[DAGCombine] add tests for bswap-shift optimization

[Chenbing.Zheng]

Summary

This patch add some tests to show some optimization opportunities for bswap-shift.

[RKSimon]

llvm/test/CodeGen/X86/combine-bswap.ll

42

Please can you add one i16 + i64 test case each as well? i16 as it lowers to a rotate and i64 so we can check its correctly handled on 32-bit targets

[Chenbing.Zheng]

address comment

[RKSimon]

LGTM

llvm/test/CodeGen/RISCV/bswap-srli-bswap.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s -check-prefix=RV32I
+; RUN: llc -mtriple=riscv64 -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s -check-prefix=RV64I
+; 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
+
+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 {
+; RV32I-LABEL: test_bswap_srli_7_bswap_i16:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slli a1, a0, 8
+; RV32I-NEXT:    slli a0, a0, 16
+; RV32I-NEXT:    srli a0, a0, 24
+; RV32I-NEXT:    or a0, a1, a0
+; RV32I-NEXT:    slli a0, a0, 16
+; RV32I-NEXT:    srli a0, a0, 16
+; RV32I-NEXT:    slli a1, a0, 1
+; RV32I-NEXT:    andi a1, a1, -256
+; RV32I-NEXT:    srli a0, a0, 15
+; RV32I-NEXT:    or a0, a1, a0
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: test_bswap_srli_7_bswap_i16:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    slli a1, a0, 8
+; RV64I-NEXT:    slli a0, a0, 48
+; RV64I-NEXT:    srli a0, a0, 56
+; RV64I-NEXT:    or a0, a1, a0
+; RV64I-NEXT:    slli a0, a0, 48
+; RV64I-NEXT:    srli a0, a0, 48
+; RV64I-NEXT:    slli a1, a0, 1
+; RV64I-NEXT:    andi a1, a1, -256
+; RV64I-NEXT:    srli a0, a0, 15
+; RV64I-NEXT:    or a0, a1, a0
+; RV64I-NEXT:    ret
+;
+; 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 {
+; RV32I-LABEL: test_bswap_srli_8_bswap_i16:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    andi a0, a0, 255
+; RV32I-NEXT:    slli a0, a0, 8
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: test_bswap_srli_8_bswap_i16:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    andi a0, a0, 255
+; RV64I-NEXT:    slli a0, a0, 8
+; RV64I-NEXT:    ret
+;
+; 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 {
+; RV32I-LABEL: test_bswap_srli_8_bswap_i32:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    srli a1, a0, 8
+; RV32I-NEXT:    lui a2, 16
+; RV32I-NEXT:    addi a2, a2, -256
+; RV32I-NEXT:    and a1, a1, a2
+; RV32I-NEXT:    srli a2, a0, 24
+; RV32I-NEXT:    or a1, a1, a2
+; RV32I-NEXT:    slli a2, a0, 8
+; RV32I-NEXT:    lui a3, 4080
+; RV32I-NEXT:    and a2, a2, a3
+; RV32I-NEXT:    slli a0, a0, 24
+; RV32I-NEXT:    or a0, a0, a2
+; RV32I-NEXT:    or a0, a0, a1
+; RV32I-NEXT:    srli a1, a0, 8
+; RV32I-NEXT:    slli a1, a1, 24
+; RV32I-NEXT:    and a2, a0, a3
+; RV32I-NEXT:    or a1, a1, a2
+; RV32I-NEXT:    srli a0, a0, 16
+; RV32I-NEXT:    andi a0, a0, -256
+; RV32I-NEXT:    or a0, a0, a1
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: test_bswap_srli_8_bswap_i32:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    srliw a1, a0, 8
+; RV64I-NEXT:    lui a2, 16
+; RV64I-NEXT:    addiw a2, a2, -256
+; RV64I-NEXT:    and a1, a1, a2
+; RV64I-NEXT:    srliw a2, a0, 24
+; RV64I-NEXT:    or a1, a1, a2
+; RV64I-NEXT:    slli a2, a0, 8
+; RV64I-NEXT:    lui a3, 4080
+; RV64I-NEXT:    and a2, a2, a3
+; RV64I-NEXT:    slliw a0, a0, 24
+; RV64I-NEXT:    or a0, a0, a2
+; RV64I-NEXT:    or a0, a0, a1
+; RV64I-NEXT:    slli a1, a0, 32
+; RV64I-NEXT:    srli a1, a1, 32
+; RV64I-NEXT:    srli a1, a1, 8
+; RV64I-NEXT:    and a2, a0, a3
+; RV64I-NEXT:    slliw a1, a1, 24
+; RV64I-NEXT:    or a1, a1, a2
+; RV64I-NEXT:    srliw a0, a0, 16
+; RV64I-NEXT:    andi a0, a0, -256
+; RV64I-NEXT:    or a0, a0, a1
+; RV64I-NEXT:    ret
+;
+; 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 {
+; RV32I-LABEL: test_bswap_srli_16_bswap_i32:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    srli a1, a0, 8
+; RV32I-NEXT:    lui a2, 16
+; RV32I-NEXT:    addi a2, a2, -256
+; RV32I-NEXT:    and a1, a1, a2
+; RV32I-NEXT:    srli a2, a0, 24
+; RV32I-NEXT:    or a1, a1, a2
