; SSE42-NEXT: pand %xmm1, %xmm0
-; SSE42-NEXT: packusdw %xmm2, %xmm0
-; SSE42-NEXT: packuswb %xmm3, %xmm0
-; SSE42-NEXT: psllw $5, %xmm0
-; SSE42-NEXT: movdqa %xmm0, %xmm1
-; SSE42-NEXT: movdqa %xmm5, %xmm2
-; SSE42-NEXT: psrlw $4, %xmm2
-; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
-; SSE42-NEXT: paddb %xmm0, %xmm1
-; SSE42-NEXT: pblendvb %xmm0, %xmm2, %xmm5
-; SSE42-NEXT: movdqa %xmm5, %xmm2
-; SSE42-NEXT: psrlw $2, %xmm2
-; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
-; SSE42-NEXT: movdqa %xmm1, %xmm0
-; SSE42-NEXT: pblendvb %xmm0, %xmm2, %xmm5
-; SSE42-NEXT: movdqa %xmm5, %xmm2
-; SSE42-NEXT: psrlw $1, %xmm2
-; SSE42-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
-; SSE42-NEXT: paddb %xmm1, %xmm1
-; SSE42-NEXT: movdqa %xmm1, %xmm0
-; SSE42-NEXT: pblendvb %xmm0, %xmm2, %xmm5
-; SSE42-NEXT: movdqa %xmm5, %xmm0
-; SSE42-NEXT: retq
-;
-; AVX2-LABEL: test_bitextract_v16i4_var:
-; AVX2: # %bb.0:
-; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; AVX2-NEXT: vpand %ymm3, %ymm2, %ymm2
-; AVX2-NEXT: vpand %ymm3, %ymm1, %ymm1
-; AVX2-NEXT: vpackusdw %ymm2, %ymm1, %ymm1
-; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-NEXT: vpackuswb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,1,3]
-; AVX2-NEXT: vpsllw $5, %xmm1, %xmm1
-; AVX2-NEXT: vpsrlw $4, %xmm0, %xmm2
-; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT: vpaddb %xmm1, %xmm1, %xmm3
-; AVX2-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vpsrlw $2, %xmm0, %xmm1
-; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX2-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: vpsrlw $1, %xmm0, %xmm1
-; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX2-NEXT: vpaddb %xmm3, %xmm3, %xmm2
-; AVX2-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: vzeroupper
-; AVX2-NEXT: retq
-;
-; AVX512-LABEL: test_bitextract_v16i4_var:
-; AVX512: # %bb.0:
-; AVX512-NEXT: vpmovdw %zmm1, %ymm1
-; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm1
-; AVX512-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0
-; AVX512-NEXT: vpmovwb %ymm0, %xmm0
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
- %result = bitextract <16 x i4>, <16 x b8> %src, <16 x i32> %off
- ret <16 x i4> %result
-}
-
define <4 x b32> @test_bitinsert_v4i8_const(<4 x b32> %base, <4 x i8> %val) {
; SSE2-LABEL: test_bitinsert_v4i8_const:
; SSE2: # %bb.0:
More information about the llvm-branch-commits
mailing list