diff --git a/core/c/libc/stdatomic.odin b/core/c/libc/stdatomic.odin
index fbe1ef7ea..9de3c4678 100644
--- a/core/c/libc/stdatomic.odin
+++ b/core/c/libc/stdatomic.odin
@@ -47,24 +47,25 @@ kill_dependency :: #force_inline proc(value: $T) -> T {
 
 // 7.17.4 Fences
 atomic_thread_fence :: #force_inline proc(order: memory_order) {
-	switch (order) {
-	case .relaxed:
-		return
-	case .consume:
-		intrinsics.atomic_fence_acq()
-	case .acquire:
-		intrinsics.atomic_fence_acq()
-	case .release:
-		intrinsics.atomic_fence_rel()
-	case .acq_rel:
-		intrinsics.atomic_fence_acqrel()
-	case .seq_cst:
-		intrinsics.atomic_fence_acqrel()
+	assert(order != .relaxed)
+	assert(order != .consume)
+	#partial switch order {
+	case .acquire: intrinsics.atomic_thread_fence(.Acquire)
+	case .release: intrinsics.atomic_thread_fence(.Release)
+	case .acq_rel: intrinsics.atomic_thread_fence(.Acq_Rel)
+	case .seq_cst: intrinsics.atomic_thread_fence(.Seq_Cst)
 	}
 }
 
 atomic_signal_fence :: #force_inline proc(order: memory_order) {
-	atomic_thread_fence(order)
+	assert(order != .relaxed)
+	assert(order != .consume)
+	#partial switch order {
+	case .acquire: intrinsics.atomic_signal_fence(.Acquire)
+	case .release: intrinsics.atomic_signal_fence(.Release)
+	case .acq_rel: intrinsics.atomic_signal_fence(.Acq_Rel)
+	case .seq_cst: intrinsics.atomic_signal_fence(.Seq_Cst)
+	}
 }
 
 // 7.17.5 Lock-free property
@@ -121,13 +122,10 @@ atomic_store_explicit :: #force_inline proc(object: ^$T, desired: T, order: memo
 	assert(order != .acquire)
 	assert(order != .acq_rel)
 
-	#partial switch (order) {
-	case .relaxed:
-		intrinsics.atomic_store_relaxed(object, desired)
-	case .release:
-		intrinsics.atomic_store_rel(object, desired)
-	case .seq_cst:
-		intrinsics.atomic_store(object, desired)
+	#partial switch order {
+	case .relaxed: intrinsics.atomic_store_explicit(object, desired, .Relaxed)
+	case .release: intrinsics.atomic_store_explicit(object, desired, .Release)
+	case .seq_cst: intrinsics.atomic_store_explicit(object, desired, .Seq_Cst)
 	}
 }
 
@@ -139,36 +137,26 @@ atomic_load_explicit :: #force_inline proc(object: ^$T, order: memory_order) {
 	assert(order != .release)
 	assert(order != .acq_rel)
 
-	#partial switch (order) {
-	case .relaxed:
-		return intrinsics.atomic_load_relaxed(object)
-	case .consume:
-		return intrinsics.atomic_load_acq(object)
-	case .acquire:
-		return intrinsics.atomic_load_acq(object)
-	case .seq_cst:
-		return intrinsics.atomic_load(object)
+	#partial switch order {
+	case .relaxed: return intrinsics.atomic_load_explicit(object, .Relaxed)
+	case .consume: return intrinsics.atomic_load_explicit(object, .Consume)
+	case .acquire: return intrinsics.atomic_load_explicit(object, .Acquire)
+	case .seq_cst: return intrinsics.atomic_load_explicit(object, .Seq_Cst)
 	}
 }
 
 atomic_exchange :: #force_inline proc(object: ^$T, desired: T) -> T {
-	return intrinsics.atomic_xchg(object, desired)
+	return intrinsics.atomic_exchange(object, desired)
 }
 
 atomic_exchange_explicit :: #force_inline proc(object: ^$T, desired: T, order: memory_order) -> T {
-	switch (order) {
-	case .relaxed:
-		return intrinsics.atomic_xchg_relaxed(object, desired)
-	case .consume:
-		return intrinsics.atomic_xchg_acq(object, desired)
-	case .acquire:
-		return intrinsics.atomic_xchg_acq(object, desired)
-	case .release:
-		return intrinsics.atomic_xchg_rel(object, desired)
-	case .acq_rel:
-		return intrinsics.atomic_xchg_acqrel(object, desired)
-	case .seq_cst:
-		return intrinsics.atomic_xchg(object, desired)
+	switch order {
+	case .relaxed: return intrinsics.atomic_exchange_explicit(object, desired, .Relaxed)
+	case .consume: return intrinsics.atomic_exchange_explicit(object, desired, .Consume)
+	case .acquire: return intrinsics.atomic_exchange_explicit(object, desired, .Acquire)
+	case .release: return intrinsics.atomic_exchange_explicit(object, desired, .Release)
+	case .acq_rel: return intrinsics.atomic_exchange_explicit(object, desired, .Acq_Rel)
+	case .seq_cst: return intrinsics.atomic_exchange_explicit(object, desired, .Seq_Cst)
 	}
 	return false
 }
@@ -189,102 +177,104 @@ atomic_exchange_explicit :: #force_inline proc(object: ^$T, desired: T, order: m
 // 	[success = seq_cst, failure = acquire] => failacq
 // 	[success = acquire, failure = relaxed] => acq_failrelaxed
 // 	[success = acq_rel, failure = relaxed] => acqrel_failrelaxed
-atomic_compare_exchange_strong :: #force_inline proc(object, expected: ^$T, desired: T) {
-	value, ok := intrinsics.atomic_cxchg(object, expected^, desired)
+atomic_compare_exchange_strong :: #force_inline proc(object, expected: ^$T, desired: T) -> bool {
+	value, ok := intrinsics.atomic_compare_exchange_strong(object, expected^, desired)
 	if !ok { expected^ = value } 
 	return ok
 }
 
-atomic_compare_exchange_strong_explicit :: #force_inline proc(object, expected: ^$T, desired: T, success, failure: memory_order) {
+atomic_compare_exchange_strong_explicit :: #force_inline proc(object, expected: ^$T, desired: T, success, failure: memory_order) -> bool {
 	assert(failure != .release)
 	assert(failure != .acq_rel)
 
 	value: T; ok: bool
-	#partial switch (failure) {
+	#partial switch failure {
 	case .seq_cst:
 		assert(success != .relaxed)
-		#partial switch (success) {
+		#partial switch success {
 		case .seq_cst:
-			value, ok := intrinsics.atomic_cxchg(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Seq_Cst, .Seq_Cst)
 		case .acquire:
-			value, ok := intrinsics.atomic_cxchg_acq(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Acquire, .Seq_Cst)
 		case .consume:
-			value, ok := intrinsics.atomic_cxchg_acq(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Consume, .Seq_Cst)
 		case .release:
-			value, ok := intrinsics.atomic_cxchg_rel(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Release, .Seq_Cst)
 		case .acq_rel:
-			value, ok := intrinsics.atomic_cxchg_acqrel(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Acq_Rel, .Seq_Cst)
 		}
 	case .relaxed:
 		assert(success != .release)
-		#partial switch (success) {
+		#partial switch success {
 		case .relaxed:
-			value, ok := intrinsics.atomic_cxchg_relaxed(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Relaxed, .Relaxed)
 		case .seq_cst:
-			value, ok := intrinsics.atomic_cxchg_failrelaxed(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Seq_Cst, .Relaxed)
 		case .acquire:
-			value, ok := intrinsics.atomic_cxchg_acq_failrelaxed(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Acquire, .Relaxed)
 		case .consume:
-			value, ok := intrinsics.atomic_cxchg_acq_failrelaxed(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Consume, .Relaxed)
 		case .acq_rel:
-			value, ok := intrinsics.atomic_cxchg_acqrel_failrelaxed(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Acq_Rel, .Relaxed)
 		}
 	case .consume:
-		fallthrough
+		assert(success == .seq_cst)
+		value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Seq_Cst, .Consume)
 	case .acquire:
 		assert(success == .seq_cst)
-		value, ok := intrinsics.atomic_cxchg_failacq(object, expected^, desired)
+		value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Seq_Cst, .Acquire)
 
 	}
 	if !ok { expected^ = value }
 	return ok
 }
 
-atomic_compare_exchange_weak :: #force_inline proc(object, expected: ^$T, desired: T) {
-	value, ok := intrinsics.atomic_cxchgweak(object, expected^, desired)
+atomic_compare_exchange_weak :: #force_inline proc(object, expected: ^$T, desired: T) -> bool {
+	value, ok := intrinsics.atomic_compare_exchange_weak(object, expected^, desired)
 	if !ok { expected^ = value }
 	return ok
 }
 
-atomic_compare_exchange_weak_explicit :: #force_inline proc(object, expected: ^$T, desited: T, success, failure: memory_order) {
+atomic_compare_exchange_weak_explicit :: #force_inline proc(object, expected: ^$T, desited: T, success, failure: memory_order) -> bool {
 	assert(failure != .release)
 	assert(failure != .acq_rel)
 
 	value: T; ok: bool
-	#partial switch (failure) {
+	#partial switch failure {
 	case .seq_cst:
 		assert(success != .relaxed)
-		#partial switch (success) {
+		#partial switch success {
 		case .seq_cst:
-			value, ok := intrinsics.atomic_cxchgweak(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Seq_Cst, .Seq_Cst)
 		case .acquire:
-			value, ok := intrinsics.atomic_cxchgweak_acq(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Acquire, .Seq_Cst)
 		case .consume:
-			value, ok := intrinsics.atomic_cxchgweak_acq(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Consume, .Seq_Cst)
 		case .release:
-			value, ok := intrinsics.atomic_cxchgweak_rel(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Release, .Seq_Cst)
 		case .acq_rel:
-			value, ok := intrinsics.atomic_cxchgweak_acqrel(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Acq_Rel, .Seq_Cst)
 		}
 	case .relaxed:
 		assert(success != .release)
-		#partial switch (success) {
+		#partial switch success {
 		case .relaxed:
-			value, ok := intrinsics.atomic_cxchgweak_relaxed(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Relaxed, .Relaxed)
 		case .seq_cst:
-			value, ok := intrinsics.atomic_cxchgweak_failrelaxed(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Seq_Cst, .Relaxed)
 		case .acquire:
-			value, ok := intrinsics.atomic_cxchgweak_acq_failrelaxed(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Acquire, .Relaxed)
 		case .consume:
-			value, ok := intrinsics.atomic_cxchgweak_acq_failrelaxed(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Consume, .Relaxed)
 		case .acq_rel:
-			value, ok := intrinsics.atomic_cxchgweak_acqrel_failrelaxed(object, expected^, desired)
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Acq_Rel, .Relaxed)
 		}
 	case .consume:
-		fallthrough
+		assert(success == .seq_cst)
+		value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Seq_Cst, .Consume)
 	case .acquire:
 		assert(success == .seq_cst)
-		value, ok := intrinsics.atomic_cxchgweak_failacq(object, expected^, desired)
+		value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Seq_Cst, .Acquire)
 
 	}
 	if !ok { expected^ = value }
@@ -297,19 +287,14 @@ atomic_fetch_add :: #force_inline proc(object: ^$T, operand: T) -> T {
 }
 
 atomic_fetch_add_explicit :: #force_inline proc(object: ^$T, operand: T, order: memory_order) -> T {
-	switch (order) {
-	case .relaxed:
-		return intrinsics.atomic_add_relaxed(object, operand)
-	case .consume:
-		return intrinsics.atomic_add_acq(object, operand)
-	case .acquire:
-		return intrinsics.atomic_add_acq(object, operand)
-	case .release:
-		return intrinsics.atomic_add_rel(object, operand)
-	case .acq_rel:
-		return intrinsics.atomic_add_acqrel(object, operand)
-	case .seq_cst:
-		return intrinsics.atomic_add(object, operand)
+	switch order {
+	case .relaxed: return intrinsics.atomic_add_explicit(object, operand, .Relaxed)
+	case .consume: return intrinsics.atomic_add_explicit(object, operand, .Consume)
+	case .acquire: return intrinsics.atomic_add_explicit(object, operand, .Acquire)
+	case .release: return intrinsics.atomic_add_explicit(object, operand, .Release)
+	case .acq_rel: return intrinsics.atomic_add_explicit(object, operand, .Acq_Rel)
+	case: fallthrough
+	case .seq_cst: return intrinsics.atomic_add_explicit(object, operand, .Seq_Cst)
 	}
 }
 
@@ -318,19 +303,14 @@ atomic_fetch_sub :: #force_inline proc(object: ^$T, operand: T) -> T {
 }
 
 atomic_fetch_sub_explicit :: #force_inline proc(object: ^$T, operand: T, order: memory_order) -> T {
-	switch (order) {
-	case .relaxed:
-		return intrinsics.atomic_sub_relaxed(object, operand)
-	case .consume:
-		return intrinsics.atomic_sub_acq(object, operand)
-	case .acquire:
-		return intrinsics.atomic_sub_acq(object, operand)
-	case .release:
-		return intrinsics.atomic_sub_rel(object, operand)
-	case .acq_rel:
-		return intrinsics.atomic_sub_acqrel(object, operand)
-	case .seq_cst:
-		return intrinsics.atomic_sub(object, operand)
+	switch order {
+	case .relaxed: return intrinsics.atomic_sub_explicit(object, operand, .Relaxed)
+	case .consume: return intrinsics.atomic_sub_explicit(object, operand, .Consume)
+	case .acquire: return intrinsics.atomic_sub_explicit(object, operand, .Acquire)
+	case .release: return intrinsics.atomic_sub_explicit(object, operand, .Release)
+	case .acq_rel: return intrinsics.atomic_sub_explicit(object, operand, .Acq_Rel)
+	case: fallthrough
+	case .seq_cst: return intrinsics.atomic_sub_explicit(object, operand, .Seq_Cst)
 	}
 }
 
@@ -339,19 +319,14 @@ atomic_fetch_or :: #force_inline proc(object: ^$T, operand: T) -> T {
 }
 
 atomic_fetch_or_explicit :: #force_inline proc(object: ^$T, operand: T, order: memory_order) -> T {
-	switch (order) {
-	case .relaxed:
-		return intrinsics.atomic_or_relaxed(object, operand)
-	case .consume:
-		return intrinsics.atomic_or_acq(object, operand)
-	case .acquire:
-		return intrinsics.atomic_or_acq(object, operand)
-	case .release:
-		return intrinsics.atomic_or_rel(object, operand)
-	case .acq_rel:
-		return intrinsics.atomic_or_acqrel(object, operand)
-	case .seq_cst:
-		return intrinsics.atomic_or(object, operand)
+	switch order {
+	case .relaxed: return intrinsics.atomic_or_explicit(object, operand, .Relaxed)
+	case .consume: return intrinsics.atomic_or_explicit(object, operand, .Consume)
+	case .acquire: return intrinsics.atomic_or_explicit(object, operand, .Acquire)
+	case .release: return intrinsics.atomic_or_explicit(object, operand, .Release)
+	case .acq_rel: return intrinsics.atomic_or_explicit(object, operand, .Acq_Rel)
+	case: fallthrough
+	case .seq_cst: return intrinsics.atomic_or_explicit(object, operand, .Seq_Cst)
 	}
 }
 
@@ -360,19 +335,14 @@ atomic_fetch_xor :: #force_inline proc(object: ^$T, operand: T) -> T {
 }
 
 atomic_fetch_xor_explicit :: #force_inline proc(object: ^$T, operand: T, order: memory_order) -> T {
-	switch (order) {
-	case .relaxed:
-		return intrinsics.atomic_xor_relaxed(object, operand)
-	case .consume:
-		return intrinsics.atomic_xor_acq(object, operand)
-	case .acquire:
-		return intrinsics.atomic_xor_acq(object, operand)
-	case .release:
-		return intrinsics.atomic_xor_rel(object, operand)
-	case .acq_rel:
-		return intrinsics.atomic_xor_acqrel(object, operand)
-	case .seq_cst:
-		return intrinsics.atomic_xor(object, operand)
+	switch order {
+	case .relaxed: return intrinsics.atomic_xor_explicit(object, operand, .Relaxed)
+	case .consume: return intrinsics.atomic_xor_explicit(object, operand, .Consume)
+	case .acquire: return intrinsics.atomic_xor_explicit(object, operand, .Acquire)
+	case .release: return intrinsics.atomic_xor_explicit(object, operand, .Release)
+	case .acq_rel: return intrinsics.atomic_xor_explicit(object, operand, .Acq_Rel)
+	case: fallthrough
+	case .seq_cst: return intrinsics.atomic_xor_explicit(object, operand, .Seq_Cst)
 	}
 }
 
