summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--src/passes/OptimizeInstructions.cpp26
-rw-r--r--test/passes/optimize-instructions.txt67
-rw-r--r--test/passes/optimize-instructions.wast75
-rw-r--r--test/unit.fromasm11
-rw-r--r--test/unit.fromasm.clamp11
-rw-r--r--test/unit.fromasm.imprecise11
6 files changed, 178 insertions, 23 deletions
diff --git a/src/passes/OptimizeInstructions.cpp b/src/passes/OptimizeInstructions.cpp
index 02475e4dc..ad828dafc 100644
--- a/src/passes/OptimizeInstructions.cpp
+++ b/src/passes/OptimizeInstructions.cpp
@@ -404,13 +404,15 @@ struct OptimizeInstructions : public WalkerPass<PostWalker<OptimizeInstructions,
if (auto* ext = Properties::getAlmostSignExt(binary)) {
Index extraShifts;
auto bits = Properties::getAlmostSignExtBits(binary, extraShifts);
- if (auto* load = getFallthrough(ext)->dynCast<Load>()) {
- // pattern match a load of 8 bits and a sign extend using a shl of 24 then shr_s of 24 as well, etc.
- if ((load->bytes == 1 && bits == 8) || (load->bytes == 2 && bits == 16)) {
- // if the value falls through, we can't alter the load, as it might be captured in a tee
- if (load->signed_ == true || load == ext) {
- load->signed_ = true;
- return removeAlmostSignExt(binary);
+ if (extraShifts == 0) {
+ if (auto* load = getFallthrough(ext)->dynCast<Load>()) {
+ // pattern match a load of 8 bits and a sign extend using a shl of 24 then shr_s of 24 as well, etc.
+ if ((load->bytes == 1 && bits == 8) || (load->bytes == 2 && bits == 16)) {
+ // if the value falls through, we can't alter the load, as it might be captured in a tee
+ if (load->signed_ == true || load == ext) {
+ load->signed_ = true;
+ return ext;
+ }
}
}
}
@@ -435,11 +437,13 @@ struct OptimizeInstructions : public WalkerPass<PostWalker<OptimizeInstructions,
}
} else if (auto* left = Properties::getSignExtValue(binary->left)) {
if (auto* right = Properties::getSignExtValue(binary->right)) {
- // we are comparing two sign-exts, so we may as well replace both with cheaper zexts
auto bits = Properties::getSignExtBits(binary->left);
- binary->left = makeZeroExt(left, bits);
- binary->right = makeZeroExt(right, bits);
- return binary;
+ if (Properties::getSignExtBits(binary->right) == bits) {
+ // we are comparing two sign-exts with the same bits, so we may as well replace both with cheaper zexts
+ binary->left = makeZeroExt(left, bits);
+ binary->right = makeZeroExt(right, bits);
+ return binary;
+ }
} else if (auto* load = binary->right->dynCast<Load>()) {
// we are comparing a load to a sign-ext, we may be able to switch to zext
auto leftBits = Properties::getSignExtBits(binary->left);
diff --git a/test/passes/optimize-instructions.txt b/test/passes/optimize-instructions.txt
index f40006f09..41d45d57f 100644
--- a/test/passes/optimize-instructions.txt
+++ b/test/passes/optimize-instructions.txt
@@ -5,6 +5,7 @@
(type $3 (func (param i32) (result i32)))
(type $4 (func (param i32 i32)))
(type $5 (func (param i32)))
+ (type $6 (func (param i32 i32) (result i32)))
(memory $0 0)
(export "load-off-2" (func $load-off-2))
(func $f (type $0) (param $i1 i32) (param $i2 i64)
@@ -1723,4 +1724,70 @@
)
)
)
+ (func $unsign-diff-sizes (type $6) (param $x i32) (param $y i32) (result i32)
+ (i32.ne
