[llvm] [AArch64] Remove redundant FMOV for zero-extended i32/i16 loads to f64 (PR #146920)
Amina Chabane via llvm-commits
llvm-commits at lists.llvm.org
Mon Jul 14 03:33:24 PDT 2025
https://github.com/Amichaxx updated https://github.com/llvm/llvm-project/pull/146920
>From b5757cb6c0525d9ff249475ff2164219cff6e0f5 Mon Sep 17 00:00:00 2001
From: Amina Chabane <amina.chabane at arm.com>
Date: Fri, 4 Jul 2025 10:31:38 +0000
Subject: [PATCH 1/2] [AArch64] Remove redundant FMOV for zero-extended i32/i16
loads to f64
Previously, a load from i32 or i16, followed by zero-extension to i64 and a
bitcast to f64, would emit separate FMOV instructions.
This patch introduces new corresponding TableGen patterns to avoid the unnecessary FMOV.
Tests added:
- load_u64_from_u32.ll
- load_u64_from_u16.ll
---
llvm/lib/Target/AArch64/AArch64InstrInfo.td | 10 +++++++++-
llvm/test/CodeGen/AArch64/load_u64_from_u16.ll | 15 +++++++++++++++
llvm/test/CodeGen/AArch64/load_u64_from_u32.ll | 14 ++++++++++++++
3 files changed, 38 insertions(+), 1 deletion(-)
create mode 100644 llvm/test/CodeGen/AArch64/load_u64_from_u16.ll
create mode 100644 llvm/test/CodeGen/AArch64/load_u64_from_u32.ll
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index efe6cc1aa8aec..fb2c7c9f6adfe 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -3913,6 +3913,14 @@ defm LDRSW : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
(SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
+// load zero-extended i32, bitcast to f64
+def : Pat <(f64 (bitconvert (i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
+ (SUBREG_TO_REG (i64 0), (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
+
+// load zero-extended i16, bitcast to f64
+def : Pat <(f64 (bitconvert (i64 (zextloadi16 (am_indexed32 GPR64sp:$Rn, uimm12s2:$offset))))),
+ (SUBREG_TO_REG (i64 0), (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
+
// Pre-fetch.
def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
[(AArch64Prefetch timm:$Rt,
@@ -10986,4 +10994,4 @@ defm FMMLA : SIMDThreeSameVectorFP8MatrixMul<"fmmla">;
include "AArch64InstrAtomics.td"
include "AArch64SVEInstrInfo.td"
include "AArch64SMEInstrInfo.td"
-include "AArch64InstrGISel.td"
+include "AArch64InstrGISel.td"
\ No newline at end of file
diff --git a/llvm/test/CodeGen/AArch64/load_u64_from_u16.ll b/llvm/test/CodeGen/AArch64/load_u64_from_u16.ll
new file mode 100644
index 0000000000000..ee02895b55221
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/load_u64_from_u16.ll
@@ -0,0 +1,15 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=aarch64-linux-gnu -o - %s | FileCheck %s
+
+define double @_Z9load_u64_from_u16_testPj(ptr %n){
+; CHECK-LABEL: _Z9load_u64_from_u16_testPj:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: ldr h0, [x0]
+; CHECK-NEXT: ret
+entry:
+ %0 = load i16, ptr %n, align 2
+ %conv = zext i16 %0 to i64
+ %1 = bitcast i64 %conv to double
+ ret double %1
+}
+
diff --git a/llvm/test/CodeGen/AArch64/load_u64_from_u32.ll b/llvm/test/CodeGen/AArch64/load_u64_from_u32.ll
new file mode 100644
index 0000000000000..ad30981012112
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/load_u64_from_u32.ll
@@ -0,0 +1,14 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=aarch64-linux-gnu -o - %s | FileCheck %s
+
+define double @_Z9load_u64_from_u32_testPj(ptr %n) {
+; CHECK-LABEL: _Z9load_u64_from_u32_testPj:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: ldr s0, [x0]
+; CHECK-NEXT: ret
+entry:
+ %0 = load i32, ptr %n, align 4
+ %conv = zext i32 %0 to i64
+ %1 = bitcast i64 %conv to double
+ ret double %1
+}
>From fabb75dfc3b041cfdba1ad9792fa2fdf44857bec Mon Sep 17 00:00:00 2001
From: Amina Chabane <amina.chabane at arm.com>
Date: Mon, 7 Jul 2025 15:30:07 +0000
Subject: [PATCH 2/2] [AArch64] This patch removes redundant FMOV instructions
for patterns where an integer is loaded from memory, zero-extended, and
bitcast to a floating-point type. Patterns are added for i8/i16/i32 to f64
and f32. A single test file (load-zext-bitcast.ll) is included for all cases,
which combines the previously created test files.
---
llvm/lib/Target/AArch64/AArch64InstrInfo.td | 14 +++-
.../test/CodeGen/AArch64/load-zext-bitcast.ll | 82 +++++++++++++++++++
.../test/CodeGen/AArch64/load_u64_from_u16.ll | 15 ----
.../test/CodeGen/AArch64/load_u64_from_u32.ll | 14 ----
4 files changed, 95 insertions(+), 30 deletions(-)
create mode 100644 llvm/test/CodeGen/AArch64/load-zext-bitcast.ll
delete mode 100644 llvm/test/CodeGen/AArch64/load_u64_from_u16.ll
delete mode 100644 llvm/test/CodeGen/AArch64/load_u64_from_u32.ll
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index fb2c7c9f6adfe..87e151a04ceae 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -3920,7 +3920,19 @@ def : Pat <(f64 (bitconvert (i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s
// load zero-extended i16, bitcast to f64
def : Pat <(f64 (bitconvert (i64 (zextloadi16 (am_indexed32 GPR64sp:$Rn, uimm12s2:$offset))))),
(SUBREG_TO_REG (i64 0), (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
-
+
+// load zero-extended i8, bitcast to f64
+def : Pat <(f64 (bitconvert (i64 (zextloadi8 (am_indexed32 GPR64sp:$Rn, uimm12s1:$offset))))),
+ (SUBREG_TO_REG (i64 0), (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
+
+// load zero-extended i16, bitcast to f32
+def : Pat <(f32 (bitconvert (i32 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
+ (SUBREG_TO_REG (i32 0), (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
+
+// load zero-extended i8, bitcast to f32
+def : Pat <(f32 (bitconvert (i32 (zextloadi8 (am_indexed16 GPR64sp:$Rn, uimm12s1:$offset))))),
+ (SUBREG_TO_REG (i32 0), (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
+
// Pre-fetch.