@@ -380,19 +350,14 @@ atomic_fetch_and :: #force_inline proc(object: ^$T, operand: T) -> T {
 	return intrinsics.atomic_and(object, operand)
 }
 atomic_fetch_and_explicit :: #force_inline proc(object: ^$T, operand: T, order: memory_order) -> T {
-	switch (order) {
-	case .relaxed:
-		return intrinsics.atomic_and_relaxed(object, operand)
-	case .consume:
-		return intrinsics.atomic_and_acq(object, operand)
-	case .acquire:
-		return intrinsics.atomic_and_acq(object, operand)
-	case .release:
-		return intrinsics.atomic_and_rel(object, operand)
-	case .acq_rel:
-		return intrinsics.atomic_and_acqrel(object, operand)
-	case .seq_cst:
-		return intrinsics.atomic_and(object, operand)
+	switch order {
+	case .relaxed: return intrinsics.atomic_and_explicit(object, operand, .Relaxed)
+	case .consume: return intrinsics.atomic_and_explicit(object, operand, .Consume)
+	case .acquire: return intrinsics.atomic_and_explicit(object, operand, .Acquire)
+	case .release: return intrinsics.atomic_and_explicit(object, operand, .Release)
+	case .acq_rel: return intrinsics.atomic_and_explicit(object, operand, .Acq_Rel)
+	case: fallthrough
+	case .seq_cst: return intrinsics.atomic_and_explicit(object, operand, .Seq_Cst)
 	}
 }
 
diff --git a/core/intrinsics/intrinsics.odin b/core/intrinsics/intrinsics.odin
index 16dd5cbc5..ce8fd96ea 100644
--- a/core/intrinsics/intrinsics.odin
+++ b/core/intrinsics/intrinsics.odin
@@ -62,77 +62,44 @@ syscall :: proc(id: uintptr, args: ..uintptr) -> uintptr ---
 
 
 // Atomics
-atomic_fence        :: proc() ---
-atomic_fence_acq    :: proc() ---
-atomic_fence_rel    :: proc() ---
-atomic_fence_acqrel :: proc() ---
+Atomic_Memory_Order :: enum {
+	Relaxed = 0, // Unordered
+	Consume = 1, // Monotonic
+	Acquire = 2,
+	Release = 3,
+	Acq_Rel = 4,
+	Seq_Cst = 5,
+}
 
-atomic_store           :: proc(dst: ^$T, val: T) ---
-atomic_store_rel       :: proc(dst: ^$T, val: T) ---
-atomic_store_relaxed   :: proc(dst: ^$T, val: T) ---
-atomic_store_unordered :: proc(dst: ^$T, val: T) ---
+atomic_thread_fence :: proc(order: Atomic_Memory_Order) ---
+atomic_signal_fence :: proc(order: Atomic_Memory_Order) ---
+
+atomic_store          :: proc(dst: ^$T, val: T) ---
+atomic_store_explicit :: proc(dst: ^$T, val: T, order: Atomic_Memory_Order) ---
 
 atomic_load           :: proc(dst: ^$T) -> T ---
-atomic_load_acq       :: proc(dst: ^$T) -> T ---
-atomic_load_relaxed   :: proc(dst: ^$T) -> T ---
-atomic_load_unordered :: proc(dst: ^$T) -> T ---
+atomic_load_explicit  :: proc(dst: ^$T, order: Atomic_Memory_Order) -> T ---
 
-atomic_add          :: proc(dst; ^$T, val: T) -> T ---
-atomic_add_acq      :: proc(dst; ^$T, val: T) -> T ---
-atomic_add_rel      :: proc(dst; ^$T, val: T) -> T ---
-atomic_add_acqrel   :: proc(dst; ^$T, val: T) -> T ---
-atomic_add_relaxed  :: proc(dst; ^$T, val: T) -> T ---
-atomic_sub          :: proc(dst; ^$T, val: T) -> T ---
-atomic_sub_acq      :: proc(dst; ^$T, val: T) -> T ---
-atomic_sub_rel      :: proc(dst; ^$T, val: T) -> T ---
-atomic_sub_acqrel   :: proc(dst; ^$T, val: T) -> T ---
-atomic_sub_relaxed  :: proc(dst; ^$T, val: T) -> T ---
-atomic_and          :: proc(dst; ^$T, val: T) -> T ---
-atomic_and_acq      :: proc(dst; ^$T, val: T) -> T ---
-atomic_and_rel      :: proc(dst; ^$T, val: T) -> T ---
-atomic_and_acqrel   :: proc(dst; ^$T, val: T) -> T ---
-atomic_and_relaxed  :: proc(dst; ^$T, val: T) -> T ---
-atomic_nand         :: proc(dst; ^$T, val: T) -> T ---
-atomic_nand_acq     :: proc(dst; ^$T, val: T) -> T ---
-atomic_nand_rel     :: proc(dst; ^$T, val: T) -> T ---
-atomic_nand_acqrel  :: proc(dst; ^$T, val: T) -> T ---
-atomic_nand_relaxed :: proc(dst; ^$T, val: T) -> T ---
-atomic_or           :: proc(dst; ^$T, val: T) -> T ---
-atomic_or_acq       :: proc(dst; ^$T, val: T) -> T ---
-atomic_or_rel       :: proc(dst; ^$T, val: T) -> T ---
-atomic_or_acqrel    :: proc(dst; ^$T, val: T) -> T ---
-atomic_or_relaxed   :: proc(dst; ^$T, val: T) -> T ---
-atomic_xor          :: proc(dst; ^$T, val: T) -> T ---
-atomic_xor_acq      :: proc(dst; ^$T, val: T) -> T ---
-atomic_xor_rel      :: proc(dst; ^$T, val: T) -> T ---
-atomic_xor_acqrel   :: proc(dst; ^$T, val: T) -> T ---
-atomic_xor_relaxed  :: proc(dst; ^$T, val: T) -> T ---
+atomic_add               :: proc(dst; ^$T, val: T) -> T ---
+atomic_add_explicit      :: proc(dst; ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_sub               :: proc(dst; ^$T, val: T) -> T ---
+atomic_sub_explicit      :: proc(dst; ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_and               :: proc(dst; ^$T, val: T) -> T ---
+atomic_and_explicit      :: proc(dst; ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_nand              :: proc(dst; ^$T, val: T) -> T ---
+atomic_nand_explicit     :: proc(dst; ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_or                :: proc(dst; ^$T, val: T) -> T ---
+atomic_or_explicit       :: proc(dst; ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_xor               :: proc(dst; ^$T, val: T) -> T ---
+atomic_xor_explicit      :: proc(dst; ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_exchange          :: proc(dst; ^$T, val: T) -> T ---
+atomic_exchange_explicit :: proc(dst; ^$T, val: T, order: Atomic_Memory_Order) -> T ---
 
-atomic_xchg         :: proc(dst; ^$T, val: T) -> T ---
-atomic_xchg_acq     :: proc(dst; ^$T, val: T) -> T ---
-atomic_xchg_rel     :: proc(dst; ^$T, val: T) -> T ---
-atomic_xchg_acqrel  :: proc(dst; ^$T, val: T) -> T ---
-atomic_xchg_relaxed :: proc(dst; ^$T, val: T) -> T ---
+atomic_compare_exchange_strong          :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
+atomic_compare_exchange_strong_explicit :: proc(dst: ^$T, old, new: T, success, failure: Atomic_Memory_Order) -> (T, bool) #optional_ok ---
+atomic_compare_exchange_weak            :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
+atomic_compare_exchange_weak_explicit   :: proc(dst: ^$T, old, new: T, success, failure: Atomic_Memory_Order) -> (T, bool) #optional_ok ---
 
-atomic_cxchg                    :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchg_acq                :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchg_rel                :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchg_acqrel             :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchg_relaxed            :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchg_failrelaxed        :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchg_failacq            :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchg_acq_failrelaxed    :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchg_acqrel_failrelaxed :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-
-atomic_cxchgweak                    :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchgweak_acq                :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchgweak_rel                :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchgweak_acqrel             :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchgweak_relaxed            :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchgweak_failrelaxed        :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchgweak_failacq            :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchgweak_acq_failrelaxed    :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
-atomic_cxchgweak_acqrel_failrelaxed :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
 
 // Constant type tests
 
diff --git a/core/mem/mem.odin b/core/mem/mem.odin
index 8eb877e75..817c0ea5f 100644
--- a/core/mem/mem.odin
+++ b/core/mem/mem.odin
@@ -16,7 +16,7 @@ zero_explicit :: proc "contextless" (data: rawptr, len: int) -> rawptr {
 	// equivalent semantics to those provided by the C11 Annex K 3.7.4.1
 	// memset_s call.
 	intrinsics.mem_zero_volatile(data, len) // Use the volatile mem_zero
-	intrinsics.atomic_fence() // Prevent reordering
+	intrinsics.atomic_thread_fence(.Seq_Cst) // Prevent reordering
 	return data
 }
 zero_item :: proc "contextless" (item: $P/^$T) {
diff --git a/core/mem/virtual/virtual_platform.odin b/core/mem/virtual/virtual_platform.odin
index d707bc427..367346f63 100644
--- a/core/mem/virtual/virtual_platform.odin
+++ b/core/mem/virtual/virtual_platform.odin
@@ -1,7 +1,7 @@
 //+private
 package mem_virtual
 
-import sync "core:sync/sync2"
+import "core:sync"
 
 Platform_Memory_Block :: struct {
 	block:      Memory_Block,
diff --git a/core/os/os2/process.odin b/core/os/os2/process.odin
index 028951fe3..7fc7a4ac0 100644
--- a/core/os/os2/process.odin
+++ b/core/os/os2/process.odin
@@ -1,6 +1,6 @@
 package os2
 
-import sync "core:sync/sync2"
+import "core:sync"
 import "core:time"
 import "core:runtime"
 
diff --git a/core/sync/atomic.odin b/core/sync/atomic.odin
index 21dcea178..f537764c4 100644
--- a/core/sync/atomic.odin
+++ b/core/sync/atomic.odin
@@ -2,167 +2,44 @@ package sync
 
 import "core:intrinsics"
 
-Ordering :: enum {
-	Relaxed, // Monotonic
-	Release,
-	Acquire,
-	Acquire_Release,
-	Sequentially_Consistent,
-}
-
-strongest_failure_ordering_table := [Ordering]Ordering{
-	.Relaxed                 = .Relaxed,
-	.Release                 = .Relaxed,
-	.Acquire                 = .Acquire,
-	.Acquire_Release         = .Acquire,
-	.Sequentially_Consistent = .Sequentially_Consistent,
-}
-
-strongest_failure_ordering :: #force_inline proc(order: Ordering) -> Ordering {
-	return strongest_failure_ordering_table[order]
-}
-
-fence :: #force_inline proc($order: Ordering) {
-	     when order == .Relaxed                 { #panic("there is no such thing as a relaxed fence") }
-	else when order == .Release                 { intrinsics.atomic_fence_rel()                       }
-	else when order == .Acquire                 { intrinsics.atomic_fence_acq()                       }
-	else when order == .Acquire_Release         { intrinsics.atomic_fence_acqrel()                    }
-	else when order == .Sequentially_Consistent { intrinsics.atomic_fence()                           }
-	else { #panic("unknown order") }
+cpu_relax :: intrinsics.cpu_relax
+
+/*
+Atomic_Memory_Order :: enum {
+	Relaxed = 0,
+	Consume = 1,
+	Acquire = 2,
+	Release = 3,
+	Acq_Rel = 4,
+	Seq_Cst = 5,
 }
+*/
+Atomic_Memory_Order :: intrinsics.Atomic_Memory_Order
 
