Skip to content

[X86][MC] Support encoding optimization & assembler relaxation about immediate operands for APX instructions #78545

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Jan 30, 2024
Merged

[X86][MC] Support encoding optimization & assembler relaxation about immediate operands for APX instructions #78545

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
a4377a9
[X86] Support encoding optimization fo immediate for ndd instructions
XinWang10 Jan 18, 2024
47c4552
support shift compress
XinWang10 Jan 26, 2024
dff1c5b
update test
XinWang10 Jan 29, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
39 changes: 31 additions & 8 deletions llvm/lib/Target/X86/MCTargetDesc/X86EncodingOptimization.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,6 +106,12 @@ bool X86::optimizeShiftRotateWithImmediateOne(MCInst &MI) {
#define TO_IMM1(FROM) \
case X86::FROM##i: \
NewOpc = X86::FROM##1; \
break; \
case X86::FROM##i_EVEX: \
NewOpc = X86::FROM##1_EVEX; \
break; \
case X86::FROM##i_ND: \
NewOpc = X86::FROM##1_ND; \
break;
switch (MI.getOpcode()) {
default:
Expand All @@ -118,6 +124,31 @@ bool X86::optimizeShiftRotateWithImmediateOne(MCInst &MI) {
TO_IMM1(RCL16r)
TO_IMM1(RCL32r)
TO_IMM1(RCL64r)
TO_IMM1(RCR8m)
TO_IMM1(RCR16m)
TO_IMM1(RCR32m)
TO_IMM1(RCR64m)
TO_IMM1(RCL8m)
TO_IMM1(RCL16m)
TO_IMM1(RCL32m)
TO_IMM1(RCL64m)
#undef TO_IMM1
#define TO_IMM1(FROM) \
case X86::FROM##i: \
NewOpc = X86::FROM##1; \
break; \
case X86::FROM##i_EVEX: \
NewOpc = X86::FROM##1_EVEX; \
break; \
case X86::FROM##i_NF: \
NewOpc = X86::FROM##1_NF; \
break; \
case X86::FROM##i_ND: \
NewOpc = X86::FROM##1_ND; \
break; \
case X86::FROM##i_NF_ND: \
NewOpc = X86::FROM##1_NF_ND; \
break;
TO_IMM1(ROR8r)
TO_IMM1(ROR16r)
TO_IMM1(ROR32r)
Expand All @@ -138,14 +169,6 @@ bool X86::optimizeShiftRotateWithImmediateOne(MCInst &MI) {
TO_IMM1(SHL16r)
TO_IMM1(SHL32r)
TO_IMM1(SHL64r)
TO_IMM1(RCR8m)
TO_IMM1(RCR16m)
TO_IMM1(RCR32m)
TO_IMM1(RCR64m)
TO_IMM1(RCL8m)
TO_IMM1(RCL16m)
TO_IMM1(RCL32m)
TO_IMM1(RCL64m)
TO_IMM1(ROR8m)
TO_IMM1(ROR16m)
TO_IMM1(ROR32m)
Expand Down
113 changes: 65 additions & 48 deletions llvm/lib/Target/X86/MCTargetDesc/X86EncodingOptimizationForImmediate.def
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,61 +12,78 @@
#ifndef ENTRY
#define ENTRY(LONG, SHORT)
#endif
ENTRY(ADC16mi, ADC16mi8)
ENTRY(ADC16ri, ADC16ri8)
ENTRY(ADC32mi, ADC32mi8)
ENTRY(ADC32ri, ADC32ri8)
ENTRY(ADC64mi32, ADC64mi8)
ENTRY(ADC64ri32, ADC64ri8)
ENTRY(SBB16mi, SBB16mi8)
ENTRY(SBB16ri, SBB16ri8)
ENTRY(SBB32mi, SBB32mi8)
ENTRY(SBB32ri, SBB32ri8)
ENTRY(SBB64mi32, SBB64mi8)
ENTRY(SBB64ri32, SBB64ri8)
ENTRY(ADD16mi, ADD16mi8)
ENTRY(ADD16ri, ADD16ri8)
ENTRY(ADD32mi, ADD32mi8)
ENTRY(ADD32ri, ADD32ri8)
ENTRY(ADD64mi32, ADD64mi8)
ENTRY(ADD64ri32, ADD64ri8)
ENTRY(AND16mi, AND16mi8)
ENTRY(AND16ri, AND16ri8)
ENTRY(AND32mi, AND32mi8)