+; RV32I-NEXT:    slli a2, a0, 8
+; RV32I-NEXT:    lui a3, 4080
+; RV32I-NEXT:    and a2, a2, a3
+; RV32I-NEXT:    slli a0, a0, 24
+; RV32I-NEXT:    or a0, a0, a2
+; RV32I-NEXT:    or a0, a0, a1
+; RV32I-NEXT:    srli a1, a0, 16
+; RV32I-NEXT:    slli a1, a1, 24
+; RV32I-NEXT:    srli a0, a0, 24
+; RV32I-NEXT:    slli a0, a0, 16
+; RV32I-NEXT:    or a0, a1, a0
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: test_bswap_srli_16_bswap_i32:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    srliw a1, a0, 8
+; RV64I-NEXT:    lui a2, 16
+; RV64I-NEXT:    addiw a2, a2, -256
+; RV64I-NEXT:    and a1, a1, a2
+; RV64I-NEXT:    srliw a2, a0, 24
+; RV64I-NEXT:    or a1, a1, a2
+; RV64I-NEXT:    slli a2, a0, 8
+; RV64I-NEXT:    lui a3, 4080
+; RV64I-NEXT:    and a2, a2, a3
+; RV64I-NEXT:    slli a0, a0, 24
+; RV64I-NEXT:    or a0, a0, a2
+; RV64I-NEXT:    or a0, a0, a1
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    srli a1, a0, 16
+; RV64I-NEXT:    srliw a0, a0, 24
+; RV64I-NEXT:    slli a0, a0, 16
+; RV64I-NEXT:    slliw a1, a1, 24
+; RV64I-NEXT:    or a0, a1, a0
+; RV64I-NEXT:    ret
+;
+; 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 {
+; RV32I-LABEL: test_bswap_srli_24_bswap_i32:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    slli a0, a0, 24
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: test_bswap_srli_24_bswap_i32:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    slliw a0, a0, 24
+; RV64I-NEXT:    ret
+;
+; 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 {
+; RV32I-LABEL: test_bswap_srli_48_bswap_i64:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    srli a1, a0, 8
+; RV32I-NEXT:    lui a2, 16
+; RV32I-NEXT:    addi a2, a2, -256
+; RV32I-NEXT:    and a1, a1, a2
+; RV32I-NEXT:    srli a2, a0, 24
+; RV32I-NEXT:    or a1, a1, a2
+; RV32I-NEXT:    slli a2, a0, 8
+; RV32I-NEXT:    lui a3, 4080
+; RV32I-NEXT:    and a2, a2, a3
+; RV32I-NEXT:    slli a0, a0, 24
+; RV32I-NEXT:    or a0, a0, a2
+; RV32I-NEXT:    or a0, a0, a1
+; RV32I-NEXT:    srli a1, a0, 16
+; RV32I-NEXT:    slli a1, a1, 24
+; RV32I-NEXT:    srli a0, a0, 24
+; RV32I-NEXT:    slli a0, a0, 16
+; RV32I-NEXT:    or a1, a1, a0
+; RV32I-NEXT:    li a0, 0
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: test_bswap_srli_48_bswap_i64:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    srli a1, a0, 24
+; RV64I-NEXT:    lui a2, 4080
+; RV64I-NEXT:    and a1, a1, a2
+; RV64I-NEXT:    srli a2, a0, 8
+; RV64I-NEXT:    li a3, 255
+; RV64I-NEXT:    slli a4, a3, 24
+; RV64I-NEXT:    and a2, a2, a4
+; RV64I-NEXT:    or a1, a2, a1
+; RV64I-NEXT:    srli a2, a0, 40
+; RV64I-NEXT:    lui a4, 16
+; RV64I-NEXT:    addiw a4, a4, -256
+; RV64I-NEXT:    and a2, a2, a4
+; RV64I-NEXT:    srli a4, a0, 56
+; RV64I-NEXT:    or a2, a2, a4
+; RV64I-NEXT:    or a1, a1, a2
+; RV64I-NEXT:    slli a2, a0, 24
+; RV64I-NEXT:    slli a4, a3, 40
+; RV64I-NEXT:    and a2, a2, a4
+; RV64I-NEXT:    srliw a4, a0, 24
+; RV64I-NEXT:    slli a4, a4, 32
+; RV64I-NEXT:    or a2, a2, a4
+; RV64I-NEXT:    slli a4, a0, 40
+; RV64I-NEXT:    slli a3, a3, 48
+; RV64I-NEXT:    and a3, a4, a3
+; RV64I-NEXT:    slli a0, a0, 56
+; RV64I-NEXT:    or a0, a0, a3
+; RV64I-NEXT:    or a0, a0, a2
+; RV64I-NEXT:    or a0, a0, a1
+; RV64I-NEXT:    srli a1, a0, 48
+; RV64I-NEXT:    slli a1, a1, 56
+; RV64I-NEXT:    srli a0, a0, 56
+; RV64I-NEXT:    slli a0, a0, 48
+; RV64I-NEXT:    or a0, a1, a0
+; RV64I-NEXT:    ret
+;
+; 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
+}

llvm/test/CodeGen/X86/combine-bswap.ll

 ; X64:       # %bb.0:
 ; X64-NEXT:    movl %edi, %eax
 ; X64-NEXT:    retq
   %b = call i32 @llvm.bswap.i32(i32 %a0)
   %c = call i32 @llvm.bswap.i32(i32 %b)
   ret i32 %c
 }
 
+; TODO: fold (bswap(srl (bswap c), x)) -> (shl c, x)
+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 i32 @test_demandedbits_bswap(i32 %a0) nounwind {
 ; X86-LABEL: test_demandedbits_bswap:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    bswapl %eax
 ; X86-NEXT:    andl $-65536, %eax # imm = 0xFFFF0000
 ; X86-NEXT:    retl
 ;