 
-atomic_store :: #force_inline proc(dst: ^$T, val: T, $order: Ordering) {
-	     when order == .Relaxed                 { intrinsics.atomic_store_relaxed(dst, val) }
-	else when order == .Release                 { intrinsics.atomic_store_rel(dst, val) }
-	else when order == .Sequentially_Consistent { intrinsics.atomic_store(dst, val) }
-	else when order == .Acquire                 { #panic("there is not such thing as an acquire store") }
-	else when order == .Acquire_Release         { #panic("there is not such thing as an acquire/release store") }
-	else { #panic("unknown order") }
-}
-
-atomic_load :: #force_inline proc(dst: ^$T, $order: Ordering) -> T {
-	     when order == .Relaxed                 { return intrinsics.atomic_load_relaxed(dst) }
-	else when order == .Acquire                 { return intrinsics.atomic_load_acq(dst) }
-	else when order == .Sequentially_Consistent { return intrinsics.atomic_load(dst) }
-	else when order == .Release                 { #panic("there is no such thing as a release load") }
-	else when order == .Acquire_Release         { #panic("there is no such thing as an acquire/release load") }
-	else { #panic("unknown order") }
-}
-
-atomic_swap :: #force_inline proc(dst: ^$T, val: T, $order: Ordering) -> T {
-	     when order == .Relaxed                 { return intrinsics.atomic_xchg_relaxed(dst, val) }
-	else when order == .Release                 { return intrinsics.atomic_xchg_rel(dst, val)     }
-	else when order == .Acquire                 { return intrinsics.atomic_xchg_acq(dst, val)     }
-	else when order == .Acquire_Release         { return intrinsics.atomic_xchg_acqrel(dst, val)  }
-	else when order == .Sequentially_Consistent { return intrinsics.atomic_xchg(dst, val)         }
-	else { #panic("unknown order") }
-}
-
-atomic_compare_exchange :: #force_inline proc(dst: ^$T, old, new: T, $success, $failure: Ordering) -> (val: T, ok: bool) {
-	when failure == .Relaxed {
-		     when success == .Relaxed                 { return intrinsics.atomic_cxchg_relaxed(dst, old, new) }
-		else when success == .Acquire                 { return intrinsics.atomic_cxchg_acq_failrelaxed(dst, old, new) }
-		else when success == .Acquire_Release         { return intrinsics.atomic_cxchg_acqrel_failrelaxed(dst, old, new) }
-		else when success == .Sequentially_Consistent { return intrinsics.atomic_cxchg_failrelaxed(dst, old, new) }
-		else when success == .Release                 { return intrinsics.atomic_cxchg_rel(dst, old, new) }
-		else { #panic("an unknown ordering combination") }
-	} else when failure == .Acquire {
-		     when success == .Release { return intrinsics.atomic_cxchg_acqrel(dst, old, new) }
-		else when success == .Acquire { return intrinsics.atomic_cxchg_acq(dst, old, new) }
-		else { #panic("an unknown ordering combination") }
-	} else when failure == .Sequentially_Consistent {
-		when success == .Sequentially_Consistent { return intrinsics.atomic_cxchg(dst, old, new) }
-		else { #panic("an unknown ordering combination") }
-	} else when failure == .Acquire_Release {
-		#panic("there is not such thing as an acquire/release failure ordering")
-	} else when failure == .Release {
-		when success == .Acquire { return instrinsics.atomic_cxchg_failacq(dst, old, new) }
-		else { #panic("an unknown ordering combination") }
-	} else {
-		return T{}, false
-	}
-
-}
-
-atomic_compare_exchange_weak :: #force_inline proc(dst: ^$T, old, new: T, $success, $failure: Ordering) -> (val: T, ok: bool) {
-	when failure == .Relaxed {
-		     when success == .Relaxed                 { return intrinsics.atomic_cxchgweak_relaxed(dst, old, new) }
-		else when success == .Acquire                 { return intrinsics.atomic_cxchgweak_acq_failrelaxed(dst, old, new) }
-		else when success == .Acquire_Release         { return intrinsics.atomic_cxchgweak_acqrel_failrelaxed(dst, old, new) }
-		else when success == .Sequentially_Consistent { return intrinsics.atomic_cxchgweak_failrelaxed(dst, old, new) }
-		else when success == .Release                 { return intrinsics.atomic_cxchgweak_rel(dst, old, new) }
-		else { #panic("an unknown ordering combination") }
-	} else when failure == .Acquire {
-		     when success == .Release { return intrinsics.atomic_cxchgweak_acqrel(dst, old, new) }
-		else when success == .Acquire { return intrinsics.atomic_cxchgweak_acq(dst, old, new) }
-		else { #panic("an unknown ordering combination") }
-	} else when failure == .Sequentially_Consistent {
-		when success == .Sequentially_Consistent { return intrinsics.atomic_cxchgweak(dst, old, new) }
-		else { #panic("an unknown ordering combination") }
-	} else when failure == .Acquire_Release {
-		#panic("there is not such thing as an acquire/release failure ordering")
-	} else when failure == .Release {
-		when success == .Acquire { return intrinsics.atomic_cxchgweak_failacq(dst, old, new) }
-		else { #panic("an unknown ordering combination") }
-	} else {
-		return T{}, false
-	}
-
-}
-
-
-atomic_add :: #force_inline proc(dst: ^$T, val: T, $order: Ordering) -> T {
-	     when order == .Relaxed                 { return intrinsics.atomic_add_relaxed(dst, val) }
-	else when order == .Release                 { return intrinsics.atomic_add_rel(dst, val) }
-	else when order == .Acquire                 { return intrinsics.atomic_add_acq(dst, val) }
-	else when order == .Acquire_Release         { return intrinsics.atomic_add_acqrel(dst, val) }
-	else when order == .Sequentially_Consistent { return intrinsics.atomic_add(dst, val) }
-	else { #panic("unknown order") }
-}
-
-atomic_sub :: #force_inline proc(dst: ^$T, val: T, $order: Ordering) -> T {
-	     when order == .Relaxed                 { return intrinsics.atomic_sub_relaxed(dst, val) }
-	else when order == .Release                 { return intrinsics.atomic_sub_rel(dst, val) }
-	else when order == .Acquire                 { return intrinsics.atomic_sub_acq(dst, val) }
-	else when order == .Acquire_Release         { return intrinsics.atomic_sub_acqrel(dst, val) }
-	else when order == .Sequentially_Consistent { return intrinsics.atomic_sub(dst, val) }
-	else { #panic("unknown order") }
-}
-
-atomic_and :: #force_inline proc(dst: ^$T, val: T, $order: Ordering) -> T {
-	     when order == .Relaxed                 { return intrinsics.atomic_and_relaxed(dst, val) }
-	else when order == .Release                 { return intrinsics.atomic_and_rel(dst, val) }
-	else when order == .Acquire                 { return intrinsics.atomic_and_acq(dst, val) }
-	else when order == .Acquire_Release         { return intrinsics.atomic_and_acqrel(dst, val) }
-	else when order == .Sequentially_Consistent { return intrinsics.atomic_and(dst, val) }
-	else { #panic("unknown order") }
-}
-
-atomic_nand :: #force_inline proc(dst: ^$T, val: T, $order: Ordering) -> T {
-	     when order == .Relaxed                 { return intrinsics.atomic_nand_relaxed(dst, val) }
-	else when order == .Release                 { return intrinsics.atomic_nand_rel(dst, val) }
-	else when order == .Acquire                 { return intrinsics.atomic_nand_acq(dst, val) }
-	else when order == .Acquire_Release         { return intrinsics.atomic_nand_acqrel(dst, val) }
-	else when order == .Sequentially_Consistent { return intrinsics.atomic_nand(dst, val) }
-	else { #panic("unknown order") }
-}
-
-atomic_or :: #force_inline proc(dst: ^$T, val: T, $order: Ordering) -> T {
-	     when order == .Relaxed                 { return intrinsics.atomic_or_relaxed(dst, val) }
-	else when order == .Release                 { return intrinsics.atomic_or_rel(dst, val) }
-	else when order == .Acquire                 { return intrinsics.atomic_or_acq(dst, val) }
-	else when order == .Acquire_Release         { return intrinsics.atomic_or_acqrel(dst, val) }
-	else when order == .Sequentially_Consistent { return intrinsics.atomic_or(dst, val) }
-	else { #panic("unknown order") }
-}
-
-atomic_xor :: #force_inline proc(dst: ^$T, val: T, $order: Ordering) -> T {
-	     when order == .Relaxed                 { return intrinsics.atomic_xor_relaxed(dst, val) }
-	else when order == .Release                 { return intrinsics.atomic_xor_rel(dst, val) }
-	else when order == .Acquire                 { return intrinsics.atomic_xor_acq(dst, val) }
-	else when order == .Acquire_Release         { return intrinsics.atomic_xor_acqrel(dst, val) }
-	else when order == .Sequentially_Consistent { return intrinsics.atomic_xor(dst, val) }
-	else { #panic("unknown order") }
-}
+atomic_thread_fence                     :: intrinsics.atomic_thread_fence
+atomic_signal_fence                     :: intrinsics.atomic_signal_fence
+atomic_store                            :: intrinsics.atomic_store
+atomic_store_explicit                   :: intrinsics.atomic_store_explicit
+atomic_load                             :: intrinsics.atomic_load
+atomic_load_explicit                    :: intrinsics.atomic_load_explicit
+atomic_add                              :: intrinsics.atomic_add
+atomic_add_explicit                     :: intrinsics.atomic_add_explicit
+atomic_sub                              :: intrinsics.atomic_sub
+atomic_sub_explicit                     :: intrinsics.atomic_sub_explicit
+atomic_and                              :: intrinsics.atomic_and
+atomic_and_explicit                     :: intrinsics.atomic_and_explicit
+atomic_nand                             :: intrinsics.atomic_nand
+atomic_nand_explicit                    :: intrinsics.atomic_nand_explicit
+atomic_or                               :: intrinsics.atomic_or
+atomic_or_explicit                      :: intrinsics.atomic_or_explicit
+atomic_xor                              :: intrinsics.atomic_xor
+atomic_xor_explicit                     :: intrinsics.atomic_xor_explicit
+atomic_exchange                         :: intrinsics.atomic_exchange
+atomic_exchange_explicit                :: intrinsics.atomic_exchange_explicit
 
+// Returns value and optional ok boolean
+atomic_compare_exchange_strong          :: intrinsics.atomic_compare_exchange_strong
+atomic_compare_exchange_strong_explicit :: intrinsics.atomic_compare_exchange_strong_explicit
+atomic_compare_exchange_weak            :: intrinsics.atomic_compare_exchange_weak
+atomic_compare_exchange_weak_explicit   :: intrinsics.atomic_compare_exchange_weak_explicit
\ No newline at end of file
diff --git a/core/sync/barrier.odin b/core/sync/barrier.odin
deleted file mode 100644
index 13c870a3b..000000000
--- a/core/sync/barrier.odin
+++ /dev/null
@@ -1,80 +0,0 @@
-package sync
-
-
-/*
-A barrier enabling multiple threads to synchronize the beginning of some computation
-Example:
-
-	package example
-
-	import "core:fmt"
-	import "core:sync"
-	import "core:thread"
-
-	barrier := &sync.Barrier{};
-
-	main :: proc() {
-		fmt.println("Start");
-
-		THREAD_COUNT :: 4;
-		threads: [THREAD_COUNT]^thread.Thread;
-
-		sync.barrier_init(barrier, THREAD_COUNT);
-		defer sync.barrier_destroy(barrier);
-
-
-		for _, i in threads {
-			threads[i] = thread.create_and_start(proc(t: ^thread.Thread) {
-				// Same messages will be printed together but without any interleaving
-				fmt.println("Getting ready!");
-				sync.barrier_wait(barrier);
-				fmt.println("Off their marks they go!");
-			});
-		}
-
-		for t in threads {
-			thread.destroy(t); // join and free thread
-		}
-		fmt.println("Finished");
-	}
-*/
-Barrier :: struct {
-	mutex: Blocking_Mutex,
-	cond:  Condition,
-	index:         int,
-	generation_id: int,
-	thread_count:  int,
-}
-
-barrier_init :: proc(b: ^Barrier, thread_count: int) {
-	blocking_mutex_init(&b.mutex)
-	condition_init(&b.cond, &b.mutex)
-	b.index = 0
-	b.generation_id = 0
-	b.thread_count = thread_count
-}
-
-barrier_destroy :: proc(b: ^Barrier) {
-	blocking_mutex_destroy(&b.mutex)
-	condition_destroy(&b.cond)
-}
-
-// Block the current thread until all threads have rendezvoused
-// Barrier can be reused after all threads rendezvoused once, and can be used continuously
-barrier_wait :: proc(b: ^Barrier) -> (is_leader: bool) {
-	blocking_mutex_lock(&b.mutex)
-	defer blocking_mutex_unlock(&b.mutex)
-	local_gen := b.generation_id
-	b.index += 1
-	if b.index < b.thread_count {
-		for local_gen == b.generation_id && b.index < b.thread_count {
-			condition_wait_for(&b.cond)
-		}
-		return false
-	}
-
-	b.index = 0
-	b.generation_id += 1
-	condition_broadcast(&b.cond)
-	return true
-}
diff --git a/core/sync/channel.odin b/core/sync/channel.odin
deleted file mode 100644
index 82b9504f4..000000000
--- a/core/sync/channel.odin
+++ /dev/null
@@ -1,889 +0,0 @@
-package sync
-
-import "core:mem"
-import "core:time"
-import "core:intrinsics"
-import "core:math/rand"
-
-_, _ :: time, rand
-
-Channel_Direction :: enum i8 {
-	Both =  0,
-	Send = +1,
-	Recv = -1,
-}
-
-Channel :: struct($T: typeid, $Direction := Channel_Direction.Both) {
-	using _internal: ^Raw_Channel,
-}
-
-channel_init :: proc(ch: ^$C/Channel($T, $D), cap := 0, allocator := context.allocator) {
-	context.allocator = allocator
-	ch._internal = raw_channel_create(size_of(T), align_of(T), cap)
-	return
-}
-
-channel_make :: proc($T: typeid, cap := 0, allocator := context.allocator) -> (ch: Channel(T, .Both)) {
-	context.allocator = allocator
-	ch._internal = raw_channel_create(size_of(T), align_of(T), cap)
-	return
-}
-
-channel_make_send :: proc($T: typeid, cap := 0, allocator := context.allocator) -> (ch: Channel(T, .Send)) {
-	context.allocator = allocator
-	ch._internal = raw_channel_create(size_of(T), align_of(T), cap)
-	return
-}
-channel_make_recv :: proc($T: typeid, cap := 0, allocator := context.allocator) -> (ch: Channel(T, .Recv)) {
-	context.allocator = allocator
-	ch._internal = raw_channel_create(size_of(T), align_of(T), cap)
-	return
-}
-
-channel_destroy :: proc(ch: $C/Channel($T, $D)) {
-	raw_channel_destroy(ch._internal)
-}
-
-channel_as_send :: proc(ch: $C/Channel($T, .Both)) -> (res: Channel(T, .Send)) {
-	res._internal = ch._internal
-	return
-}
-
-channel_as_recv :: proc(ch: $C/Channel($T, .Both)) -> (res: Channel(T, .Recv)) {
-	res._internal = ch._internal
-	return
-}
-
-
-channel_len :: proc(ch: $C/Channel($T, $D)) -> int {
-	return ch._internal.len if ch._internal != nil else 0
-}
-channel_cap :: proc(ch: $C/Channel($T, $D)) -> int {
-	return ch._internal.cap if ch._internal != nil else 0
-}
-
-
-channel_send :: proc(ch: $C/Channel($T, $D), msg: T, loc := #caller_location) where D >= .Both {
-	msg := msg
-	_ = raw_channel_send_impl(ch._internal, &msg, /*block*/true, loc)
-}
-channel_try_send :: proc(ch: $C/Channel($T, $D), msg: T, loc := #caller_location) -> bool where D >= .Both {
-	msg := msg
-	return raw_channel_send_impl(ch._internal, &msg, /*block*/false, loc)
-}
-
-channel_recv :: proc(ch: $C/Channel($T, $D), loc := #caller_location) -> (msg: T) where D <= .Both {
-	c := ch._internal
-	if c == nil {
-		panic(message="cannot recv message; channel is nil", loc=loc)
-	}
-	mutex_lock(&c.mutex)
-	raw_channel_recv_impl(c, &msg, loc)
-	mutex_unlock(&c.mutex)
-	return
-}
-channel_try_recv :: proc(ch: $C/Channel($T, $D), loc := #caller_location) -> (msg: T, ok: bool) where D <= .Both {
-	c := ch._internal
-	if c != nil && mutex_try_lock(&c.mutex) {
-		if c.len > 0 {
-			raw_channel_recv_impl(c, &msg, loc)
-			ok = true
-		}
-		mutex_unlock(&c.mutex)
-	}
-	return
-}
-channel_try_recv_ptr :: proc(ch: $C/Channel($T, $D), msg: ^T, loc := #caller_location) -> (ok: bool) where D <= .Both {
-	res: T
-	res, ok = channel_try_recv(ch, loc)
-	if ok && msg != nil {
-		msg^ = res
-	}
-	return
-}
-
-
-channel_is_nil :: proc(ch: $C/Channel($T, $D)) -> bool {
-	return ch._internal == nil
-}
-channel_is_open :: proc(ch: $C/Channel($T, $D)) -> bool {
-	c := ch._internal
-	return c != nil && !c.closed
-}
-
-
-channel_eq :: proc(a, b: $C/Channel($T, $D)) -> bool {
-	return a._internal == b._internal
-}
-channel_ne :: proc(a, b: $C/Channel($T, $D)) -> bool {
-	return a._internal != b._internal
-}
-
-
-channel_can_send :: proc(ch: $C/Channel($T, $D)) -> (ok: bool) where D >= .Both {
-	return raw_channel_can_send(ch._internal)
-}
-channel_can_recv :: proc(ch: $C/Channel($T, $D)) -> (ok: bool) where D <= .Both {
-	return raw_channel_can_recv(ch._internal)
-}
-
-
-channel_peek :: proc(ch: $C/Channel($T, $D)) -> int {
-	c := ch._internal
-	if c == nil {
-		return -1
-	}
-	if intrinsics.atomic_load(&c.closed) {
-		return -1
-	}
-	return intrinsics.atomic_load(&c.len)
-}
-
-
-channel_close :: proc(ch: $C/Channel($T, $D), loc := #caller_location) {
-	raw_channel_close(ch._internal, loc)
-}
-
-
-channel_iterator :: proc(ch: $C/Channel($T, $D)) -> (msg: T, ok: bool) where D <= .Both {
-	c := ch._internal
-	if c == nil {
-		return
-	}
-
-	if !c.closed || c.len > 0 {
-		msg, ok = channel_recv(ch), true
-	}
-	return
-}
-channel_drain :: proc(ch: $C/Channel($T, $D)) where D >= .Both {
-	raw_channel_drain(ch._internal)
-}
-
-
-channel_move :: proc(dst: $C1/Channel($T, $D1) src: $C2/Channel(T, $D2)) where D1 <= .Both, D2 >= .Both {
-	for msg in channel_iterator(src) {
-		channel_send(dst, msg)
-	}
-}
-
-
-Raw_Channel_Wait_Queue :: struct {
-	next: ^Raw_Channel_Wait_Queue,
-	state: ^uintptr,
-}
-
-
-Raw_Channel :: struct {
-	closed:      bool,
-	ready:       bool, // ready to recv
-	data_offset: u16,  // data is stored at the end of this data structure
-	elem_size:   u32,
-	len, cap:    int,
-	read, write: int,
-	mutex:       Mutex,
-	cond:        Condition,
-	allocator:   mem.Allocator,
-
-	sendq: ^Raw_Channel_Wait_Queue,
-	recvq: ^Raw_Channel_Wait_Queue,
-}
-
-raw_channel_wait_queue_insert :: proc(head: ^^Raw_Channel_Wait_Queue, val: ^Raw_Channel_Wait_Queue) {
-	val.next = head^
-	head^ = val
-}
-raw_channel_wait_queue_remove :: proc(head: ^^Raw_Channel_Wait_Queue, val: ^Raw_Channel_Wait_Queue) {
-	p := head
-	for p^ != nil && p^ != val {
-		p = &p^.next
-	}
-	if p != nil {
-		p^ = p^.next
-	}
-}
-
-
-raw_channel_create :: proc(elem_size, elem_align: int, cap := 0) -> ^Raw_Channel {
-	assert(int(u32(elem_size)) == elem_size)
-
-	s := size_of(Raw_Channel)
-	s = mem.align_forward_int(s, elem_align)
-	data_offset := uintptr(s)
-	s += elem_size * max(cap, 1)
-
-	a := max(elem_align, align_of(Raw_Channel))
-
-	c := (^Raw_Channel)(mem.alloc(s, a))
-	if c == nil {
-		return nil
-	}
-
-	c.data_offset = u16(data_offset)
-	c.elem_size = u32(elem_size)
-	c.len, c.cap = 0, max(cap, 0)
-	c.read, c.write = 0, 0
-	mutex_init(&c.mutex)
-	condition_init(&c.cond, &c.mutex)
-	c.allocator = context.allocator
-	c.closed = false
-
-	return c
-}
-
-
-raw_channel_destroy :: proc(c: ^Raw_Channel) {
-	if c == nil {
-		return
-	}
-	context.allocator = c.allocator
-	intrinsics.atomic_store(&c.closed, true)
-
-	condition_destroy(&c.cond)
-	mutex_destroy(&c.mutex)
-	free(c)
-}
-
-raw_channel_close :: proc(c: ^Raw_Channel, loc := #caller_location) {
-	if c == nil {
-		panic(message="cannot close nil channel", loc=loc)
-	}
-	mutex_lock(&c.mutex)
-	defer mutex_unlock(&c.mutex)
-	intrinsics.atomic_store(&c.closed, true)
-
-	// Release readers and writers
-	raw_channel_wait_queue_broadcast(c.recvq)
-	raw_channel_wait_queue_broadcast(c.sendq)
-	condition_broadcast(&c.cond)
-}
-
-
-
-raw_channel_send_impl :: proc(c: ^Raw_Channel, msg: rawptr, block: bool, loc := #caller_location) -> bool {
-	send :: proc(c: ^Raw_Channel, src: rawptr) {
-		data := uintptr(c) + uintptr(c.data_offset)
-		dst := data + uintptr(c.write * int(c.elem_size))
-		mem.copy(rawptr(dst), src, int(c.elem_size))
-		c.len += 1
-		c.write = (c.write + 1) % max(c.cap, 1)
-	}
-
-	switch {
-	case c == nil:
-		panic(message="cannot send message; channel is nil", loc=loc)
-	case c.closed:
-		panic(message="cannot send message; channel is closed", loc=loc)
-	}
-
-	mutex_lock(&c.mutex)
-	defer mutex_unlock(&c.mutex)
-
-	if c.cap > 0 {
-		if !block && c.len >= c.cap {
-			return false
-		}
-
-		for c.len >= c.cap {
-			condition_wait_for(&c.cond)
-		}
-	} else if c.len > 0 { // TODO(bill): determine correct behaviour
-		if !block {
-			return false
-		}
-		condition_wait_for(&c.cond)
-	} else if c.len == 0 && !block {
-		return false
-	}
-
-	send(c, msg)
-	condition_signal(&c.cond)
-	raw_channel_wait_queue_signal(c.recvq)
-
-	return true
-}
-
-raw_channel_recv_impl :: proc(c: ^Raw_Channel, res: rawptr, loc := #caller_location) {
-	recv :: proc(c: ^Raw_Channel, dst: rawptr, loc := #caller_location) {
-		if c.len < 1 {
-			panic(message="cannot recv message; channel is empty", loc=loc)
-		}
-		c.len -= 1
-
-		data := uintptr(c) + uintptr(c.data_offset)
-		src := data + uintptr(c.read * int(c.elem_size))
-		mem.copy(dst, rawptr(src), int(c.elem_size))
-		c.read = (c.read + 1) % max(c.cap, 1)
-	}
-
-	if c == nil {
-		panic(message="cannot recv message; channel is nil", loc=loc)
-	}
-	intrinsics.atomic_store(&c.ready, true)
-	for c.len < 1 {
-		raw_channel_wait_queue_signal(c.sendq)
-		condition_wait_for(&c.cond)
-	}
-	intrinsics.atomic_store(&c.ready, false)
-	recv(c, res, loc)
-	if c.cap > 0 {
-		if c.len == c.cap - 1 {
-			// NOTE(bill): Only signal on the last one
-			condition_signal(&c.cond)
-		}
-	} else {
-		condition_signal(&c.cond)
-	}
-}
-
-
-raw_channel_can_send :: proc(c: ^Raw_Channel) -> (ok: bool) {
-	if c == nil {
-		return false
-	}
-	mutex_lock(&c.mutex)
-	switch {
-	case c.closed:
-		ok = false
-	case c.cap > 0:
-		ok = c.ready && c.len < c.cap
-	case:
-		ok = c.ready && c.len == 0
-	}
-	mutex_unlock(&c.mutex)
-	return
-}
-raw_channel_can_recv :: proc(c: ^Raw_Channel) -> (ok: bool) {
-	if c == nil {
-		return false
-	}
-	mutex_lock(&c.mutex)
-	ok = c.len > 0
-	mutex_unlock(&c.mutex)
-	return
-}
-
-
-raw_channel_drain :: proc(c: ^Raw_Channel) {
-	if c == nil {
-		return
-	}
-	mutex_lock(&c.mutex)
-	c.len   = 0
-	c.read  = 0
-	c.write = 0
-	mutex_unlock(&c.mutex)
-}
-
-
-
-MAX_SELECT_CHANNELS :: 64
-SELECT_MAX_TIMEOUT :: max(time.Duration)
-
-Select_Command :: enum {
-	Recv,
-	Send,
-}
-
-Select_Channel :: struct {
-	channel: ^Raw_Channel,
-	command: Select_Command,
-}
-
-
-
-select :: proc(channels: ..Select_Channel) -> (index: int) {
-	return select_timeout(SELECT_MAX_TIMEOUT, ..channels)
-}
-select_timeout :: proc(timeout: time.Duration, channels: ..Select_Channel) -> (index: int) {
-	switch len(channels) {
-	case 0:
-		panic("sync: select with no channels")
-	}
-
-	assert(len(channels) <= MAX_SELECT_CHANNELS)
-
-	backing: [MAX_SELECT_CHANNELS]int
-	queues:  [MAX_SELECT_CHANNELS]Raw_Channel_Wait_Queue
-	candidates := backing[:]
-	cap := len(channels)
-	candidates = candidates[:cap]
-
-	count := u32(0)
-	for c, i in channels {
-		if c.channel == nil {
-			continue
-		}
-		switch c.command {
-		case .Recv:
-			if raw_channel_can_recv(c.channel) {
-				candidates[count] = i
-				count += 1
-			}
-		case .Send:
-			if raw_channel_can_send(c.channel) {
-				candidates[count] = i
-				count += 1
-			}
-		}
-	}
-
-	if count == 0 {
-		wait_state: uintptr = 0
-		for _, i in channels {
-			q := &queues[i]
-			q.state = &wait_state
-		}
-
-		for c, i in channels {
-			if c.channel == nil {
-				continue
-			}
-			q := &queues[i]
-			switch c.command {
-			case .Recv: raw_channel_wait_queue_insert(&c.channel.recvq, q)
-			case .Send: raw_channel_wait_queue_insert(&c.channel.sendq, q)
-			}
-		}
-		raw_channel_wait_queue_wait_on(&wait_state, timeout)
-		for c, i in channels {
-			if c.channel == nil {
-				continue
-			}
-			q := &queues[i]
-			switch c.command {
-			case .Recv: raw_channel_wait_queue_remove(&c.channel.recvq, q)
-			case .Send: raw_channel_wait_queue_remove(&c.channel.sendq, q)
-			}
-		}
-
-		for c, i in channels {
-			switch c.command {
-			case .Recv:
-				if raw_channel_can_recv(c.channel) {
-					candidates[count] = i
-					count += 1
-				}
-			case .Send:
-				if raw_channel_can_send(c.channel) {
-					candidates[count] = i
-					count += 1
-				}
-			}
-		}
-		if count == 0 && timeout == SELECT_MAX_TIMEOUT {
-			index = -1
-			return
-		}
-
-		assert(count != 0)
-	}
-
-	t := time.now()
-	r := rand.create(transmute(u64)t)
-	i := rand.uint32(&r)
-
-	index = candidates[i % count]
-	return
-}
-
-select_recv :: proc(channels: ..^Raw_Channel) -> (index: int) {
-	switch len(channels) {
-	case 0:
-		panic("sync: select with no channels")
-	}
-
-	assert(len(channels) <= MAX_SELECT_CHANNELS)
-
-	backing: [MAX_SELECT_CHANNELS]int
-	queues:  [MAX_SELECT_CHANNELS]Raw_Channel_Wait_Queue
-	candidates := backing[:]
-	cap := len(channels)
-	candidates = candidates[:cap]
-
-	count := u32(0)
-	for c, i in channels {
-		if raw_channel_can_recv(c) {
-			candidates[count] = i
-			count += 1
-		}
-	}
-
-	if count == 0 {
-		state: uintptr
-		for c, i in channels {
-			q := &queues[i]
-			q.state = &state
-			raw_channel_wait_queue_insert(&c.recvq, q)
-		}
-		raw_channel_wait_queue_wait_on(&state, SELECT_MAX_TIMEOUT)
-		for c, i in channels {
-			q := &queues[i]
-			raw_channel_wait_queue_remove(&c.recvq, q)
-		}
-
-		for c, i in channels {
-			if raw_channel_can_recv(c) {
-				candidates[count] = i
-				count += 1
-			}
-		}
-		assert(count != 0)
-	}
-
-	t := time.now()
-	r := rand.create(transmute(u64)t)
-	i := rand.uint32(&r)
-
-	index = candidates[i % count]
-	return
-}
-
-select_recv_msg :: proc(channels: ..$C/Channel($T, $D)) -> (msg: T, index: int) {
-	switch len(channels) {
-	case 0:
-		panic("sync: select with no channels")
-	}
-
-	assert(len(channels) <= MAX_SELECT_CHANNELS)
-
-	queues:  [MAX_SELECT_CHANNELS]Raw_Channel_Wait_Queue
-	candidates: [MAX_SELECT_CHANNELS]int
-
-	count := u32(0)
-	for c, i in channels {
-		if raw_channel_can_recv(c) {
-			candidates[count] = i
-			count += 1
-		}
-	}
-
-	if count == 0 {
-		state: uintptr
-		for c, i in channels {
-			q := &queues[i]
-			q.state = &state
-			raw_channel_wait_queue_insert(&c.recvq, q)
-		}
-		raw_channel_wait_queue_wait_on(&state, SELECT_MAX_TIMEOUT)
-		for c, i in channels {
-			q := &queues[i]
-			raw_channel_wait_queue_remove(&c.recvq, q)
-		}
-
-		for c, i in channels {
-			if raw_channel_can_recv(c) {
-				candidates[count] = i
-				count += 1
-			}
-		}
-		assert(count != 0)
-	}
-
-	t := time.now()
-	r := rand.create(transmute(u64)t)
-	i := rand.uint32(&r)
-
-	index = candidates[i % count]
-	msg = channel_recv(channels[index])
-
-	return
-}
-
-select_send_msg :: proc(msg: $T, channels: ..$C/Channel(T, $D)) -> (index: int) {
-	switch len(channels) {
-	case 0:
-		panic("sync: select with no channels")
-	}
-
-	assert(len(channels) <= MAX_SELECT_CHANNELS)
-
-	backing: [MAX_SELECT_CHANNELS]int
-	queues:  [MAX_SELECT_CHANNELS]Raw_Channel_Wait_Queue
-	candidates := backing[:]
-	cap := len(channels)
-	candidates = candidates[:cap]
-
-	count := u32(0)
-	for c, i in channels {
-		if raw_channel_can_recv(c) {
-			candidates[count] = i
-			count += 1
-		}
-	}
-
-	if count == 0 {
-		state: uintptr
-		for c, i in channels {
-			q := &queues[i]
-			q.state = &state
-			raw_channel_wait_queue_insert(&c.recvq, q)
-		}
-		raw_channel_wait_queue_wait_on(&state, SELECT_MAX_TIMEOUT)
-		for c, i in channels {
-			q := &queues[i]
-			raw_channel_wait_queue_remove(&c.recvq, q)
-		}
-
-		for c, i in channels {
-			if raw_channel_can_recv(c) {
-				candidates[count] = i
-				count += 1
-			}
-		}
-		assert(count != 0)
-	}
-
-	t := time.now()
-	r := rand.create(transmute(u64)t)
-	i := rand.uint32(&r)
-
-	index = candidates[i % count]
-
-	if msg != nil {
-		channel_send(channels[index], msg)
-	}
-
-	return
-}
-
-select_send :: proc(channels: ..^Raw_Channel) -> (index: int) {
-	switch len(channels) {
-	case 0:
-		panic("sync: select with no channels")
-	}
-
-	assert(len(channels) <= MAX_SELECT_CHANNELS)
-	candidates: [MAX_SELECT_CHANNELS]int
-	queues: [MAX_SELECT_CHANNELS]Raw_Channel_Wait_Queue
-
-	count := u32(0)
-	for c, i in channels {
-		if raw_channel_can_send(c) {
-			candidates[count] = i
-			count += 1
-		}
-	}
-
-	if count == 0 {
-		state: uintptr
-		for c, i in channels {
-			q := &queues[i]
-			q.state = &state
-			raw_channel_wait_queue_insert(&c.sendq, q)
-		}
-		raw_channel_wait_queue_wait_on(&state, SELECT_MAX_TIMEOUT)
-		for c, i in channels {
-			q := &queues[i]
-			raw_channel_wait_queue_remove(&c.sendq, q)
-		}
-
-		for c, i in channels {
-			if raw_channel_can_send(c) {
-				candidates[count] = i
-				count += 1
-			}
-		}
-		assert(count != 0)
-	}
-
-	t := time.now()
-	r := rand.create(transmute(u64)t)
-	i := rand.uint32(&r)
-
-	index = candidates[i % count]
-	return
-}
-
-select_try :: proc(channels: ..Select_Channel) -> (index: int) {
-	switch len(channels) {
-	case 0:
-		panic("sync: select with no channels")
-	}
-
-	assert(len(channels) <= MAX_SELECT_CHANNELS)
-
-	backing: [MAX_SELECT_CHANNELS]int
-	candidates := backing[:]
-	cap := len(channels)
-	candidates = candidates[:cap]
-
-	count := u32(0)
-	for c, i in channels {
-		switch c.command {
-		case .Recv:
-			if raw_channel_can_recv(c.channel) {
-				candidates[count] = i
-				count += 1
-			}
-		case .Send:
-			if raw_channel_can_send(c.channel) {
-				candidates[count] = i
-				count += 1
-			}
-		}
-	}
-
-	if count == 0 {
-		index = -1
-		return
-	}
-
-	t := time.now()
-	r := rand.create(transmute(u64)t)
-	i := rand.uint32(&r)
-
-	index = candidates[i % count]
-	return
-}
-
-
-select_try_recv :: proc(channels: ..^Raw_Channel) -> (index: int) {
-	switch len(channels) {
-	case 0:
-		index = -1
-		return
-	case 1:
-		index = -1
-		if raw_channel_can_recv(channels[0]) {
-			index = 0
-		}
-		return
-	}
-
-	assert(len(channels) <= MAX_SELECT_CHANNELS)
-	candidates: [MAX_SELECT_CHANNELS]int
-
-	count := u32(0)
-	for c, i in channels {
-		if raw_channel_can_recv(c) {
-			candidates[count] = i
-			count += 1
-		}
-	}
-
-	if count == 0 {
-		index = -1
-		return
-	}
-
-	t := time.now()
-	r := rand.create(transmute(u64)t)
-	i := rand.uint32(&r)
-
-	index = candidates[i % count]
-	return
-}
-
-
-select_try_send :: proc(channels: ..^Raw_Channel) -> (index: int) #no_bounds_check {
-	switch len(channels) {
-	case 0:
-		return -1
-	case 1:
-		if raw_channel_can_send(channels[0]) {
-			return 0
-		}
-		return -1
-	}
-
-	assert(len(channels) <= MAX_SELECT_CHANNELS)
-	candidates: [MAX_SELECT_CHANNELS]int
-
-	count := u32(0)
-	for c, i in channels {
-		if raw_channel_can_send(c) {
-			candidates[count] = i
-			count += 1
-		}
-	}
-
-	if count == 0 {
-		index = -1
-		return
-	}
-
-	t := time.now()
-	r := rand.create(transmute(u64)t)
-	i := rand.uint32(&r)
-
-	index = candidates[i % count]
-	return
-}
-
-select_try_recv_msg :: proc(channels: ..$C/Channel($T, $D)) -> (msg: T, index: int) {
-	switch len(channels) {
-	case 0:
-		index = -1
-		return
-	case 1:
-		ok: bool
-		if msg, ok = channel_try_recv(channels[0]); ok {
-			index = 0
-		}
-		return
-	}
-
-	assert(len(channels) <= MAX_SELECT_CHANNELS)
-	candidates: [MAX_SELECT_CHANNELS]int
-
-	count := u32(0)
-	for c, i in channels {
-		if channel_can_recv(c) {
-			candidates[count] = i
-			count += 1
-		}
-	}
-
-	if count == 0 {
-		index = -1
-		return
-	}
-
-	t := time.now()
-	r := rand.create(transmute(u64)t)
-	i := rand.uint32(&r)
-
-	index = candidates[i % count]
-	msg = channel_recv(channels[index])
-	return
-}
-
-select_try_send_msg :: proc(msg: $T, channels: ..$C/Channel(T, $D)) -> (index: int) {
-	index = -1
-	switch len(channels) {
-	case 0:
-		return
-	case 1:
-		if channel_try_send(channels[0], msg) {
-			index = 0
-		}
-		return
-	}
-
-
-	assert(len(channels) <= MAX_SELECT_CHANNELS)
-	candidates: [MAX_SELECT_CHANNELS]int
-
-	count := u32(0)
-	for c, i in channels {
-		if raw_channel_can_send(c) {
-			candidates[count] = i
-			count += 1
-		}
-	}
-
-	if count == 0 {
-		index = -1
-		return
-	}
-
-	t := time.now()
-	r := rand.create(transmute(u64)t)
-	i := rand.uint32(&r)
-
-	index = candidates[i % count]
-	channel_send(channels[index], msg)
-	return
-}
-
diff --git a/core/sync/channel_unix.odin b/core/sync/channel_unix.odin
deleted file mode 100644
index 47aa46004..000000000
--- a/core/sync/channel_unix.odin
+++ /dev/null
@@ -1,16 +0,0 @@
-// +build linux, darwin, freebsd, openbsd
-package sync
-
-import "core:time"
-
-raw_channel_wait_queue_wait_on :: proc(state: ^uintptr, timeout: time.Duration) {
-	// stub
-}
-
-raw_channel_wait_queue_signal :: proc(q: ^Raw_Channel_Wait_Queue) {
-	// stub
-}
-
-raw_channel_wait_queue_broadcast :: proc(q: ^Raw_Channel_Wait_Queue) {
-	// stub
-}
diff --git a/core/sync/channel_windows.odin b/core/sync/channel_windows.odin
deleted file mode 100644
index 5d469ffff..000000000
--- a/core/sync/channel_windows.odin
+++ /dev/null
@@ -1,33 +0,0 @@
-package sync
-
-import "core:intrinsics"
-import win32 "core:sys/windows"
-import "core:time"
-
-raw_channel_wait_queue_wait_on :: proc(state: ^uintptr, timeout: time.Duration) {
-	ms: win32.DWORD = win32.INFINITE
-	if max(time.Duration) != SELECT_MAX_TIMEOUT {
-		ms = win32.DWORD((max(time.duration_nanoseconds(timeout), 0) + 999999)/1000000)
-	}
-
-	v := intrinsics.atomic_load(state)
-	for v == 0 {
-		win32.WaitOnAddress(state, &v, size_of(state^), ms)
-		v = intrinsics.atomic_load(state)
-	}
-	intrinsics.atomic_store(state, 0)
-}
-
-raw_channel_wait_queue_signal :: proc(q: ^Raw_Channel_Wait_Queue) {
-	for x := q; x != nil; x = x.next {
-		intrinsics.atomic_add(x.state, 1)
-		win32.WakeByAddressSingle(x.state)
-	}
-}
-
-raw_channel_wait_queue_broadcast :: proc(q: ^Raw_Channel_Wait_Queue) {
-	for x := q; x != nil; x = x.next {
-		intrinsics.atomic_add(x.state, 1)
-		win32.WakeByAddressAll(x.state)
-	}
-}
diff --git a/core/sync/sync2/extended.odin b/core/sync/extended.odin
similarity index 85%
rename from core/sync/sync2/extended.odin
rename to core/sync/extended.odin
index deb48a22d..24f7c096c 100644
--- a/core/sync/sync2/extended.odin
+++ b/core/sync/extended.odin
@@ -1,4 +1,4 @@
-package sync2
+package sync
 
 import "core:time"
 
@@ -146,10 +146,10 @@ Auto_Reset_Event :: struct {
 }
 
 auto_reset_event_signal :: proc(e: ^Auto_Reset_Event) {
-	old_status := atomic_load_relaxed(&e.status)
+	old_status := atomic_load_explicit(&e.status, .Relaxed)
 	for {
 		new_status := old_status + 1 if old_status < 1 else 1
-		if _, ok := atomic_compare_exchange_weak_release(&e.status, old_status, new_status); ok {
+		if _, ok := atomic_compare_exchange_weak_explicit(&e.status, old_status, new_status, .Release, .Relaxed); ok {
 			break
 		}
 
@@ -160,7 +160,7 @@ auto_reset_event_signal :: proc(e: ^Auto_Reset_Event) {
 }
 
 auto_reset_event_wait :: proc(e: ^Auto_Reset_Event) {
-	old_status := atomic_sub_acquire(&e.status, 1)
+	old_status := atomic_sub_explicit(&e.status, 1, .Acquire)
 	if old_status < 1 {
 		sema_wait(&e.sema)
 	}
@@ -174,14 +174,14 @@ Ticket_Mutex :: struct {
 }
 
 ticket_mutex_lock :: #force_inline proc(m: ^Ticket_Mutex) {
-	ticket := atomic_add_relaxed(&m.ticket, 1)
-	for ticket != atomic_load_acquire(&m.serving) {
+	ticket := atomic_add_explicit(&m.ticket, 1, .Relaxed)
+	for ticket != atomic_load_explicit(&m.serving, .Acquire) {
 		cpu_relax()
 	}
 }
 
 ticket_mutex_unlock :: #force_inline proc(m: ^Ticket_Mutex) {
-	atomic_add_relaxed(&m.serving, 1)
+	atomic_add_explicit(&m.serving, 1, .Relaxed)
 }
 @(deferred_in=ticket_mutex_unlock)
 ticket_mutex_guard :: proc(m: ^Ticket_Mutex) -> bool {
@@ -196,18 +196,18 @@ Benaphore :: struct {
 }
 
 benaphore_lock :: proc(b: ^Benaphore) {
-	if atomic_add_acquire(&b.counter, 1) > 1 {
+	if atomic_add_explicit(&b.counter, 1, .Acquire) > 1 {
 		sema_wait(&b.sema)
 	}
 }
 
 benaphore_try_lock :: proc(b: ^Benaphore) -> bool {
-	v, _ := atomic_compare_exchange_strong_acquire(&b.counter, 1, 0)
+	v, _ := atomic_compare_exchange_strong_explicit(&b.counter, 1, 0, .Acquire, .Acquire)
 	return v == 0
 }
 
 benaphore_unlock :: proc(b: ^Benaphore) {
-	if atomic_sub_release(&b.counter, 1) > 0 {
+	if atomic_sub_explicit(&b.counter, 1, .Release) > 0 {
 		sema_post(&b.sema)
 	}
 }
@@ -227,7 +227,7 @@ Recursive_Benaphore :: struct {
 
 recursive_benaphore_lock :: proc(b: ^Recursive_Benaphore) {
 	tid := current_thread_id()
-	if atomic_add_acquire(&b.counter, 1) > 1 {
+	if atomic_add_explicit(&b.counter, 1, .Acquire) > 1 {
 		if tid != b.owner {
 			sema_wait(&b.sema)
 		}
@@ -240,10 +240,10 @@ recursive_benaphore_lock :: proc(b: ^Recursive_Benaphore) {
 recursive_benaphore_try_lock :: proc(b: ^Recursive_Benaphore) -> bool {
 	tid := current_thread_id()
 	if b.owner == tid {
-		atomic_add_acquire(&b.counter, 1)
+		atomic_add_explicit(&b.counter, 1, .Acquire)
 	}
 
-	if v, _ := atomic_compare_exchange_strong_acquire(&b.counter, 1, 0); v != 0 {
+	if v, _ := atomic_compare_exchange_strong_explicit(&b.counter, 1, 0, .Acquire, .Acquire); v != 0 {
 		return false
 	}
 	// inside the lock
@@ -260,7 +260,7 @@ recursive_benaphore_unlock :: proc(b: ^Recursive_Benaphore) {
 	if recursion == 0 {
 		b.owner = 0
 	}
-	if atomic_sub_release(&b.counter, 1) > 0 {
+	if atomic_sub_explicit(&b.counter, 1, .Release) > 0 {
 		if recursion == 0 {
 			sema_post(&b.sema)
 		}
@@ -293,12 +293,12 @@ once_do :: proc(o: ^Once, fn: proc()) {
 		defer mutex_unlock(&o.m)
 		if !o.done {
 			fn()
-			atomic_store_release(&o.done, true)
+			atomic_store_explicit(&o.done, true, .Release)
 		}
 	}
 
 	
-	if atomic_load_acquire(&o.done) == false {
+	if atomic_load_explicit(&o.done, .Acquire) == false {
 		do_slow(o, fn)
 	}
 }
diff --git a/core/sync/sync2/futex_darwin.odin b/core/sync/futex_darwin.odin
similarity index 99%
rename from core/sync/sync2/futex_darwin.odin
rename to core/sync/futex_darwin.odin
index 9dad8d375..88e354827 100644
--- a/core/sync/sync2/futex_darwin.odin
+++ b/core/sync/futex_darwin.odin
@@ -1,6 +1,6 @@
 //+private
 //+build darwin
-package sync2
+package sync
 
 import "core:c"
 import "core:time"
diff --git a/core/sync/sync2/futex_freebsd.odin b/core/sync/futex_freebsd.odin
similarity index 98%
rename from core/sync/sync2/futex_freebsd.odin
rename to core/sync/futex_freebsd.odin
index 2cbdb4aaa..2e1d065bc 100644
--- a/core/sync/sync2/futex_freebsd.odin
+++ b/core/sync/futex_freebsd.odin
@@ -1,6 +1,6 @@
 //+private
 //+build freebsd
-package sync2
+package sync
 
 import "core:c"
 import "core:os"
diff --git a/core/sync/sync2/futex_linux.odin b/core/sync/futex_linux.odin
similarity index 94%
rename from core/sync/sync2/futex_linux.odin
rename to core/sync/futex_linux.odin
index fca28cace..c429a9d64 100644
--- a/core/sync/sync2/futex_linux.odin
+++ b/core/sync/futex_linux.odin
@@ -1,6 +1,6 @@
 //+private
 //+build linux
-package sync2
+package sync
 
 import "core:c"
 import "core:time"
@@ -14,12 +14,6 @@ FUTEX_PRIVATE_FLAG :: 128
 FUTEX_WAIT_PRIVATE :: (FUTEX_WAIT | FUTEX_PRIVATE_FLAG)
 FUTEX_WAKE_PRIVATE :: (FUTEX_WAKE | FUTEX_PRIVATE_FLAG)
 
-foreign import libc "system:c"
-
-foreign libc {
-	__errno_location :: proc "c" () -> ^c.int ---	
-}
-
 ESUCCESS  :: 0
 EINTR     :: -4
 EAGAIN    :: -11
diff --git a/core/sync/sync2/futex_openbsd.odin b/core/sync/futex_openbsd.odin
similarity index 99%
rename from core/sync/sync2/futex_openbsd.odin
rename to core/sync/futex_openbsd.odin
index dbc80747b..6ac9d3efb 100644
--- a/core/sync/sync2/futex_openbsd.odin
+++ b/core/sync/futex_openbsd.odin
@@ -1,6 +1,6 @@
 //+private
 //+build openbsd
-package sync2
+package sync
 
 import "core:c"
 import "core:os"
diff --git a/core/sync/sync2/futex_windows.odin b/core/sync/futex_windows.odin
similarity index 98%
rename from core/sync/sync2/futex_windows.odin
rename to core/sync/futex_windows.odin
index 200a119ff..1c9d8b845 100644
--- a/core/sync/sync2/futex_windows.odin
+++ b/core/sync/futex_windows.odin
@@ -1,6 +1,6 @@
 //+private
 //+build windows
-package sync2
+package sync
 
 import "core:time"
 
diff --git a/core/sync/sync2/primitives.odin b/core/sync/primitives.odin
similarity index 99%
rename from core/sync/sync2/primitives.odin
rename to core/sync/primitives.odin
index 6d056d439..483f85343 100644
--- a/core/sync/sync2/primitives.odin
+++ b/core/sync/primitives.odin
@@ -1,4 +1,4 @@
-package sync2
+package sync
 
 import "core:time"
 
diff --git a/core/sync/sync2/primitives_atomic.odin b/core/sync/primitives_atomic.odin
similarity index 94%
rename from core/sync/sync2/primitives_atomic.odin
rename to core/sync/primitives_atomic.odin
index dd082029b..11fff4e60 100644
--- a/core/sync/sync2/primitives_atomic.odin
+++ b/core/sync/primitives_atomic.odin
@@ -1,4 +1,4 @@
-package sync2
+package sync
 
 import "core:time"
 
@@ -24,7 +24,7 @@ atomic_mutex_lock :: proc(m: ^Atomic_Mutex) {
 		new_state := curr_state // Make a copy of it
 
 		spin_lock: for spin in 0..<i32(100) {
-			state, ok := atomic_compare_exchange_weak_acquire(&m.state, .Unlocked, new_state)
+			state, ok := atomic_compare_exchange_weak_explicit(&m.state, .Unlocked, new_state, .Acquire, .Consume)
 			if ok {
 				return
 			}
@@ -42,7 +42,7 @@ atomic_mutex_lock :: proc(m: ^Atomic_Mutex) {
 		new_state = .Waiting
 
 		for {
-			if atomic_exchange_acquire(&m.state, .Waiting) == .Unlocked {
+			if atomic_exchange_explicit(&m.state, .Waiting, .Acquire) == .Unlocked {
 				return
 			}
 			
@@ -52,7 +52,7 @@ atomic_mutex_lock :: proc(m: ^Atomic_Mutex) {
 	}
 
 
-	if v := atomic_exchange_acquire(&m.state, .Locked); v != .Unlocked {
+	if v := atomic_exchange_explicit(&m.state, .Locked, .Acquire); v != .Unlocked {
 		lock_slow(m, v)
 	}
 }
@@ -65,7 +65,7 @@ atomic_mutex_unlock :: proc(m: ^Atomic_Mutex) {
 	}
 
 
-	switch atomic_exchange_release(&m.state, .Unlocked) {
+	switch atomic_exchange_explicit(&m.state, .Unlocked, .Release) {
 	case .Unlocked:
 		unreachable()
 	case .Locked:
@@ -77,7 +77,7 @@ atomic_mutex_unlock :: proc(m: ^Atomic_Mutex) {
 
 // atomic_mutex_try_lock tries to lock m, will return true on success, and false on failure
 atomic_mutex_try_lock :: proc(m: ^Atomic_Mutex) -> bool {
-	_, ok := atomic_compare_exchange_strong_acquire(&m.state, .Unlocked, .Locked)
+	_, ok := atomic_compare_exchange_strong_explicit(&m.state, .Unlocked, .Locked, .Acquire, .Consume)
 	return ok
 }
 
@@ -290,7 +290,7 @@ Queue_Item :: struct {
 
 @(private="file")
 queue_item_wait :: proc(item: ^Queue_Item) {
-	for atomic_load_acquire(&item.futex) == 0 {
+	for atomic_load_explicit(&item.futex, .Acquire) == 0 {
 		futex_wait(&item.futex, 0)
 		cpu_relax()
 	}
@@ -298,7 +298,7 @@ queue_item_wait :: proc(item: ^Queue_Item) {
 @(private="file")
 queue_item_wait_with_timeout :: proc(item: ^Queue_Item, duration: time.Duration) -> bool {
 	start := time.tick_now()
-	for atomic_load_acquire(&item.futex) == 0 {
+	for atomic_load_explicit(&item.futex, .Acquire) == 0 {
 		remaining := duration - time.tick_since(start)
 		if remaining < 0 {
 			return false
@@ -312,7 +312,7 @@ queue_item_wait_with_timeout :: proc(item: ^Queue_Item, duration: time.Duration)
 }
 @(private="file")
 queue_item_signal :: proc(item: ^Queue_Item) {
-	atomic_store_release(&item.futex, 1)
+	atomic_store_explicit(&item.futex, 1, .Release)
 	futex_signal(&item.futex)
 }
 
diff --git a/core/sync/sync2/primitives_darwin.odin b/core/sync/primitives_darwin.odin
similarity index 98%
rename from core/sync/sync2/primitives_darwin.odin
rename to core/sync/primitives_darwin.odin
index 66995bd01..514f66f3e 100644
--- a/core/sync/sync2/primitives_darwin.odin
+++ b/core/sync/primitives_darwin.odin
@@ -1,6 +1,6 @@
 //+build darwin
 //+private
-package sync2
+package sync
 
 import "core:c"
 import "core:time"
diff --git a/core/sync/primitives_freebsd.odin b/core/sync/primitives_freebsd.odin
new file mode 100644
index 000000000..b88fca181
--- /dev/null
+++ b/core/sync/primitives_freebsd.odin
@@ -0,0 +1,46 @@
+//+build freebsd
+//+private
+package sync
+
+import "core:os"
+import "core:time"
+
+_current_thread_id :: proc "contextless" () -> int {
+	return os.current_thread_id()
+}
+
+_Mutex :: struct {
+	mutex: Atomic_Mutex,
+}
+
+_mutex_lock :: proc(m: ^Mutex) {
+	atomic_mutex_lock(&m.impl.mutex)
+}
+
+_mutex_unlock :: proc(m: ^Mutex) {
+	atomic_mutex_unlock(&m.impl.mutex)
+}
+
+_mutex_try_lock :: proc(m: ^Mutex) -> bool {
+	return atomic_mutex_try_lock(&m.impl.mutex)
+}
+
+_Cond :: struct {
+	cond: Atomic_Cond,
+}
+
+_cond_wait :: proc(c: ^Cond, m: ^Mutex) {
+	atomic_cond_wait(&c.impl.cond, &m.impl.mutex)
+}
+
+_cond_wait_with_timeout :: proc(c: ^Cond, m: ^Mutex, duration: time.Duration) -> bool {
+	return atomic_cond_wait_with_timeout(&c.impl.cond, &m.impl.mutex, duration)
+}
+
+_cond_signal :: proc(c: ^Cond) {
+	atomic_cond_signal(&c.impl.cond)
+}
+
+_cond_broadcast :: proc(c: ^Cond) {
+	atomic_cond_broadcast(&c.impl.cond)
+}
diff --git a/core/sync/primitives_internal.odin b/core/sync/primitives_internal.odin
new file mode 100644
index 000000000..15967d45e
--- /dev/null
+++ b/core/sync/primitives_internal.odin
@@ -0,0 +1,125 @@
+//+private
+package sync
+
+when #config(ODIN_SYNC_RECURSIVE_MUTEX_USE_FUTEX, true) {
+	_Recursive_Mutex :: struct {
+		owner:     Futex,
+		recursion: i32,
+	}
+
+	_recursive_mutex_lock :: proc(m: ^Recursive_Mutex) {
+		tid := Futex(current_thread_id())
+		for {
+			prev_owner := atomic_compare_exchange_strong_explicit(&m.impl.owner, tid, 0, .Acquire, .Acquire)
+			switch prev_owner {
+			case 0, tid:
+				m.impl.recursion += 1
+				// inside the lock
+				return
+			}
+
+			futex_wait(&m.impl.owner, u32(prev_owner))
+		}
+	}
+
+	_recursive_mutex_unlock :: proc(m: ^Recursive_Mutex) {
+		m.impl.recursion -= 1
+		if m.impl.recursion != 0 {
+			return
+		}
+		atomic_exchange_explicit(&m.impl.owner, 0, .Release)
+		
+		futex_signal(&m.impl.owner)
+		// outside the lock
+
+	}
+
+	_recursive_mutex_try_lock :: proc(m: ^Recursive_Mutex) -> bool {
+		tid := Futex(current_thread_id())
+		prev_owner := atomic_compare_exchange_strong_explicit(&m.impl.owner, tid, 0, .Acquire, .Acquire)
+		switch prev_owner {
+		case 0, tid:
+			m.impl.recursion += 1
+			// inside the lock
+			return true
+		}
+		return false
+	}
+} else {
+	_Recursive_Mutex :: struct {
+		owner:     int,
+		recursion: int,
+		mutex:     Mutex,
+	}
+
+	_recursive_mutex_lock :: proc(m: ^Recursive_Mutex) {
+		tid := current_thread_id()
+		if tid != m.impl.owner {
+			mutex_lock(&m.impl.mutex)
+		}
+		// inside the lock
+		m.impl.owner = tid
+		m.impl.recursion += 1
+	}
+
+	_recursive_mutex_unlock :: proc(m: ^Recursive_Mutex) {
+		tid := current_thread_id()
+		assert(tid == m.impl.owner)
+		m.impl.recursion -= 1
+		recursion := m.impl.recursion
+		if recursion == 0 {
+			m.impl.owner = 0
+		}
+		if recursion == 0 {
+			mutex_unlock(&m.impl.mutex)
+		}
+		// outside the lock
+
+	}
+
+	_recursive_mutex_try_lock :: proc(m: ^Recursive_Mutex) -> bool {
+		tid := current_thread_id()
+		if m.impl.owner == tid {
+			return mutex_try_lock(&m.impl.mutex)
+		}
+		if !mutex_try_lock(&m.impl.mutex) {
+			return false
+		}
+		// inside the lock
+		m.impl.owner = tid
+		m.impl.recursion += 1
+		return true
+	}
+}
+
+
+when ODIN_OS != .Windows {
+	_RW_Mutex :: struct {
+		mutex: Atomic_RW_Mutex,
+	}
+
+	_rw_mutex_lock :: proc(rw: ^RW_Mutex) {
+		atomic_rw_mutex_lock(&rw.impl.mutex)
+	}
+
+	_rw_mutex_unlock :: proc(rw: ^RW_Mutex) {
+		atomic_rw_mutex_unlock(&rw.impl.mutex)
+	}
+
+	_rw_mutex_try_lock :: proc(rw: ^RW_Mutex) -> bool {
+		return atomic_rw_mutex_try_lock(&rw.impl.mutex)
+	}
+
+	_rw_mutex_shared_lock :: proc(rw: ^RW_Mutex) {
+		atomic_rw_mutex_shared_lock(&rw.impl.mutex)
+	}
+
+	_rw_mutex_shared_unlock :: proc(rw: ^RW_Mutex) {
+		atomic_rw_mutex_shared_unlock(&rw.impl.mutex)
+	}
+
+	_rw_mutex_try_shared_lock :: proc(rw: ^RW_Mutex) -> bool {
+		return atomic_rw_mutex_try_shared_lock(&rw.impl.mutex)
+	}
+
+}
\ No newline at end of file
diff --git a/core/sync/primitives_linux.odin b/core/sync/primitives_linux.odin
new file mode 100644
index 000000000..0a9f0cc33
--- /dev/null
+++ b/core/sync/primitives_linux.odin
@@ -0,0 +1,47 @@
+//+build linux
+//+private
+package sync
+
+import "core:sys/unix"
+import "core:time"
+
+_current_thread_id :: proc "contextless" () -> int {
+	return unix.sys_gettid()
+}
+
+
+_Mutex :: struct {
+	mutex: Atomic_Mutex,
+}
+
+_mutex_lock :: proc(m: ^Mutex) {
+	atomic_mutex_lock(&m.impl.mutex)
+}
+
+_mutex_unlock :: proc(m: ^Mutex) {
+	atomic_mutex_unlock(&m.impl.mutex)
+}
+
+_mutex_try_lock :: proc(m: ^Mutex) -> bool {
+	return atomic_mutex_try_lock(&m.impl.mutex)
+}
+
+_Cond :: struct {
+	cond: Atomic_Cond,
+}
+
+_cond_wait :: proc(c: ^Cond, m: ^Mutex) {
+	atomic_cond_wait(&c.impl.cond, &m.impl.mutex)
+}
+
+_cond_wait_with_timeout :: proc(c: ^Cond, m: ^Mutex, duration: time.Duration) -> bool {
+	return atomic_cond_wait_with_timeout(&c.impl.cond, &m.impl.mutex, duration)
+}
+
+_cond_signal :: proc(c: ^Cond) {
+	atomic_cond_signal(&c.impl.cond)
+}
+
+_cond_broadcast :: proc(c: ^Cond) {
+	atomic_cond_broadcast(&c.impl.cond)
+}
diff --git a/core/sync/primitives_openbsd.odin b/core/sync/primitives_openbsd.odin
new file mode 100644
index 000000000..7794016f8
--- /dev/null
+++ b/core/sync/primitives_openbsd.odin
@@ -0,0 +1,46 @@
+//+build openbsd
+//+private
+package sync
+
+import "core:os"
+import "core:time"
+
+_current_thread_id :: proc "contextless" () -> int {
+	return os.current_thread_id()
+}
+
+_Mutex :: struct {
+	mutex: Atomic_Mutex,
+}
+
+_mutex_lock :: proc(m: ^Mutex) {
+	atomic_mutex_lock(&m.impl.mutex)
+}
+
+_mutex_unlock :: proc(m: ^Mutex) {
+	atomic_mutex_unlock(&m.impl.mutex)
+}
+
+_mutex_try_lock :: proc(m: ^Mutex) -> bool {
+	return atomic_mutex_try_lock(&m.impl.mutex)
+}
+
+_Cond :: struct {
+	cond: Atomic_Cond,
+}
+
+_cond_wait :: proc(c: ^Cond, m: ^Mutex) {
+	atomic_cond_wait(&c.impl.cond, &m.impl.mutex)
+}
+
+_cond_wait_with_timeout :: proc(c: ^Cond, m: ^Mutex, duration: time.Duration) -> bool {
+	return atomic_cond_wait_with_timeout(&c.impl.cond, &m.impl.mutex, duration)
+}
+
+_cond_signal :: proc(c: ^Cond) {
+	atomic_cond_signal(&c.impl.cond)
+}
+
+_cond_broadcast :: proc(c: ^Cond) {
+	atomic_cond_broadcast(&c.impl.cond)
+}
diff --git a/core/sync/sync2/primitives_windows.odin b/core/sync/primitives_windows.odin
similarity index 99%
rename from core/sync/sync2/primitives_windows.odin
rename to core/sync/primitives_windows.odin
index ca7a5c9ee..055167892 100644
--- a/core/sync/sync2/primitives_windows.odin
+++ b/core/sync/primitives_windows.odin
@@ -1,6 +1,6 @@
 //+build windows
 //+private
-package sync2
+package sync
 
 import "core:time"
 import win32 "core:sys/windows"
diff --git a/core/sync/sync2/sema_internal.odin b/core/sync/sema_internal.odin
similarity index 99%
rename from core/sync/sync2/sema_internal.odin
rename to core/sync/sema_internal.odin
index f4027e908..8cf157708 100644
--- a/core/sync/sync2/sema_internal.odin
+++ b/core/sync/sema_internal.odin
@@ -1,5 +1,5 @@
 //+private
-package sync2
+package sync
 
 import "core:time"
 
diff --git a/core/sync/sync.odin b/core/sync/sync.odin
deleted file mode 100644
index 05c86a868..000000000
--- a/core/sync/sync.odin
+++ /dev/null
@@ -1,123 +0,0 @@
-package sync
-
-import "core:intrinsics"
-
-cpu_relax :: #force_inline proc "contextless" () {
-	intrinsics.cpu_relax()
-}
-
-Condition_Mutex_Ptr :: union{^Mutex, ^Blocking_Mutex}
-
-
-Ticket_Mutex :: struct {
-	ticket:  u64,
-	serving: u64,
-}
-
-ticket_mutex_init :: proc(m: ^Ticket_Mutex) {
-	atomic_store(&m.ticket,  0, .Relaxed)
-	atomic_store(&m.serving, 0, .Relaxed)
-}
-
-ticket_mutex_lock :: #force_inline proc(m: ^Ticket_Mutex) {
-	ticket := atomic_add(&m.ticket, 1, .Relaxed)
-	for ticket != atomic_load(&m.serving, .Acquire) {
-		intrinsics.cpu_relax()
-	}
-}
-
-ticket_mutex_unlock :: #force_inline proc(m: ^Ticket_Mutex) {
-	atomic_add(&m.serving, 1, .Relaxed)
-}
-
-
-Benaphore :: struct {
-	counter: int,
-	sema: Semaphore,
-}
-
-benaphore_init :: proc(b: ^Benaphore) {
-	intrinsics.atomic_store(&b.counter, 0)
-	semaphore_init(&b.sema)
-}
-
-benaphore_destroy :: proc(b: ^Benaphore) {
-	semaphore_destroy(&b.sema)
-}
-
-benaphore_lock :: proc(b: ^Benaphore) {
-	if intrinsics.atomic_add_acq(&b.counter, 1) > 1 {
-		semaphore_wait_for(&b.sema)
-	}
-}
-
-benaphore_try_lock :: proc(b: ^Benaphore) -> bool {
-	v, _ := intrinsics.atomic_cxchg_acq(&b.counter, 1, 0)
-	return v == 0
-}
-
-benaphore_unlock :: proc(b: ^Benaphore) {
-	if intrinsics.atomic_sub_rel(&b.counter, 1) > 0 {
-		semaphore_post(&b.sema)
-	}
-}
-
-Recursive_Benaphore :: struct {
-	counter:   int,
-	owner:     int,
-	recursion: int,
-	sema: Semaphore,
-}
-
-recursive_benaphore_init :: proc(b: ^Recursive_Benaphore) {
-	intrinsics.atomic_store(&b.counter, 0)
-	semaphore_init(&b.sema)
-}
-
-recursive_benaphore_destroy :: proc(b: ^Recursive_Benaphore) {
-	semaphore_destroy(&b.sema)
-}
-
-recursive_benaphore_lock :: proc(b: ^Recursive_Benaphore) {
-	tid := current_thread_id()
-	if intrinsics.atomic_add_acq(&b.counter, 1) > 1 {
-		if tid != b.owner {
-			semaphore_wait_for(&b.sema)
-		}
-	}
-	// inside the lock
-	b.owner = tid
-	b.recursion += 1
-}
-
-recursive_benaphore_try_lock :: proc(b: ^Recursive_Benaphore) -> bool {
-	tid := current_thread_id()
-	if b.owner == tid {
-		intrinsics.atomic_add_acq(&b.counter, 1)
-	} else {
-		v, _ := intrinsics.atomic_cxchg_acq(&b.counter, 1, 0)
-		if v != 0 {
-			return false
-		}
-		// inside the lock
-		b.owner = tid
-	}
-	b.recursion += 1
-	return true
-}
-
-recursive_benaphore_unlock :: proc(b: ^Recursive_Benaphore) {
-	tid := current_thread_id()
-	assert(tid == b.owner)
-	b.recursion -= 1
-	recursion := b.recursion
-	if recursion == 0 {
-		b.owner = 0
-	}
-	if intrinsics.atomic_sub_rel(&b.counter, 1) > 0 {
-		if recursion == 0 {
-			semaphore_post(&b.sema)
-		}
-	}
-	// outside the lock
-}
diff --git a/core/sync/sync2/atomic.odin b/core/sync/sync2/atomic.odin
deleted file mode 100644
index fe19f17c8..000000000
--- a/core/sync/sync2/atomic.odin
+++ /dev/null
@@ -1,79 +0,0 @@
-package sync2
-
-import "core:intrinsics"
-
-cpu_relax :: intrinsics.cpu_relax
-
-atomic_fence         :: intrinsics.atomic_fence
-atomic_fence_acquire :: intrinsics.atomic_fence_acq
-atomic_fence_release :: intrinsics.atomic_fence_rel
-atomic_fence_acqrel  :: intrinsics.atomic_fence_acqrel
-
-atomic_store           :: intrinsics.atomic_store
-atomic_store_release   :: intrinsics.atomic_store_rel
-atomic_store_relaxed   :: intrinsics.atomic_store_relaxed
-atomic_store_unordered :: intrinsics.atomic_store_unordered
-
-atomic_load           :: intrinsics.atomic_load
-atomic_load_acquire   :: intrinsics.atomic_load_acq
-atomic_load_relaxed   :: intrinsics.atomic_load_relaxed
-atomic_load_unordered :: intrinsics.atomic_load_unordered
-
-atomic_add          :: intrinsics.atomic_add
-atomic_add_acquire  :: intrinsics.atomic_add_acq
-atomic_add_release  :: intrinsics.atomic_add_rel
-atomic_add_acqrel   :: intrinsics.atomic_add_acqrel
-atomic_add_relaxed  :: intrinsics.atomic_add_relaxed
-atomic_sub          :: intrinsics.atomic_sub
-atomic_sub_acquire  :: intrinsics.atomic_sub_acq
-atomic_sub_release  :: intrinsics.atomic_sub_rel
-atomic_sub_acqrel   :: intrinsics.atomic_sub_acqrel
-atomic_sub_relaxed  :: intrinsics.atomic_sub_relaxed
-atomic_and          :: intrinsics.atomic_and
-atomic_and_acquire  :: intrinsics.atomic_and_acq
-atomic_and_release  :: intrinsics.atomic_and_rel
-atomic_and_acqrel   :: intrinsics.atomic_and_acqrel
-atomic_and_relaxed  :: intrinsics.atomic_and_relaxed
-atomic_nand         :: intrinsics.atomic_nand
-atomic_nand_acquire :: intrinsics.atomic_nand_acq
-atomic_nand_release :: intrinsics.atomic_nand_rel
-atomic_nand_acqrel  :: intrinsics.atomic_nand_acqrel
-atomic_nand_relaxed :: intrinsics.atomic_nand_relaxed
-atomic_or           :: intrinsics.atomic_or
-atomic_or_acquire   :: intrinsics.atomic_or_acq
-atomic_or_release   :: intrinsics.atomic_or_rel
-atomic_or_acqrel    :: intrinsics.atomic_or_acqrel
-atomic_or_relaxed   :: intrinsics.atomic_or_relaxed
-atomic_xor          :: intrinsics.atomic_xor
-atomic_xor_acquire  :: intrinsics.atomic_xor_acq
-atomic_xor_release  :: intrinsics.atomic_xor_rel
-atomic_xor_acqrel   :: intrinsics.atomic_xor_acqrel
-atomic_xor_relaxed  :: intrinsics.atomic_xor_relaxed
-
-atomic_exchange         :: intrinsics.atomic_xchg
-atomic_exchange_acquire :: intrinsics.atomic_xchg_acq
-atomic_exchange_release :: intrinsics.atomic_xchg_rel
-atomic_exchange_acqrel  :: intrinsics.atomic_xchg_acqrel
-atomic_exchange_relaxed :: intrinsics.atomic_xchg_relaxed
-
-// Returns value and optional ok boolean
-atomic_compare_exchange_strong                     :: intrinsics.atomic_cxchg
-atomic_compare_exchange_strong_acquire             :: intrinsics.atomic_cxchg_acq
-atomic_compare_exchange_strong_release             :: intrinsics.atomic_cxchg_rel
-atomic_compare_exchange_strong_acqrel              :: intrinsics.atomic_cxchg_acqrel
-atomic_compare_exchange_strong_relaxed             :: intrinsics.atomic_cxchg_relaxed
-atomic_compare_exchange_strong_failrelaxed         :: intrinsics.atomic_cxchg_failrelaxed
-atomic_compare_exchange_strong_failacquire         :: intrinsics.atomic_cxchg_failacq
-atomic_compare_exchange_strong_acquire_failrelaxed :: intrinsics.atomic_cxchg_acq_failrelaxed
-atomic_compare_exchange_strong_acqrel_failrelaxed  :: intrinsics.atomic_cxchg_acqrel_failrelaxed
-
-// Returns value and optional ok boolean
-atomic_compare_exchange_weak                     :: intrinsics.atomic_cxchgweak
-atomic_compare_exchange_weak_acquire             :: intrinsics.atomic_cxchgweak_acq
-atomic_compare_exchange_weak_release             :: intrinsics.atomic_cxchgweak_rel
-atomic_compare_exchange_weak_acqrel              :: intrinsics.atomic_cxchgweak_acqrel
-atomic_compare_exchange_weak_relaxed             :: intrinsics.atomic_cxchgweak_relaxed
-atomic_compare_exchange_weak_failrelaxed         :: intrinsics.atomic_cxchgweak_failrelaxed
-atomic_compare_exchange_weak_failacquire         :: intrinsics.atomic_cxchgweak_failacq
-atomic_compare_exchange_weak_acquire_failrelaxed :: intrinsics.atomic_cxchgweak_acq_failrelaxed
-atomic_compare_exchange_weak_acqrel_failrelaxed  :: intrinsics.atomic_cxchgweak_acqrel_failrelaxed
diff --git a/core/sync/sync2/primitives_freebsd.odin b/core/sync/sync2/primitives_freebsd.odin
deleted file mode 100644
index 2a25a18a4..000000000
--- a/core/sync/sync2/primitives_freebsd.odin
+++ /dev/null
@@ -1,9 +0,0 @@
-//+build freebsd
-//+private
-package sync2
-
-import "core:os"
-
-_current_thread_id :: proc "contextless" () -> int {
-	return os.current_thread_id()
-}
diff --git a/core/sync/sync2/primitives_internal.odin b/core/sync/sync2/primitives_internal.odin
deleted file mode 100644
index eb692c6ae..000000000
--- a/core/sync/sync2/primitives_internal.odin
+++ /dev/null
@@ -1,184 +0,0 @@
-//+private
-package sync2
-
-when #config(ODIN_SYNC_RECURSIVE_MUTEX_USE_FUTEX, true) {
-	_Recursive_Mutex :: struct {
-		owner:     Futex,
-		recursion: i32,
-	}
-
-	_recursive_mutex_lock :: proc(m: ^Recursive_Mutex) {
-		tid := Futex(current_thread_id())
-		for {
-			prev_owner := atomic_compare_exchange_strong_acquire(&m.impl.owner, tid, 0)
-			switch prev_owner {
-			case 0, tid:
-				m.impl.recursion += 1
-				// inside the lock
-				return
-			}
-
-			futex_wait(&m.impl.owner, u32(prev_owner))
-		}
-	}
-
-	_recursive_mutex_unlock :: proc(m: ^Recursive_Mutex) {
-		m.impl.recursion -= 1
-		if m.impl.recursion != 0 {
-			return
-		}
-		atomic_exchange_release(&m.impl.owner, 0)
-		
-		futex_signal(&m.impl.owner)
-		// outside the lock
-
-	}
-
-	_recursive_mutex_try_lock :: proc(m: ^Recursive_Mutex) -> bool {
-		tid := Futex(current_thread_id())
-		prev_owner := atomic_compare_exchange_strong_acquire(&m.impl.owner, tid, 0)
-		switch prev_owner {
-		case 0, tid:
-			m.impl.recursion += 1
-			// inside the lock
-			return true
-		}
-		return false
-	}
-} else {
-	_Recursive_Mutex :: struct {
-		owner:     int,
-		recursion: int,
-		mutex:     Mutex,
-	}
-
-	_recursive_mutex_lock :: proc(m: ^Recursive_Mutex) {
-		tid := current_thread_id()
-		if tid != m.impl.owner {
-			mutex_lock(&m.impl.mutex)
-		}
-		// inside the lock
-		m.impl.owner = tid
-		m.impl.recursion += 1
-	}
-
-	_recursive_mutex_unlock :: proc(m: ^Recursive_Mutex) {
-		tid := current_thread_id()
-		assert(tid == m.impl.owner)
-		m.impl.recursion -= 1
-		recursion := m.impl.recursion
-		if recursion == 0 {
-			m.impl.owner = 0
-		}
-		if recursion == 0 {
-			mutex_unlock(&m.impl.mutex)
-		}
-		// outside the lock
-
-	}
-
-	_recursive_mutex_try_lock :: proc(m: ^Recursive_Mutex) -> bool {
-		tid := current_thread_id()
-		if m.impl.owner == tid {
-			return mutex_try_lock(&m.impl.mutex)
-		}
-		if !mutex_try_lock(&m.impl.mutex) {
-			return false
-		}
-		// inside the lock
-		m.impl.owner = tid
-		m.impl.recursion += 1
-		return true
-	}
-}
-
-
-when ODIN_OS != .Windows {
-	RW_Mutex_State :: distinct uint
-	RW_Mutex_State_Half_Width :: size_of(RW_Mutex_State)*8/2
-	RW_Mutex_State_Is_Writing :: RW_Mutex_State(1)
-	RW_Mutex_State_Writer     :: RW_Mutex_State(1)<<1
-	RW_Mutex_State_Reader     :: RW_Mutex_State(1)<<RW_Mutex_State_Half_Width
-
-	RW_Mutex_State_Writer_Mask :: RW_Mutex_State(1<<(RW_Mutex_State_Half_Width-1) - 1) << 1
-	RW_Mutex_State_Reader_Mask :: RW_Mutex_State(1<<(RW_Mutex_State_Half_Width-1) - 1) << RW_Mutex_State_Half_Width
-
-
-	_RW_Mutex :: struct {
-		// NOTE(bill): pthread_rwlock_t cannot be used since pthread_rwlock_destroy is required on some platforms
-		// TODO(bill): Can we determine which platforms exactly?
-		state: RW_Mutex_State,
-		mutex: Mutex,
-		sema:  Sema,
-	}
-
-	_rw_mutex_lock :: proc(rw: ^RW_Mutex) {
-		_ = atomic_add(&rw.impl.state, RW_Mutex_State_Writer)
-		mutex_lock(&rw.impl.mutex)
-
-		state := atomic_or(&rw.impl.state, RW_Mutex_State_Writer)
-		if state & RW_Mutex_State_Reader_Mask != 0 {
-			sema_wait(&rw.impl.sema)
-		}
-	}
-
-	_rw_mutex_unlock :: proc(rw: ^RW_Mutex) {
-		_ = atomic_and(&rw.impl.state, ~RW_Mutex_State_Is_Writing)
-		mutex_unlock(&rw.impl.mutex)
-	}
-
-	_rw_mutex_try_lock :: proc(rw: ^RW_Mutex) -> bool {
-		if mutex_try_lock(&rw.impl.mutex) {
-			state := atomic_load(&rw.impl.state)
-			if state & RW_Mutex_State_Reader_Mask == 0 {
-				_ = atomic_or(&rw.impl.state, RW_Mutex_State_Is_Writing)
-				return true
-			}
-
-			mutex_unlock(&rw.impl.mutex)
-		}
-		return false
-	}
-
-	_rw_mutex_shared_lock :: proc(rw: ^RW_Mutex) {
-		state := atomic_load(&rw.impl.state)
-		for state & (RW_Mutex_State_Is_Writing|RW_Mutex_State_Writer_Mask) == 0 {
-			ok: bool
-			state, ok = atomic_compare_exchange_weak(&rw.impl.state, state, state + RW_Mutex_State_Reader)
-			if ok {
-				return
-			}
-		}
-
-		mutex_lock(&rw.impl.mutex)
-		_ = atomic_add(&rw.impl.state, RW_Mutex_State_Reader)
-		mutex_unlock(&rw.impl.mutex)
-	}
-
-	_rw_mutex_shared_unlock :: proc(rw: ^RW_Mutex) {
-		state := atomic_sub(&rw.impl.state, RW_Mutex_State_Reader)
-
-		if (state & RW_Mutex_State_Reader_Mask == RW_Mutex_State_Reader) &&
-		   (state & RW_Mutex_State_Is_Writing != 0) {
-		   	sema_post(&rw.impl.sema)
-		}
-	}
-
-	_rw_mutex_try_shared_lock :: proc(rw: ^RW_Mutex) -> bool {
-		state := atomic_load(&rw.impl.state)
-		if state & (RW_Mutex_State_Is_Writing|RW_Mutex_State_Writer_Mask) == 0 {
-			_, ok := atomic_compare_exchange_strong(&rw.impl.state, state, state + RW_Mutex_State_Reader)
-			if ok {
-				return true
-			}
-		}
-		if mutex_try_lock(&rw.impl.mutex) {
-			_ = atomic_add(&rw.impl.state, RW_Mutex_State_Reader)
-			mutex_unlock(&rw.impl.mutex)
-			return true
-		}
-
-		return false
-	}
-
-}
\ No newline at end of file
diff --git a/core/sync/sync2/primitives_linux.odin b/core/sync/sync2/primitives_linux.odin
deleted file mode 100644
index 89ed97985..000000000
--- a/core/sync/sync2/primitives_linux.odin
+++ /dev/null
@@ -1,9 +0,0 @@
-//+build linux
-//+private
-package sync2
-
-import "core:sys/unix"
-
-_current_thread_id :: proc "contextless" () -> int {
-	return unix.sys_gettid()
-}
diff --git a/core/sync/sync2/primitives_openbsd.odin b/core/sync/sync2/primitives_openbsd.odin
deleted file mode 100644
index ef122b02e..000000000
--- a/core/sync/sync2/primitives_openbsd.odin
+++ /dev/null
@@ -1,9 +0,0 @@
-//+build openbsd
-//+private
-package sync2
-
-import "core:os"
-
-_current_thread_id :: proc "contextless" () -> int {
-	return os.current_thread_id()
-}
diff --git a/core/sync/sync2/primitives_pthreads.odin b/core/sync/sync2/primitives_pthreads.odin
deleted file mode 100644
index 28053f9cc..000000000
--- a/core/sync/sync2/primitives_pthreads.odin
+++ /dev/null
@@ -1,58 +0,0 @@
-//+build linux, freebsd, openbsd
-//+private
-package sync2
-
-import "core:time"
-import "core:sys/unix"
-
-_Mutex_State :: enum i32 {
-	Unlocked = 0,
-	Locked   = 1,
-	Waiting  = 2,
-}
-_Mutex :: struct {
-	pthread_mutex: unix.pthread_mutex_t,
-}
-
-_mutex_lock :: proc(m: ^Mutex) {
-	err := unix.pthread_mutex_lock(&m.impl.pthread_mutex)
-	assert(err == 0)
-}
-
-_mutex_unlock :: proc(m: ^Mutex) {
-	err := unix.pthread_mutex_unlock(&m.impl.pthread_mutex)
-	assert(err == 0)
-}
-
-_mutex_try_lock :: proc(m: ^Mutex) -> bool {
-	err := unix.pthread_mutex_trylock(&m.impl.pthread_mutex)
-	return err == 0
-}
-
-_Cond :: struct {
-	pthread_cond: unix.pthread_cond_t,
-}
-
-_cond_wait :: proc(c: ^Cond, m: ^Mutex) {
-	err := unix.pthread_cond_wait(&c.impl.pthread_cond, &m.impl.pthread_mutex)
-	assert(err == 0)
-}
-
-
-_cond_wait_with_timeout :: proc(c: ^Cond, m: ^Mutex, duration: time.Duration) -> bool {
-	tv_sec  := i64(duration/1e9)
-	tv_nsec := i64(duration%1e9)
-	err := unix.pthread_cond_timedwait(&c.impl.pthread_cond, &m.impl.pthread_mutex, &{tv_sec, tv_nsec})
-	return err == 0
-}
-
-
-_cond_signal :: proc(c: ^Cond) {
-	err := unix.pthread_cond_signal(&c.impl.pthread_cond)
