--- /dev/null
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+zbb -verify-machineinstrs < %s \
+; RUN: | FileCheck %s -check-prefixes=RV32ZB
+; RUN: llc -mtriple=riscv64 -mattr=+zbb -verify-machineinstrs < %s \
+; RUN: | FileCheck %s -check-prefixes=RV64ZB
+; RUN: llc -mtriple=riscv32 -mattr=+zbkb -verify-machineinstrs < %s \
+; RUN: | FileCheck %s -check-prefixes=RV32ZB
+; RUN: llc -mtriple=riscv64 -mattr=+zbkb -verify-machineinstrs < %s \
+; RUN: | FileCheck %s -check-prefixes=RV64ZB
+
+; TODO: These tests can be optmised, with x%8 == 0
+; fold (bswap(srl (bswap c), x)) -> (shl c, x)
+; fold (bswap(shl (bswap c), x)) -> (srl c, x)
+
+declare i16 @llvm.bswap.i16(i16)
+declare i32 @llvm.bswap.i32(i32)
+declare i64 @llvm.bswap.i64(i64)
+
+define i16 @test_bswap_srli_7_bswap_i16(i16 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_srli_7_bswap_i16:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: srli a0, a0, 23
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: srli a0, a0, 16
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_srli_7_bswap_i16:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 55
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 48
+; RV64ZB-NEXT: ret
+ %1 = call i16 @llvm.bswap.i16(i16 %a)
+ %2 = lshr i16 %1, 7
+ %3 = call i16 @llvm.bswap.i16(i16 %2)
+ ret i16 %3
+}
+
+define i16 @test_bswap_srli_8_bswap_i16(i16 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_srli_8_bswap_i16:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: andi a0, a0, 255
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: srli a0, a0, 16
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_srli_8_bswap_i16:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: andi a0, a0, 255
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 48
+; RV64ZB-NEXT: ret
+ %1 = call i16 @llvm.bswap.i16(i16 %a)
+ %2 = lshr i16 %1, 8
+ %3 = call i16 @llvm.bswap.i16(i16 %2)
+ ret i16 %3
+}
+
+define i32 @test_bswap_srli_8_bswap_i32(i32 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_srli_8_bswap_i32:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: srli a0, a0, 8
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_srli_8_bswap_i32:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 40
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 32
+; RV64ZB-NEXT: ret
+ %1 = call i32 @llvm.bswap.i32(i32 %a)
+ %2 = lshr i32 %1, 8
+ %3 = call i32 @llvm.bswap.i32(i32 %2)
+ ret i32 %3
+}
+
+define i32 @test_bswap_srli_16_bswap_i32(i32 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_srli_16_bswap_i32:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: srli a0, a0, 16
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_srli_16_bswap_i32:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 48
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 32
+; RV64ZB-NEXT: ret
+ %1 = call i32 @llvm.bswap.i32(i32 %a)
+ %2 = lshr i32 %1, 16
+ %3 = call i32 @llvm.bswap.i32(i32 %2)
+ ret i32 %3
+}
+
+define i32 @test_bswap_srli_24_bswap_i32(i32 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_srli_24_bswap_i32:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: andi a0, a0, 255
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_srli_24_bswap_i32:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: andi a0, a0, 255
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 32
+; RV64ZB-NEXT: ret
+ %1 = call i32 @llvm.bswap.i32(i32 %a)
+ %2 = lshr i32 %1, 24
+ %3 = call i32 @llvm.bswap.i32(i32 %2)
+ ret i32 %3
+}
+
+define i64 @test_bswap_srli_48_bswap_i64(i64 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_srli_48_bswap_i64:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: srli a0, a0, 16
+; RV32ZB-NEXT: rev8 a1, a0
+; RV32ZB-NEXT: li a0, 0
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_srli_48_bswap_i64:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 48
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: ret