def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
[(AArch64Prefetch timm:$Rt,
diff --git a/llvm/test/CodeGen/AArch64/load-zext-bitcast.ll b/llvm/test/CodeGen/AArch64/load-zext-bitcast.ll
new file mode 100644
index 0000000000000..1a83930dfafa2
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/load-zext-bitcast.ll
@@ -0,0 +1,82 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=aarch64-linux-gnu -o - %s | FileCheck %s
+
+; load zero-extended i32, bitcast to f64
+define double @_Z9load_u64_from_u32_testPj(ptr %n){
+; CHECK-LABEL: _Z9load_u64_from_u32_testPj:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: ldr s0, [x0]
+; CHECK-NEXT: ret
+entry:
+ %0 = load i32, ptr %n, align 4
+ %conv = zext i32 %0 to i64
+ %1 = bitcast i64 %conv to double
+ ret double %1
+}
+
+; load zero-extended i16, bitcast to f64
+define double @_Z9load_u64_from_u16_testPj(ptr %n){
+; CHECK-LABEL: _Z9load_u64_from_u16_testPj:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: ldr h0, [x0]
+; CHECK-NEXT: ret
+entry:
+ %0 = load i16, ptr %n, align 2
+ %conv = zext i16 %0 to i64
+ %1 = bitcast i64 %conv to double
+ ret double %1
+}
+
+; load zero-extended i8, bitcast to f64
+define double @_Z16load_u64_from_u8Ph(ptr %n){
+; CHECK-LABEL: _Z16load_u64_from_u8Ph:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: ldr b0, [x0]
+; CHECK-NEXT: ret
+entry:
+ %0 = load i8, ptr %n, align 1
+ %conv = zext i8 %0 to i64
+ %1 = bitcast i64 %conv to double
+ ret double %1
+}
+
+; load zero-extended i16, bitcast to f32
+define float @_Z17load_u32_from_u16Pt(ptr %n){
+; CHECK-LABEL: _Z17load_u32_from_u16Pt:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: ldr h0, [x0]
+; CHECK-NEXT: ret
+entry:
+ %0 = load i16, ptr %n, align 2
+ %conv = zext i16 %0 to i32
+ %1 = bitcast i32 %conv to float
+ ret float %1
+}
+
+; load zero-extended i8, bitcast to f32
+define float @_Z16load_u32_from_u8Ph(ptr %n){
+; CHECK-LABEL: _Z16load_u32_from_u8Ph:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: ldr b0, [x0]
+; CHECK-NEXT: ret
+entry:
+ %0 = load i8, ptr %n, align 1
+ %conv = zext i8 %0 to i32
+ %1 = bitcast i32 %conv to float
+ ret float %1
+}
+
+; load zero-extended i8, bitcast to f16
+define half @_Z16load_u16_from_u8Ph(ptr %n){
+; CHECK-LABEL: _Z16load_u16_from_u8Ph:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: ldr b0, [x0]
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $s0
+; CHECK-NEXT: ret
+entry:
+ %0 = load i8, ptr %n, align 1
+ %conv = zext i8 %0 to i16
+ %1 = bitcast i16 %conv to half
+ ret half %1
+}
+
diff --git a/llvm/test/CodeGen/AArch64/load_u64_from_u16.ll b/llvm/test/CodeGen/AArch64/load_u64_from_u16.ll
deleted file mode 100644
index ee02895b55221..0000000000000
--- a/llvm/test/CodeGen/AArch64/load_u64_from_u16.ll
+++ /dev/null
@@ -1,15 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc -mtriple=aarch64-linux-gnu -o - %s | FileCheck %s
-
-define double @_Z9load_u64_from_u16_testPj(ptr %n){
-; CHECK-LABEL: _Z9load_u64_from_u16_testPj:
-; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: ldr h0, [x0]
-; CHECK-NEXT: ret
-entry:
- %0 = load i16, ptr %n, align 2
- %conv = zext i16 %0 to i64
- %1 = bitcast i64 %conv to double
- ret double %1
-}
-
diff --git a/llvm/test/CodeGen/AArch64/load_u64_from_u32.ll b/llvm/test/CodeGen/AArch64/load_u64_from_u32.ll
deleted file mode 100644
index ad30981012112..0000000000000
--- a/llvm/test/CodeGen/AArch64/load_u64_from_u32.ll
+++ /dev/null
@@ -1,14 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc -mtriple=aarch64-linux-gnu -o - %s | FileCheck %s
-
-define double @_Z9load_u64_from_u32_testPj(ptr %n) {
-; CHECK-LABEL: _Z9load_u64_from_u32_testPj:
-; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: ldr s0, [x0]
-; CHECK-NEXT: ret
-entry:
- %0 = load i32, ptr %n, align 4
- %conv = zext i32 %0 to i64
- %1 = bitcast i64 %conv to double
- ret double %1
-}
More information about the llvm-commits
mailing list