+ (i32.shr_s
+ (i32.shl
+ (call $unsign-diff-sizes
+ (i32.const -1)
+ (i32.const 5)
+ )
+ (i32.const 24)
+ )
+ (i32.const 24)
+ )
+ (i32.shr_s
+ (i32.shl
+ (call $unsign-diff-sizes
+ (i32.const 1)
+ (i32.const 2006)
+ )
+ (i32.const 16)
+ )
+ (i32.const 16)
+ )
+ )
+ )
+ (func $unsign-same-sizes (type $6) (param $x i32) (param $y i32) (result i32)
+ (i32.ne
+ (i32.and
+ (call $unsign-same-sizes
+ (i32.const -1)
+ (i32.const 5)
+ )
+ (i32.const 255)
+ )
+ (i32.and
+ (call $unsign-same-sizes
+ (i32.const 1)
+ (i32.const 2006)
+ )
+ (i32.const 255)
+ )
+ )
+ )
+ (func $fuzz-almost-sign-ext (type $1)
+ (drop
+ (i32.shr_s
+ (i32.shl
+ (i32.load16_u
+ (i32.const 2278)
+ )
+ (i32.const 17)
+ )
+ (i32.const 16)
+ )
+ )
+ (drop
+ (i32.shr_s
+ (i32.shl
+ (i32.load16_u
+ (i32.const 2278)
+ )
+ (i32.const 17)
+ )
+ (i32.const 16)
+ )
+ )
+ )
)
diff --git a/test/passes/optimize-instructions.wast b/test/passes/optimize-instructions.wast
index 51bd076d1..8ef6f67a5 100644
--- a/test/passes/optimize-instructions.wast
+++ b/test/passes/optimize-instructions.wast
@@ -2106,4 +2106,79 @@
)
)
)
+ (func $unsign-diff-sizes (param $x i32) (param $y i32) (result i32)
+ (i32.ne
+ (i32.shr_s
+ (i32.shl
+ (call $unsign-diff-sizes
+ (i32.const -1)
+ (i32.const 5)
+ )
+ (i32.const 24)
+ )
+ (i32.const 24)
+ )
+ (i32.shr_s
+ (i32.shl
+ (call $unsign-diff-sizes
+ (i32.const 1)
+ (i32.const 2006)
+ )
+ (i32.const 16)
+ )
+ (i32.const 16)
+ )
+ )
+ )
+ (func $unsign-same-sizes (param $x i32) (param $y i32) (result i32)
+ (i32.ne
+ (i32.shr_s
+ (i32.shl
+ (call $unsign-same-sizes
+ (i32.const -1)
+ (i32.const 5)
+ )
+ (i32.const 24)
+ )
+ (i32.const 24)
+ )
+ (i32.shr_s
+ (i32.shl
+ (call $unsign-same-sizes
+ (i32.const 1)
+ (i32.const 2006)
+ )
+ (i32.const 24)
+ )
+ (i32.const 24)
+ )
+ )
+ )
+ (func $fuzz-almost-sign-ext
+ (drop
+ (i32.shr_s
+ (i32.shl
+ (i32.load16_u
+ (i32.const 2278)
+ )
+ (i32.const 17)
+ )
+ (i32.const 16)
+ )
+ )
+ (drop
+ (i32.shr_s
+ (i32.shl
+ (i32.shl
+ (i32.load16_u
+ (i32.const 2278)
+ )
+ (i32.const 1)
+ )
+ (i32.const 16)
+ )
+ (i32.const 16)
+ )
+ )
+ )
)
diff --git a/test/unit.fromasm b/test/unit.fromasm
index 58272fae8..fd0fb27a0 100644
--- a/test/unit.fromasm
+++ b/test/unit.fromasm
@@ -696,11 +696,14 @@
)
)
(call $loadSigned
- (i32.shl
- (i32.load16_s
- (get_local $0)
+ (i32.shr_s
+ (i32.shl
+ (i32.load16_u
+ (get_local $0)
+ )
+ (i32.const 24)
)
- (i32.const 8)
+ (i32.const 16)
)
)
)
diff --git a/test/unit.fromasm.clamp b/test/unit.fromasm.clamp
index 645f65e3e..ffe26a3a2 100644
--- a/test/unit.fromasm.clamp
+++ b/test/unit.fromasm.clamp
@@ -720,11 +720,14 @@
)
)
(call $loadSigned
- (i32.shl
- (i32.load16_s
- (get_local $0)
+ (i32.shr_s
+ (i32.shl
+ (i32.load16_u
+ (get_local $0)
+ )
+ (i32.const 24)
)
- (i32.const 8)
+ (i32.const 16)
)
)
)
diff --git a/test/unit.fromasm.imprecise b/test/unit.fromasm.imprecise
index 0b03e0016..e94c07f7f 100644
--- a/test/unit.fromasm.imprecise
+++ b/test/unit.fromasm.imprecise
@@ -664,11 +664,14 @@
)
)
(call $loadSigned
- (i32.shl
- (i32.load16_s
- (get_local $0)
+ (i32.shr_s
+ (i32.shl
+ (i32.load16_u
+ (get_local $0)
+ )
+ (i32.const 24)
)
- (i32.const 8)
+ (i32.const 16)
)
)
)
generated by cgit v1.2.3 (git 2.39.1) at 2025-08-08 13:41:33 +0000