+ %1 = call i64 @llvm.bswap.i64(i64 %a)
+ %2 = lshr i64 %1, 48
+ %3 = call i64 @llvm.bswap.i64(i64 %2)
+ ret i64 %3
+}
+
+define i16 @test_bswap_shli_7_bswap_i16(i16 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_shli_7_bswap_i16:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: srli a0, a0, 9
+; RV32ZB-NEXT: andi a0, a0, -128
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: srli a0, a0, 16
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_shli_7_bswap_i16:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 41
+; RV64ZB-NEXT: andi a0, a0, -128
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 48
+; RV64ZB-NEXT: ret
+ %1 = call i16 @llvm.bswap.i16(i16 %a)
+ %2 = shl i16 %1, 7
+ %3 = call i16 @llvm.bswap.i16(i16 %2)
+ ret i16 %3
+}
+
+define i16 @test_bswap_shli_8_bswap_i16(i16 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_shli_8_bswap_i16:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: andi a0, a0, -256
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: srli a0, a0, 16
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_shli_8_bswap_i16:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: andi a0, a0, -256
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 48
+; RV64ZB-NEXT: ret
+ %1 = call i16 @llvm.bswap.i16(i16 %a)
+ %2 = shl i16 %1, 8
+ %3 = call i16 @llvm.bswap.i16(i16 %2)
+ ret i16 %3
+}
+
+define i32 @test_bswap_shli_8_bswap_i32(i32 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_shli_8_bswap_i32:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: slli a0, a0, 8
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_shli_8_bswap_i32:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 24
+; RV64ZB-NEXT: andi a0, a0, -256
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 32
+; RV64ZB-NEXT: ret
+ %1 = call i32 @llvm.bswap.i32(i32 %a)
+ %2 = shl i32 %1, 8
+ %3 = call i32 @llvm.bswap.i32(i32 %2)
+ ret i32 %3
+}
+
+define i32 @test_bswap_shli_16_bswap_i32(i32 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_shli_16_bswap_i32:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: slli a0, a0, 16
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_shli_16_bswap_i32:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 32
+; RV64ZB-NEXT: slli a0, a0, 16
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 32
+; RV64ZB-NEXT: ret
+ %1 = call i32 @llvm.bswap.i32(i32 %a)
+ %2 = shl i32 %1, 16
+ %3 = call i32 @llvm.bswap.i32(i32 %2)
+ ret i32 %3
+}
+
+define i32 @test_bswap_shli_24_bswap_i32(i32 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_shli_24_bswap_i32:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: lui a1, 1044480
+; RV32ZB-NEXT: and a0, a0, a1
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_shli_24_bswap_i32:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: lui a1, 1044480
+; RV64ZB-NEXT: and a0, a0, a1
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: srli a0, a0, 32
+; RV64ZB-NEXT: ret
+ %1 = call i32 @llvm.bswap.i32(i32 %a)
+ %2 = shl i32 %1, 24
+ %3 = call i32 @llvm.bswap.i32(i32 %2)
+ ret i32 %3
+}
+
+define i64 @test_bswap_shli_48_bswap_i64(i64 %a) nounwind {
+; RV32ZB-LABEL: test_bswap_shli_48_bswap_i64:
+; RV32ZB: # %bb.0:
+; RV32ZB-NEXT: rev8 a0, a1
+; RV32ZB-NEXT: slli a0, a0, 16
+; RV32ZB-NEXT: rev8 a0, a0
+; RV32ZB-NEXT: li a1, 0
+; RV32ZB-NEXT: ret
+;
+; RV64ZB-LABEL: test_bswap_shli_48_bswap_i64:
+; RV64ZB: # %bb.0:
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: slli a0, a0, 48
+; RV64ZB-NEXT: rev8 a0, a0
+; RV64ZB-NEXT: ret
+ %1 = call i64 @llvm.bswap.i64(i64 %a)
+ %2 = shl i64 %1, 48
+ %3 = call i64 @llvm.bswap.i64(i64 %2)
+ ret i64 %3
+}
; These tests just check that the plumbing is in place for @llvm.bswap. The
; actual output is massive at the moment as llvm.bswap is not yet legal.
+declare i16 @llvm.bswap.i16(i16) readnone
declare i32 @llvm.bswap.i32(i32) readnone
