OpcodeDispatcher: Improve output of SHLX/SHRX/SARX

We can remove some unnecessary moves for the 32-bit cases and
collapse the operations down to a single instruction.
This commit is contained in:
Lioncache committed 2023-09-15 21:05:50 -04:00
1 parent 4604c01986
commit e9be291cec
2 files changed
+77 -10

No files matched your search

@@ -2627,12 +2627,16 @@ void OpDispatchBuilder::BLSRBMIOp(OpcodeArgs) {
GenerateFlags_BLSR(Op, Result, Src);
}
// Handles SARX, SHLX, and SHRX
void OpDispatchBuilder::BMI2Shift(OpcodeArgs) {
// Handles SARX, SHLX, and SHRX
auto* Src = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, -1);
auto* Shift = LoadSource(GPRClass, Op, Op->Src[1], Op->Flags, -1);
// In the event the source is a memory operand, use the
// exact width instead of the GPR size.
const auto GPRSize = CTX->GetGPRSize();
const auto Size = GetSrcSize(Op);
const auto SrcSize = Op->Src[0].IsGPR() ? GPRSize : Size;
auto* Src = LoadSource_WithOpSize(GPRClass, Op, Op->Src[0], SrcSize, Op->Flags, -1);
auto* Shift = LoadSource_WithOpSize(GPRClass, Op, Op->Src[1], GPRSize, Op->Flags, -1);
auto* Result = [&]() -> OrderedNode* {
// SARX
+69 -6
View File
@@ -3985,6 +3985,16 @@
]
},
"shlx eax, ebx, ecx": {
"ExpectedInstructionCount": 1,
"Optimal": "Yes",
"Comment": [
"Map 2 0b01 0xf7 32-bit"
],
"ExpectedArm64ASM": [
"lsl w4, w7, w5"
]
},
"shlx eax, [ebx], ecx": {
"ExpectedInstructionCount": 3,
"Optimal": "No",
"Comment": [
@@ -3992,8 +4002,8 @@
],
"ExpectedArm64ASM": [
"mov w20, w7",
"mov w21, w5",
"lsl w4, w20, w21"
"ldr w20, [x20]",
"lsl w4, w20, w5"
]
},
"shlx rax, rbx, rcx": {
@@ -4006,7 +4016,28 @@
"lsl x4, x7, x5"
]
},
"shlx rax, [rbx], rcx": {
"ExpectedInstructionCount": 2,
"Optimal": "Yes",
"Comment": [
"Map 2 0b01 0xf7 64-bit"
],
"ExpectedArm64ASM": [
"ldr x20, [x7]",
"lsl x4, x20, x5"
]
},
"sarx eax, ebx, ecx": {
"ExpectedInstructionCount": 1,
"Optimal": "Yes",
"Comment": [
"Map 2 0b10 0xf7 32-bit"
],
"ExpectedArm64ASM": [
"asr w4, w7, w5"
]
},
"sarx eax, [ebx], ecx": {
"ExpectedInstructionCount": 3,
"Optimal": "No",
"Comment": [
@@ -4014,8 +4045,8 @@
],
"ExpectedArm64ASM": [
"mov w20, w7",
"mov w21, w5",
"asr w4, w20, w21"
"ldr w20, [x20]",
"asr w4, w20, w5"
]
},
"sarx rax, rbx, rcx": {
@@ -4028,7 +4059,28 @@
"asr x4, x7, x5"
]
},
"sarx rax, [rbx], rcx": {
"ExpectedInstructionCount": 2,
"Optimal": "Yes",
"Comment": [
"Map 2 0b10 0xf7 64-bit"
],
"ExpectedArm64ASM": [
"ldr x20, [x7]",
"asr x4, x20, x5"
]
},
"shrx eax, ebx, ecx": {
"ExpectedInstructionCount": 1,
"Optimal": "Yes",
"Comment": [
"Map 2 0b11 0xf7 32-bit"
],
"ExpectedArm64ASM": [
"lsr w4, w7, w5"
]
},
"shrx eax, [ebx], ecx": {
"ExpectedInstructionCount": 3,
"Optimal": "No",
"Comment": [
@@ -4036,8 +4088,8 @@
],
"ExpectedArm64ASM": [
"mov w20, w7",
"mov w21, w5",
"lsr w4, w20, w21"
"ldr w20, [x20]",
"lsr w4, w20, w5"
]
},
"shrx rax, rbx, rcx": {
@@ -4049,6 +4101,17 @@
"ExpectedArm64ASM": [
"lsr x4, x7, x5"
]
},
"shrx rax, [rbx], rcx": {
"ExpectedInstructionCount": 2,
"Optimal": "Yes",
"Comment": [
"Map 2 0b11 0xf7 64-bit"
],
"ExpectedArm64ASM": [
"ldr x20, [x7]",
"lsr x4, x20, x5"
]
}
}
}