ENTRY(AND32ri, AND32ri8)
ENTRY(AND64mi32, AND64mi8)
ENTRY(AND64ri32, AND64ri8)
ENTRY(OR16mi, OR16mi8)
ENTRY(OR16ri, OR16ri8)
ENTRY(OR32mi, OR32mi8)
ENTRY(OR32ri, OR32ri8)
ENTRY(OR64mi32, OR64mi8)
ENTRY(OR64ri32, OR64ri8)
ENTRY(SUB16mi, SUB16mi8)
ENTRY(SUB16ri, SUB16ri8)
ENTRY(SUB32mi, SUB32mi8)
ENTRY(SUB32ri, SUB32ri8)
ENTRY(SUB64mi32, SUB64mi8)
ENTRY(SUB64ri32, SUB64ri8)
ENTRY(XOR16mi, XOR16mi8)
ENTRY(XOR16ri, XOR16ri8)
ENTRY(XOR32mi, XOR32mi8)
ENTRY(XOR32ri, XOR32ri8)
ENTRY(XOR64mi32, XOR64mi8)
ENTRY(XOR64ri32, XOR64ri8)
ENTRY(CMP16mi, CMP16mi8)
ENTRY(CMP16ri, CMP16ri8)
ENTRY(CMP32mi, CMP32mi8)
ENTRY(CMP32ri, CMP32ri8)
ENTRY(CMP64mi32, CMP64mi8)
ENTRY(CMP64ri32, CMP64ri8)
ENTRY(IMUL16rmi, IMUL16rmi8)
ENTRY(IMUL16rri, IMUL16rri8)
ENTRY(IMUL32rmi, IMUL32rmi8)
ENTRY(IMUL32rri, IMUL32rri8)
ENTRY(IMUL64rmi32, IMUL64rmi8)
ENTRY(IMUL64rri32, IMUL64rri8)
ENTRY(PUSH16i, PUSH16i8)
ENTRY(PUSH32i, PUSH32i8)
ENTRY(PUSH64i32, PUSH64i8)
#define ENTRYS(LONG, SHORT) \
ENTRY(LONG, SHORT) \
ENTRY(LONG##_EVEX, SHORT##_EVEX) \
ENTRY(LONG##_NF, SHORT##_NF) \
ENTRY(LONG##_ND, SHORT##_ND) \
ENTRY(LONG##_NF_ND, SHORT##_NF_ND)
ENTRYS(ADD16mi, ADD16mi8)
ENTRYS(ADD16ri, ADD16ri8)
ENTRYS(ADD32mi, ADD32mi8)
ENTRYS(ADD32ri, ADD32ri8)
ENTRYS(ADD64mi32, ADD64mi8)
ENTRYS(ADD64ri32, ADD64ri8)
ENTRYS(AND16mi, AND16mi8)
ENTRYS(AND16ri, AND16ri8)
ENTRYS(AND32mi, AND32mi8)
ENTRYS(AND32ri, AND32ri8)
ENTRYS(AND64mi32, AND64mi8)
ENTRYS(AND64ri32, AND64ri8)
ENTRYS(OR16mi, OR16mi8)
ENTRYS(OR16ri, OR16ri8)
ENTRYS(OR32mi, OR32mi8)
ENTRYS(OR32ri, OR32ri8)
ENTRYS(OR64mi32, OR64mi8)
ENTRYS(OR64ri32, OR64ri8)
ENTRYS(SUB16mi, SUB16mi8)
ENTRYS(SUB16ri, SUB16ri8)
ENTRYS(SUB32mi, SUB32mi8)
ENTRYS(SUB32ri, SUB32ri8)
ENTRYS(SUB64mi32, SUB64mi8)
ENTRYS(SUB64ri32, SUB64ri8)
ENTRYS(XOR16mi, XOR16mi8)
ENTRYS(XOR16ri, XOR16ri8)
ENTRYS(XOR32mi, XOR32mi8)
ENTRYS(XOR32ri, XOR32ri8)
ENTRYS(XOR64mi32, XOR64mi8)
ENTRYS(XOR64ri32, XOR64ri8)
#undef ENTRYS
#define ENTRYS(LONG, SHORT) \
ENTRY(LONG, SHORT) \
ENTRY(LONG##_EVEX, SHORT##_EVEX) \
ENTRY(LONG##_NF, SHORT##_NF)
ENTRYS(IMUL16rmi, IMUL16rmi8)
ENTRYS(IMUL16rri, IMUL16rri8)
ENTRYS(IMUL32rmi, IMUL32rmi8)
ENTRYS(IMUL32rri, IMUL32rri8)
ENTRYS(IMUL64rmi32, IMUL64rmi8)
ENTRYS(IMUL64rri32, IMUL64rri8)
#undef ENTRYS
#define ENTRYS(LONG, SHORT) \
ENTRY(LONG, SHORT) \
ENTRY(LONG##_EVEX, SHORT##_EVEX) \
ENTRY(LONG##_ND, SHORT##_ND)
ENTRYS(ADC16mi, ADC16mi8)
ENTRYS(ADC16ri, ADC16ri8)
ENTRYS(ADC32mi, ADC32mi8)
ENTRYS(ADC32ri, ADC32ri8)
ENTRYS(ADC64mi32, ADC64mi8)
ENTRYS(ADC64ri32, ADC64ri8)
ENTRYS(SBB16mi, SBB16mi8)
ENTRYS(SBB16ri, SBB16ri8)
ENTRYS(SBB32mi, SBB32mi8)
ENTRYS(SBB32ri, SBB32ri8)
ENTRYS(SBB64mi32, SBB64mi8)
ENTRYS(SBB64ri32, SBB64ri8)
#undef ENTRYS
#undef ENTRY
18 changes: 9 additions & 9 deletions llvm/test/CodeGen/X86/apx/adc.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -113,7 +113,7 @@ define i16 @adc16ri8(i16 %a, i16 %x, i16 %y) nounwind {
; CHECK-LABEL: adc16ri8:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpw %si, %dx # encoding: [0x66,0x39,0xf2]
; CHECK-NEXT: adcw $0, %di, %ax # encoding: [0x62,0xf4,0x7d,0x18,0x81,0xd7,0x00,0x00]
; CHECK-NEXT: adcw $0, %di, %ax # encoding: [0x62,0xf4,0x7d,0x18,0x83,0xd7,0x00]
; CHECK-NEXT: addl $123, %eax # EVEX TO LEGACY Compression encoding: [0x83,0xc0,0x7b]
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NEXT: retq # encoding: [0xc3]
Expand All @@ -128,7 +128,7 @@ define i32 @adc32ri8(i32 %a, i32 %x, i32 %y) nounwind {
; CHECK-LABEL: adc32ri8:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpl %esi, %edx # encoding: [0x39,0xf2]
; CHECK-NEXT: adcl $123, %edi, %eax # encoding: [0x62,0xf4,0x7c,0x18,0x81,0xd7,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: adcl $123, %edi, %eax # encoding: [0x62,0xf4,0x7c,0x18,0x83,0xd7,0x7b]
; CHECK-NEXT: retq # encoding: [0xc3]
%s = add i32 %a, 123
%k = icmp ugt i32 %x, %y
Expand All @@ -141,7 +141,7 @@ define i64 @adc64ri8(i64 %a, i64 %x, i64 %y) nounwind {
; CHECK-LABEL: adc64ri8:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpq %rsi, %rdx # encoding: [0x48,0x39,0xf2]
; CHECK-NEXT: adcq $123, %rdi, %rax # encoding: [0x62,0xf4,0xfc,0x18,0x81,0xd7,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: adcq $123, %rdi, %rax # encoding: [0x62,0xf4,0xfc,0x18,0x83,0xd7,0x7b]
; CHECK-NEXT: retq # encoding: [0xc3]
%s = add i64 %a, 123
%k = icmp ugt i64 %x, %y
Expand All @@ -167,7 +167,7 @@ define i16 @adc16ri(i16 %a, i16 %x, i16 %y) nounwind {
; CHECK-LABEL: adc16ri:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpw %si, %dx # encoding: [0x66,0x39,0xf2]
; CHECK-NEXT: adcw $0, %di, %ax # encoding: [0x62,0xf4,0x7d,0x18,0x81,0xd7,0x00,0x00]
; CHECK-NEXT: adcw $0, %di, %ax # encoding: [0x62,0xf4,0x7d,0x18,0x83,0xd7,0x00]
; CHECK-NEXT: addl $1234, %eax # EVEX TO LEGACY Compression encoding: [0x05,0xd2,0x04,0x00,0x00]
; CHECK-NEXT: # imm = 0x4D2
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
Expand Down Expand Up @@ -267,7 +267,7 @@ define i16 @adc16mi8(ptr %ptr, i16 %x, i16 %y) nounwind {
; CHECK-LABEL: adc16mi8:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpw %si, %dx # encoding: [0x66,0x39,0xf2]
; CHECK-NEXT: adcw $0, (%rdi), %ax # encoding: [0x62,0xf4,0x7d,0x18,0x81,0x17,0x00,0x00]
; CHECK-NEXT: adcw $0, (%rdi), %ax # encoding: [0x62,0xf4,0x7d,0x18,0x83,0x17,0x00]
; CHECK-NEXT: addl $123, %eax # EVEX TO LEGACY Compression encoding: [0x83,0xc0,0x7b]
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NEXT: retq # encoding: [0xc3]
Expand All @@ -283,7 +283,7 @@ define i32 @adc32mi8(ptr %ptr, i32 %x, i32 %y) nounwind {
; CHECK-LABEL: adc32mi8:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpl %esi, %edx # encoding: [0x39,0xf2]
; CHECK-NEXT: adcl $123, (%rdi), %eax # encoding: [0x62,0xf4,0x7c,0x18,0x81,0x17,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: adcl $123, (%rdi), %eax # encoding: [0x62,0xf4,0x7c,0x18,0x83,0x17,0x7b]
; CHECK-NEXT: retq # encoding: [0xc3]
%a = load i32, ptr %ptr
%s = add i32 %a, 123
Expand All @@ -297,7 +297,7 @@ define i64 @adc64mi8(ptr %ptr, i64 %x, i64 %y) nounwind {
; CHECK-LABEL: adc64mi8:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpq %rsi, %rdx # encoding: [0x48,0x39,0xf2]
; CHECK-NEXT: adcq $123, (%rdi), %rax # encoding: [0x62,0xf4,0xfc,0x18,0x81,0x17,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: adcq $123, (%rdi), %rax # encoding: [0x62,0xf4,0xfc,0x18,0x83,0x17,0x7b]
; CHECK-NEXT: retq # encoding: [0xc3]
%a = load i64, ptr %ptr
%s = add i64 %a, 123
Expand Down Expand Up @@ -325,7 +325,7 @@ define i16 @adc16mi(ptr %ptr, i16 %x, i16 %y) nounwind {
; CHECK-LABEL: adc16mi:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpw %si, %dx # encoding: [0x66,0x39,0xf2]
; CHECK-NEXT: adcw $0, (%rdi), %ax # encoding: [0x62,0xf4,0x7d,0x18,0x81,0x17,0x00,0x00]
; CHECK-NEXT: adcw $0, (%rdi), %ax # encoding: [0x62,0xf4,0x7d,0x18,0x83,0x17,0x00]
; CHECK-NEXT: addl $1234, %eax # EVEX TO LEGACY Compression encoding: [0x05,0xd2,0x04,0x00,0x00]
; CHECK-NEXT: # imm = 0x4D2
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
Expand Down Expand Up @@ -447,7 +447,7 @@ define void @adc16mi_legacy(ptr %ptr, i16 %x, i16 %y) nounwind {
; CHECK-LABEL: adc16mi_legacy:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpw %si, %dx # encoding: [0x66,0x39,0xf2]
; CHECK-NEXT: adcw $0, (%rdi), %ax # encoding: [0x62,0xf4,0x7d,0x18,0x81,0x17,0x00,0x00]
; CHECK-NEXT: adcw $0, (%rdi), %ax # encoding: [0x62,0xf4,0x7d,0x18,0x83,0x17,0x00]
; CHECK-NEXT: addl $1234, %eax # EVEX TO LEGACY Compression encoding: [0x05,0xd2,0x04,0x00,0x00]
; CHECK-NEXT: # imm = 0x4D2
; CHECK-NEXT: movw %ax, (%rdi) # encoding: [0x66,0x89,0x07]
Expand Down
16 changes: 8 additions & 8 deletions llvm/test/CodeGen/X86/apx/add.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,7 +89,7 @@ entry:
define i16 @add16ri8(i16 noundef %a) {
; CHECK-LABEL: add16ri8:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: addl $123, %edi, %eax # encoding: [0x62,0xf4,0x7c,0x18,0x81,0xc7,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: addl $123, %edi, %eax # encoding: [0x62,0xf4,0x7c,0x18,0x83,0xc7,0x7b]
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NEXT: retq # encoding: [0xc3]
entry:
Expand All @@ -100,7 +100,7 @@ entry:
define i32 @add32ri8(i32 noundef %a) {
; CHECK-LABEL: add32ri8:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: addl $123, %edi, %eax # encoding: [0x62,0xf4,0x7c,0x18,0x81,0xc7,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: addl $123, %edi, %eax # encoding: [0x62,0xf4,0x7c,0x18,0x83,0xc7,0x7b]
; CHECK-NEXT: retq # encoding: [0xc3]
entry:
%add = add i32 %a, 123
Expand All @@ -110,7 +110,7 @@ entry:
define i64 @add64ri8(i64 noundef %a) {
; CHECK-LABEL: add64ri8:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: addq $123, %rdi, %rax # encoding: [0x62,0xf4,0xfc,0x18,0x81,0xc7,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: addq $123, %rdi, %rax # encoding: [0x62,0xf4,0xfc,0x18,0x83,0xc7,0x7b]
; CHECK-NEXT: retq # encoding: [0xc3]
entry:
%add = add i64 %a, 123
Expand Down Expand Up @@ -221,7 +221,7 @@ entry:
define i32 @add32mi8(ptr %a) {
; CHECK-LABEL: add32mi8:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: addl $123, (%rdi), %eax # encoding: [0x62,0xf4,0x7c,0x18,0x81,0x07,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: addl $123, (%rdi), %eax # encoding: [0x62,0xf4,0x7c,0x18,0x83,0x07,0x7b]
; CHECK-NEXT: retq # encoding: [0xc3]
entry:
%t= load i32, ptr %a
Expand All @@ -232,7 +232,7 @@ entry:
define i64 @add64mi8(ptr %a) {
; CHECK-LABEL: add64mi8:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: addq $123, (%rdi), %rax # encoding: [0x62,0xf4,0xfc,0x18,0x81,0x07,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: addq $123, (%rdi), %rax # encoding: [0x62,0xf4,0xfc,0x18,0x83,0x07,0x7b]
; CHECK-NEXT: retq # encoding: [0xc3]
entry:
%t= load i64, ptr %a
Expand Down Expand Up @@ -405,7 +405,7 @@ entry:
define i16 @addflag16ri8(i16 noundef %a) {
; CHECK-LABEL: addflag16ri8:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: addw $123, %di, %cx # encoding: [0x62,0xf4,0x75,0x18,0x81,0xc7,0x7b,0x00]
; CHECK-NEXT: addw $123, %di, %cx # encoding: [0x62,0xf4,0x75,0x18,0x83,0xc7,0x7b]
; CHECK-NEXT: movl $65535, %eax # encoding: [0xb8,0xff,0xff,0x00,0x00]
; CHECK-NEXT: # imm = 0xFFFF
; CHECK-NEXT: cmovael %ecx, %eax # encoding: [0x0f,0x43,0xc1]
Expand All @@ -419,7 +419,7 @@ entry:
define i32 @addflag32ri8(i32 noundef %a) {
; CHECK-LABEL: addflag32ri8:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: addl $123, %edi, %ecx # encoding: [0x62,0xf4,0x74,0x18,0x81,0xc7,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: addl $123, %edi, %ecx # encoding: [0x62,0xf4,0x74,0x18,0x83,0xc7,0x7b]
; CHECK-NEXT: movl $-1, %eax # encoding: [0xb8,0xff,0xff,0xff,0xff]
; CHECK-NEXT: cmovael %ecx, %eax # encoding: [0x0f,0x43,0xc1]
; CHECK-NEXT: retq # encoding: [0xc3]
Expand All @@ -431,7 +431,7 @@ entry:
define i64 @addflag64ri8(i64 noundef %a) {
; CHECK-LABEL: addflag64ri8:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: addq $123, %rdi, %rcx # encoding: [0x62,0xf4,0xf4,0x18,0x81,0xc7,0x7b,0x00,0x00,0x00]
; CHECK-NEXT: addq $123, %rdi, %rcx # encoding: [0x62,0xf4,0xf4,0x18,0x83,0xc7,0x7b]
; CHECK-NEXT: movq $-1, %rax # encoding: [0x48,0xc7,0xc0,0xff,0xff,0xff,0xff]
; CHECK-NEXT: cmovaeq %rcx, %rax # encoding: [0x48,0x0f,0x43,0xc1]
; CHECK-NEXT: retq # encoding: [0xc3]
Expand Down
Loading