declare i64 @llvm.bswap.i64(i64) readnone
declare i32 @llvm.bswap.v4i32(i32) readnone
ret i32 %c
}
+; TODO: fold (bswap(srl (bswap c), x)) -> (shl c, x)
+define i16 @test_bswap_srli_8_bswap_i16(i16 %a) nounwind {
+; X86-LABEL: test_bswap_srli_8_bswap_i16:
+; X86: # %bb.0:
+; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: rolw $8, %ax
+; X86-NEXT: # kill: def $ax killed $ax killed $eax
+; X86-NEXT: retl
+;
+; X64-LABEL: test_bswap_srli_8_bswap_i16:
+; X64: # %bb.0:
+; X64-NEXT: movzbl %dil, %eax
+; X64-NEXT: rolw $8, %ax
+; X64-NEXT: # kill: def $ax killed $ax killed $eax
+; X64-NEXT: retq
+ %1 = call i16 @llvm.bswap.i16(i16 %a)
+ %2 = lshr i16 %1, 8
+ %3 = call i16 @llvm.bswap.i16(i16 %2)
+ ret i16 %3
+}
+
+define i32 @test_bswap_srli_8_bswap_i32(i32 %a) nounwind {
+; X86-LABEL: test_bswap_srli_8_bswap_i32:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: bswapl %eax
+; X86-NEXT: shrl $8, %eax
+; X86-NEXT: bswapl %eax
+; X86-NEXT: retl
+;
+; X64-LABEL: test_bswap_srli_8_bswap_i32:
+; X64: # %bb.0:
+; X64-NEXT: movl %edi, %eax
+; X64-NEXT: bswapl %eax
+; X64-NEXT: shrl $8, %eax
+; X64-NEXT: bswapl %eax
+; X64-NEXT: retq
+ %1 = call i32 @llvm.bswap.i32(i32 %a)
+ %2 = lshr i32 %1, 8
+ %3 = call i32 @llvm.bswap.i32(i32 %2)
+ ret i32 %3
+}
+
+define i64 @test_bswap_srli_16_bswap_i64(i64 %a) nounwind {
+; X86-LABEL: test_bswap_srli_16_bswap_i64:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT: bswapl %edx
+; X86-NEXT: bswapl %eax
+; X86-NEXT: shrdl $16, %eax, %edx
+; X86-NEXT: shrl $16, %eax
+; X86-NEXT: bswapl %edx
+; X86-NEXT: bswapl %eax
+; X86-NEXT: retl
+;
+; X64-LABEL: test_bswap_srli_16_bswap_i64:
+; X64: # %bb.0:
+; X64-NEXT: movq %rdi, %rax
+; X64-NEXT: bswapq %rax
+; X64-NEXT: shrq $16, %rax
+; X64-NEXT: bswapq %rax
+; X64-NEXT: retq
+ %1 = call i64 @llvm.bswap.i64(i64 %a)
+ %2 = lshr i64 %1, 16
+ %3 = call i64 @llvm.bswap.i64(i64 %2)
+ ret i64 %3
+}
+
+; TODO: fold (bswap(shl (bswap c), x)) -> (srl c, x)
+define i16 @test_bswap_shli_8_bswap_i16(i16 %a) nounwind {
+; X86-LABEL: test_bswap_shli_8_bswap_i16:
+; X86: # %bb.0:
+; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: shll $8, %eax
+; X86-NEXT: rolw $8, %ax
+; X86-NEXT: # kill: def $ax killed $ax killed $eax
+; X86-NEXT: retl
+;
+; X64-LABEL: test_bswap_shli_8_bswap_i16:
+; X64: # %bb.0:
+; X64-NEXT: movl %edi, %eax
+; X64-NEXT: andl $65280, %eax # imm = 0xFF00
+; X64-NEXT: rolw $8, %ax
+; X64-NEXT: # kill: def $ax killed $ax killed $eax
+; X64-NEXT: retq
+ %1 = call i16 @llvm.bswap.i16(i16 %a)
+ %2 = shl i16 %1, 8
+ %3 = call i16 @llvm.bswap.i16(i16 %2)
+ ret i16 %3
+}
+
+define i32 @test_bswap_shli_8_bswap_i32(i32 %a) nounwind {
+; X86-LABEL: test_bswap_shli_8_bswap_i32:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: bswapl %eax
+; X86-NEXT: shll $8, %eax
+; X86-NEXT: bswapl %eax
+; X86-NEXT: retl
+;
+; X64-LABEL: test_bswap_shli_8_bswap_i32:
+; X64: # %bb.0:
+; X64-NEXT: movl %edi, %eax
+; X64-NEXT: bswapl %eax
+; X64-NEXT: shll $8, %eax
+; X64-NEXT: bswapl %eax
+; X64-NEXT: retq
+ %1 = call i32 @llvm.bswap.i32(i32 %a)
+ %2 = shl i32 %1, 8
+ %3 = call i32 @llvm.bswap.i32(i32 %2)
+ ret i32 %3
+}
+
+define i64 @test_bswap_shli_16_bswap_i64(i64 %a) nounwind {
+; X86-LABEL: test_bswap_shli_16_bswap_i64:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT: bswapl %ecx
+; X86-NEXT: bswapl %eax
+; X86-NEXT: shldl $16, %ecx, %eax
+; X86-NEXT: bswapl %eax
+; X86-NEXT: rolw $8, %cx
+; X86-NEXT: movzwl %cx, %edx
+; X86-NEXT: retl
+;
+; X64-LABEL: test_bswap_shli_16_bswap_i64:
+; X64: # %bb.0:
+; X64-NEXT: movq %rdi, %rax
+; X64-NEXT: bswapq %rax
+; X64-NEXT: shlq $16, %rax
+; X64-NEXT: bswapq %rax
+; X64-NEXT: retq
+ %1 = call i64 @llvm.bswap.i64(i64 %a)
+ %2 = shl i64 %1, 16
+ %3 = call i64 @llvm.bswap.i64(i64 %2)
+ ret i64 %3
+}
+
define i32 @test_demandedbits_bswap(i32 %a0) nounwind {
; X86-LABEL: test_demandedbits_bswap:
; X86: # %bb.0: