summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--src/passes/OptimizeInstructions.cpp18
-rw-r--r--test/emcc_hello_world.fromasm38
-rw-r--r--test/emcc_hello_world.fromasm.imprecise38
-rw-r--r--test/passes/optimize-instructions.txt80
-rw-r--r--test/passes/optimize-instructions.wast86
5 files changed, 210 insertions, 50 deletions
diff --git a/src/passes/OptimizeInstructions.cpp b/src/passes/OptimizeInstructions.cpp
index 026f3f7ec..49d025f82 100644
--- a/src/passes/OptimizeInstructions.cpp
+++ b/src/passes/OptimizeInstructions.cpp
@@ -515,6 +515,24 @@ struct OptimizeInstructions : public WalkerPass<PostWalker<OptimizeInstructions,
binary->left = makeZeroExt(left, bits);
binary->right = makeZeroExt(right, bits);
return binary;
+ } else if (auto* load = getFallthroughDynCast<Load>(binary->right)) {
+ // we are comparing a load to a sign-ext, we may be able to switch to zext
+ auto leftBits = getSignExtBits(binary->left);
+ if (load->signed_ && leftBits == load->bytes * 8) {
+ load->signed_ = false;
+ binary->left = makeZeroExt(left, leftBits);
+ return binary;
+ }
+ }
+ } else if (auto* load = getFallthroughDynCast<Load>(binary->left)) {
+ if (auto* right = getSignExt(binary->right)) {
+ // we are comparing a load to a sign-ext, we may be able to switch to zext
+ auto rightBits = getSignExtBits(binary->right);
+ if (load->signed_ && rightBits == load->bytes * 8) {
+ load->signed_ = false;
+ binary->right = makeZeroExt(right, rightBits);
+ return binary;
+ }
}
}
// note that both left and right may be consts, but then we let precompute compute the constant result
diff --git a/test/emcc_hello_world.fromasm b/test/emcc_hello_world.fromasm
index c314d887c..ecbadd304 100644
--- a/test/emcc_hello_world.fromasm
+++ b/test/emcc_hello_world.fromasm
@@ -1864,15 +1864,12 @@
(loop $while-in
(if
(i32.eq
- (i32.load8_s
+ (i32.load8_u
(get_local $2)
)
- (i32.shr_s
- (i32.shl
- (get_local $4)
- (i32.const 24)
- )
- (i32.const 24)
+ (i32.and
+ (get_local $4)
+ (i32.const 255)
)
)
(block
@@ -1941,20 +1938,14 @@
)
(if
(i32.ne
- (i32.load8_s
+ (i32.load8_u
(get_local $2)
)
- (i32.shr_s
- (i32.shl
- (tee_local $1
- (i32.and
- (get_local $1)
- (i32.const 255)
- )
- )
- (i32.const 24)
+ (tee_local $1
+ (i32.and
+ (get_local $1)
+ (i32.const 255)
)
- (i32.const 24)
)
)
(block
@@ -2035,15 +2026,12 @@
(loop $while-in5
(br_if $label$break$L8
(i32.eq
- (i32.load8_s
+ (i32.load8_u
(get_local $2)
)
- (i32.shr_s
- (i32.shl
- (get_local $1)
- (i32.const 24)
- )
- (i32.const 24)
+ (i32.and
+ (get_local $1)
+ (i32.const 255)
)
)
)
diff --git a/test/emcc_hello_world.fromasm.imprecise b/test/emcc_hello_world.fromasm.imprecise
index 1fa81ba43..52fa498da 100644
--- a/test/emcc_hello_world.fromasm.imprecise
+++ b/test/emcc_hello_world.fromasm.imprecise
@@ -1861,15 +1861,12 @@
(loop $while-in
(if
(i32.eq
- (i32.load8_s
+ (i32.load8_u
(get_local $2)
)
- (i32.shr_s
- (i32.shl
- (get_local $4)
- (i32.const 24)
- )
- (i32.const 24)
+ (i32.and
+ (get_local $4)
+ (i32.const 255)
)
)
(block
@@ -1938,20 +1935,14 @@
)
(if
(i32.ne
- (i32.load8_s
+ (i32.load8_u
(get_local $2)
)
- (i32.shr_s
- (i32.shl
- (tee_local $1
- (i32.and
- (get_local $1)
- (i32.const 255)
- )
- )
- (i32.const 24)
+ (tee_local $1
+ (i32.and
+ (get_local $1)
+ (i32.const 255)
)
- (i32.const 24)
)
)
(block
@@ -2032,15 +2023,12 @@
(loop $while-in5
(br_if $label$break$L8
(i32.eq
- (i32.load8_s
+ (i32.load8_u
(get_local $2)
)
- (i32.shr_s
- (i32.shl
- (get_local $1)
- (i32.const 24)
- )
- (i32.const 24)
+ (i32.and
+ (get_local $1)
+ (i32.const 255)
)
)
)
diff --git a/test/passes/optimize-instructions.txt b/test/passes/optimize-instructions.txt
index 550859185..d70e6a9c5 100644
--- a/test/passes/optimize-instructions.txt
+++ b/test/passes/optimize-instructions.txt
@@ -1643,4 +1643,84 @@
)
)
)
+ (func $compare-load-s-sign-extend (type $4) (param $0 i32) (param $1 i32)
+ (drop
+ (i32.eq
+ (i32.load8_u
+ (get_local $0)
+ )
+ (i32.and
+ (get_local $1)
+ (i32.const 255)
+ )
+ )
+ )
+ (drop
+ (i32.eq
+ (i32.and
+ (get_local $1)
+ (i32.const 255)
+ )
+ (i32.load8_u
+ (get_local $0)
+ )
+ )
+ )
+ (drop
+ (i32.eq
+ (i32.load8_u
+ (get_local $0)
+ )
+ (i32.shr_s
+ (i32.shl
+ (get_local $1)
+ (i32.const 24)
+ )
+ (i32.const 24)
+ )
+ )
+ )
+ (drop
+ (i32.eq
+ (i32.load8_s
+ (get_local $0)
+ )
+ (i32.shr_s
+ (i32.shl
+ (get_local $1)
+ (i32.const 16)
+ )
+ (i32.const 16)
+ )
+ )
+ )
+ (drop
+ (i32.eq
+ (i32.shr_s
+ (i32.shl
+ (get_local $1)
+ (i32.const 24)
+ )
+ (i32.const 24)
+ )
+ (i32.load8_u
+ (get_local $0)
+ )
+ )
+ )
+ (drop
+ (i32.eq
+ (i32.shr_s
+ (i32.shl
+ (get_local $1)
+ (i32.const 16)
+ )
+ (i32.const 16)
+ )
+ (i32.load8_s
+ (get_local $0)
+ )
+ )
+ )
+ )
)
diff --git a/test/passes/optimize-instructions.wast b/test/passes/optimize-instructions.wast
index 2f2333867..51bd076d1 100644
--- a/test/passes/optimize-instructions.wast
+++ b/test/passes/optimize-instructions.wast
@@ -2020,4 +2020,90 @@
)
)
)
+ (func $compare-load-s-sign-extend (param $0 i32) (param $1 i32)
+ (drop
+ (i32.eq
+ (i32.load8_s
+ (get_local $0)
+ )
+ (i32.shr_s
+ (i32.shl
+ (get_local $1)
+ (i32.const 24)
+ )
+ (i32.const 24)
+ )
+ )
+ )
+ (drop
+ (i32.eq
+ (i32.shr_s
+ (i32.shl
+ (get_local $1)
+ (i32.const 24)
+ )
+ (i32.const 24)
+ )
+ (i32.load8_s
+ (get_local $0) ;; flip order, we should canonicalize
+ )
+ )
+ )
+ (drop
+ (i32.eq
+ (i32.load8_u ;; unsigned, bad
+ (get_local $0)
+ )
+ (i32.shr_s
+ (i32.shl
+ (get_local $1)
+ (i32.const 24)
+ )
+ (i32.const 24)
+ )
+ )
+ )
+ (drop
+ (i32.eq
+ (i32.load8_s
+ (get_local $0)
+ )
+ (i32.shr_s
+ (i32.shl
+ (get_local $1)
+ (i32.const 16) ;; wrong size
+ )
+ (i32.const 16)
+ )
+ )
+ )
+ (drop
+ (i32.eq
+ (i32.shr_s
+ (i32.shl
+ (get_local $1)
+ (i32.const 24)
+ )
+ (i32.const 24)
+ )
+ (i32.load8_u ;; unsigned, bad
+ (get_local $0)
+ )
+ )
+ )
+ (drop
+ (i32.eq
+ (i32.shr_s
+ (i32.shl
+ (get_local $1)
+ (i32.const 16) ;; wrong size
+ )
+ (i32.const 16)
+ )
+ (i32.load8_s
+ (get_local $0)
+ )
+ )
+ )
+ )
)
generated by cgit v1.2.3 (git 2.39.1) at 2025-08-07 22:12:06 +0000