-	assert(err == 0)
-}
-
-_cond_broadcast :: proc(c: ^Cond) {
-	err := unix.pthread_cond_broadcast(&c.impl.pthread_cond)
-	assert(err == 0)
-}
diff --git a/core/sync/sync_darwin.odin b/core/sync/sync_darwin.odin
deleted file mode 100644
index f3bb4d5a3..000000000
--- a/core/sync/sync_darwin.odin
+++ /dev/null
@@ -1,54 +0,0 @@
-package sync
-
-import "core:sys/darwin"
-
-import "core:c"
-
-foreign import pthread "System.framework"
-
-current_thread_id :: proc "contextless" () -> int {
-	tid: u64
-	// NOTE(Oskar): available from OSX 10.6 and iOS 3.2.
-	// For older versions there is `syscall(SYS_thread_selfid)`, but not really
-	// the same thing apparently.
-	foreign pthread { pthread_threadid_np :: proc "c" (rawptr, ^u64) -> c.int --- }
-	pthread_threadid_np(nil, &tid)
-	return int(tid)
-}
-
-
-// The Darwin docs say it best:
-// A semaphore is much like a lock, except that a finite number of threads can hold it simultaneously.
-// Semaphores can be thought of as being much like piles of tokens; multiple threads can take these tokens, 
-// but when there are none left, a thread must wait until another thread returns one.
-Semaphore :: struct #align 16 {
-	handle: darwin.semaphore_t,
-}
-// TODO(tetra): Only marked with alignment because we cannot mark distinct integers with alignments.
-// See core/sys/unix/pthread_linux.odin/pthread_t.
-
-semaphore_init :: proc(s: ^Semaphore, initial_count := 0) {
-	ct := darwin.mach_task_self()
-	res := darwin.semaphore_create(ct, &s.handle, 0, c.int(initial_count))
-	assert(res == 0)
-}
-
-semaphore_destroy :: proc(s: ^Semaphore) {
-	ct := darwin.mach_task_self()
-	res := darwin.semaphore_destroy(ct, s.handle)
-	assert(res == 0)
-	s.handle = {}
-}
-
-semaphore_post :: proc(s: ^Semaphore, count := 1) {
-	// NOTE: SPEED: If there's one syscall to do this, we should use it instead of the loop.
-	for in 0..<count {
-		res := darwin.semaphore_signal(s.handle)
-		assert(res == 0)
-	}
-}
-
-semaphore_wait_for :: proc(s: ^Semaphore) {
-	res := darwin.semaphore_wait(s.handle)
-	assert(res == 0)
-}
diff --git a/core/sync/sync_freebsd.odin b/core/sync/sync_freebsd.odin
deleted file mode 100644
index 20e6bfceb..000000000
--- a/core/sync/sync_freebsd.odin
+++ /dev/null
@@ -1,40 +0,0 @@
-package sync
-
-import "core:sys/unix"
-import "core:intrinsics"
-
-
-current_thread_id :: proc "contextless" () -> int {
-	SYS_GETTID :: 186
-	return int(intrinsics.syscall(SYS_GETTID))
-}
-
-
-// The Darwin docs say it best:
-// A semaphore is much like a lock, except that a finite number of threads can hold it simultaneously.
-// Semaphores can be thought of as being much like piles of tokens; multiple threads can take these tokens, 
-// but when there are none left, a thread must wait until another thread returns one.
-Semaphore :: struct #align 16 {
-	handle: unix.sem_t,
-}
-
-semaphore_init :: proc(s: ^Semaphore, initial_count := 0) {
-	assert(unix.sem_init(&s.handle, 0, u32(initial_count)) == 0)
-}
-
-semaphore_destroy :: proc(s: ^Semaphore) {
-	assert(unix.sem_destroy(&s.handle) == 0)
-	s.handle = {}
-}
-
-semaphore_post :: proc(s: ^Semaphore, count := 1) {
-	// NOTE: SPEED: If there's one syscall to do this, we should use it instead of the loop.
-	for in 0..<count {
-		assert(unix.sem_post(&s.handle) == 0)
-	}
-}
-
-semaphore_wait_for :: proc(s: ^Semaphore) {
-	assert(unix.sem_wait(&s.handle) == 0)
-}
-
diff --git a/core/sync/sync_linux.odin b/core/sync/sync_linux.odin
deleted file mode 100644
index 340437c11..000000000
--- a/core/sync/sync_linux.odin
+++ /dev/null
@@ -1,36 +0,0 @@
-package sync
-
-import "core:sys/unix"
-
-current_thread_id :: proc "contextless" () -> int {
-	return unix.sys_gettid()
-}
-
-
-// The Darwin docs say it best:
-// A semaphore is much like a lock, except that a finite number of threads can hold it simultaneously.
-// Semaphores can be thought of as being much like piles of tokens; multiple threads can take these tokens, 
-// but when there are none left, a thread must wait until another thread returns one.
-Semaphore :: struct #align 16 {
-	handle: unix.sem_t,
-}
-
-semaphore_init :: proc(s: ^Semaphore, initial_count := 0) {
-	assert(unix.sem_init(&s.handle, 0, u32(initial_count)) == 0)
-}
-
-semaphore_destroy :: proc(s: ^Semaphore) {
-	assert(unix.sem_destroy(&s.handle) == 0)
-	s.handle = {}
-}
-
-semaphore_post :: proc(s: ^Semaphore, count := 1) {
-    // NOTE: SPEED: If there's one syscall to do this, we should use it instead of the loop.
-    for in 0..<count {
-	    assert(unix.sem_post(&s.handle) == 0)
-    }
-}
-
-semaphore_wait_for :: proc(s: ^Semaphore) {
-	assert(unix.sem_wait(&s.handle) == 0)
-}
diff --git a/core/sync/sync_openbsd.odin b/core/sync/sync_openbsd.odin
deleted file mode 100644
index 926655f5b..000000000
--- a/core/sync/sync_openbsd.odin
+++ /dev/null
@@ -1,36 +0,0 @@
-package sync
-
-import "core:sys/unix"
-import "core:os"
-
-current_thread_id :: proc "contextless" () -> int {
-	return os.current_thread_id()
-}
-
-// The Darwin docs say it best:
-// A semaphore is much like a lock, except that a finite number of threads can hold it simultaneously.
-// Semaphores can be thought of as being much like piles of tokens; multiple threads can take these tokens, 
-// but when there are none left, a thread must wait until another thread returns one.
-Semaphore :: struct #align 16 {
-	handle: unix.sem_t,
-}
-
-semaphore_init :: proc(s: ^Semaphore, initial_count := 0) {
-	assert(unix.sem_init(&s.handle, 0, u32(initial_count)) == 0)
-}
-
-semaphore_destroy :: proc(s: ^Semaphore) {
-	assert(unix.sem_destroy(&s.handle) == 0)
-	s.handle = {}
-}
-
-semaphore_post :: proc(s: ^Semaphore, count := 1) {
-    // NOTE: SPEED: If there's one syscall to do this, we should use it instead of the loop.
-    for in 0..<count {
-	    assert(unix.sem_post(&s.handle) == 0)
-    }
-}
-
-semaphore_wait_for :: proc(s: ^Semaphore) {
-	assert(unix.sem_wait(&s.handle) == 0)
-}
diff --git a/core/sync/sync_unix.odin b/core/sync/sync_unix.odin
deleted file mode 100644
index b8e9d32cc..000000000
--- a/core/sync/sync_unix.odin
+++ /dev/null
@@ -1,248 +0,0 @@
-// +build linux, darwin, freebsd, openbsd
-package sync
-
-import "core:sys/unix"
-import "core:time"
-
-// A recursive lock that can only be held by one thread at once
-Mutex :: struct {
-	handle: unix.pthread_mutex_t,
-}
-
-
-mutex_init :: proc(m: ^Mutex) {
-	// NOTE(tetra, 2019-11-01): POSIX OOM if we cannot init the attrs or the mutex.
-	attrs: unix.pthread_mutexattr_t
-	assert(unix.pthread_mutexattr_init(&attrs) == 0)
-	defer unix.pthread_mutexattr_destroy(&attrs) // ignores destruction error
-	unix.pthread_mutexattr_settype(&attrs, unix.PTHREAD_MUTEX_RECURSIVE)
-
-	assert(unix.pthread_mutex_init(&m.handle, &attrs) == 0)
-}
-
-mutex_destroy :: proc(m: ^Mutex) {
-	assert(unix.pthread_mutex_destroy(&m.handle) == 0)
-	m.handle = {}
-}
-
-mutex_lock :: proc(m: ^Mutex) {
-	assert(unix.pthread_mutex_lock(&m.handle) == 0)
-}
-
-// Returns false if someone else holds the lock.
-mutex_try_lock :: proc(m: ^Mutex) -> bool {
-	return unix.pthread_mutex_trylock(&m.handle) == 0
-}
-
-mutex_unlock :: proc(m: ^Mutex) {
-	assert(unix.pthread_mutex_unlock(&m.handle) == 0)
-}
-
-
-Blocking_Mutex :: struct {
-	handle: unix.pthread_mutex_t,
-}
-
-
-blocking_mutex_init :: proc(m: ^Blocking_Mutex) {
-	// NOTE(tetra, 2019-11-01): POSIX OOM if we cannot init the attrs or the mutex.
-	attrs: unix.pthread_mutexattr_t
-	assert(unix.pthread_mutexattr_init(&attrs) == 0)
-	defer unix.pthread_mutexattr_destroy(&attrs) // ignores destruction error
-
-	assert(unix.pthread_mutex_init(&m.handle, &attrs) == 0)
-}
-
-blocking_mutex_destroy :: proc(m: ^Blocking_Mutex) {
-	assert(unix.pthread_mutex_destroy(&m.handle) == 0)
-	m.handle = {}
-}
-
-blocking_mutex_lock :: proc(m: ^Blocking_Mutex) {
-	assert(unix.pthread_mutex_lock(&m.handle) == 0)
-}
-
-// Returns false if someone else holds the lock.
-blocking_mutex_try_lock :: proc(m: ^Blocking_Mutex) -> bool {
-	return unix.pthread_mutex_trylock(&m.handle) == 0
-}
-
-blocking_mutex_unlock :: proc(m: ^Blocking_Mutex) {
-	assert(unix.pthread_mutex_unlock(&m.handle) == 0)
-}
-
-
-
-// Blocks until signalled, and then lets past exactly
-// one thread.
-Condition :: struct {
-	handle: unix.pthread_cond_t,
-	mutex:  Condition_Mutex_Ptr,
-
-	// NOTE(tetra, 2019-11-11): Used to mimic the more sane behavior of Windows' AutoResetEvent.
-	// This means that you may signal the condition before anyone is waiting to cause the
-	// next thread that tries to wait to just pass by uninterrupted, without sleeping.
-	// Without this, signalling a condition will only wake up a thread which is already waiting,
-	// but not one that is about to wait, which can cause your program to become out of sync in
-	// ways that are hard to debug or fix.
-	flag: bool, // atomically mutated
-}
-
-condition_init :: proc(c: ^Condition, mutex: Condition_Mutex_Ptr) -> bool {
-	// NOTE(tetra, 2019-11-01): POSIX OOM if we cannot init the attrs or the condition.
-	attrs: unix.pthread_condattr_t
-	if unix.pthread_condattr_init(&attrs) != 0 {
-		return false
-	}
-	defer unix.pthread_condattr_destroy(&attrs) // ignores destruction error
-
-	c.flag = false
-	c.mutex = mutex
-	return unix.pthread_cond_init(&c.handle, &attrs) == 0
-}
-
-condition_destroy :: proc(c: ^Condition) {
-	assert(unix.pthread_cond_destroy(&c.handle) == 0)
-	c.handle = {}
-}
-
-// Awaken exactly one thread who is waiting on the condition
-condition_signal :: proc(c: ^Condition) -> bool {
-	switch m in c.mutex {
-	case ^Mutex:
-		mutex_lock(m)
-		defer mutex_unlock(m)
-		atomic_swap(&c.flag, true, .Sequentially_Consistent)
-		return unix.pthread_cond_signal(&c.handle) == 0
-	case ^Blocking_Mutex:
-		blocking_mutex_lock(m)
-		defer blocking_mutex_unlock(m)
-		atomic_swap(&c.flag, true, .Sequentially_Consistent)
-		return unix.pthread_cond_signal(&c.handle) == 0
-	}
-	return false
-}
-
-// Awaken all threads who are waiting on the condition
-condition_broadcast :: proc(c: ^Condition) -> bool {
-	return unix.pthread_cond_broadcast(&c.handle) == 0
-}
-
-// Wait for the condition to be signalled.
-// Does not block if the condition has been signalled and no one
-// has waited on it yet.
-condition_wait_for :: proc(c: ^Condition) -> bool {
-	switch m in c.mutex {
-	case ^Mutex:
-		mutex_lock(m)
-		defer mutex_unlock(m)
-		// NOTE(tetra): If a thread comes by and steals the flag immediately after the signal occurs,
-		// the thread that gets signalled and wakes up, discovers that the flag was taken and goes
-		// back to sleep.
-		// Though this overall behavior is the most sane, there may be a better way to do this that means that
-		// the first thread to wait, gets the flag first.
-		if atomic_swap(&c.flag, false, .Sequentially_Consistent) {
-			return true
-		}
-		for {
-			if unix.pthread_cond_wait(&c.handle, &m.handle) != 0 {
-				return false
-			}
-			if atomic_swap(&c.flag, false, .Sequentially_Consistent) {
-				return true
-			}
-		}
-		return false
-
-	case ^Blocking_Mutex:
-		blocking_mutex_lock(m)
-		defer blocking_mutex_unlock(m)
-		// NOTE(tetra): If a thread comes by and steals the flag immediately after the signal occurs,
-		// the thread that gets signalled and wakes up, discovers that the flag was taken and goes
-		// back to sleep.
-		// Though this overall behavior is the most sane, there may be a better way to do this that means that
-		// the first thread to wait, gets the flag first.
-		if atomic_swap(&c.flag, false, .Sequentially_Consistent) {
-			return true
-		}
-		for {
-			if unix.pthread_cond_wait(&c.handle, &m.handle) != 0 {
-				return false
-			}
-			if atomic_swap(&c.flag, false, .Sequentially_Consistent) {
-				return true
-			}
-		}
-		return false
-	}
-	return false
-}
-
-// Wait for the condition to be signalled.
-// Does not block if the condition has been signalled and no one
-// has waited on it yet.
-condition_wait_for_timeout :: proc(c: ^Condition, duration: time.Duration) -> bool {
-	switch m in c.mutex {
-	case ^Mutex:
-		mutex_lock(m)
-		defer mutex_unlock(m)
-		// NOTE(tetra): If a thread comes by and steals the flag immediately after the signal occurs,
-		// the thread that gets signalled and wakes up, discovers that the flag was taken and goes
-		// back to sleep.
-		// Though this overall behavior is the most sane, there may be a better way to do this that means that
-		// the first thread to wait, gets the flag first.
-		if atomic_swap(&c.flag, false, .Sequentially_Consistent) {
-			return true
-		}
-
-		ns := time.duration_nanoseconds(duration)
-		timeout: time.TimeSpec
-		timeout.tv_sec  = ns / 1e9
-		timeout.tv_nsec = ns % 1e9
-
-		for {
-			if unix.pthread_cond_timedwait(&c.handle, &m.handle, &timeout) != 0 {
-				return false
-			}
-			if atomic_swap(&c.flag, false, .Sequentially_Consistent) {
-				return true
-			}
-		}
-		return false
-
-	case ^Blocking_Mutex:
-		blocking_mutex_lock(m)
-		defer blocking_mutex_unlock(m)
-		// NOTE(tetra): If a thread comes by and steals the flag immediately after the signal occurs,
-		// the thread that gets signalled and wakes up, discovers that the flag was taken and goes
-		// back to sleep.
-		// Though this overall behavior is the most sane, there may be a better way to do this that means that
-		// the first thread to wait, gets the flag first.
-		if atomic_swap(&c.flag, false, .Sequentially_Consistent) {
-			return true
-		}
-
-		ns := time.duration_nanoseconds(duration)
-
-		timeout: time.TimeSpec
-		timeout.tv_sec  = ns / 1e9
-		timeout.tv_nsec = ns % 1e9
-
-		for {
-			if unix.pthread_cond_timedwait(&c.handle, &m.handle, &timeout) != 0 {
-				return false
-			}
-			if atomic_swap(&c.flag, false, .Sequentially_Consistent) {
-				return true
-			}
-		}
-		return false
-	}
-	return false
-}
-
-
-
-thread_yield :: proc() {
-	unix.sched_yield()
-}
diff --git a/core/sync/sync2/sync_util.odin b/core/sync/sync_util.odin
similarity index 99%
rename from core/sync/sync2/sync_util.odin
rename to core/sync/sync_util.odin
index 013bf511a..4add9064d 100644
--- a/core/sync/sync2/sync_util.odin
+++ b/core/sync/sync_util.odin
@@ -1,4 +1,4 @@
-package sync2
+package sync
 
 /*
 Example:
diff --git a/core/sync/sync_windows.odin b/core/sync/sync_windows.odin
deleted file mode 100644
index 0a7cf71b2..000000000
--- a/core/sync/sync_windows.odin
+++ /dev/null
@@ -1,180 +0,0 @@
-// +build windows
-package sync
-
-import win32 "core:sys/windows"
-import "core:time"
-
-current_thread_id :: proc "contextless" () -> int {
-	return int(win32.GetCurrentThreadId())
-}
-
-
-// When waited upon, blocks until the internal count is greater than zero, then subtracts one.
-// Posting to the semaphore increases the count by one, or the provided amount.
-Semaphore :: struct {
-	_handle: win32.HANDLE,
-}
-
-semaphore_init :: proc(s: ^Semaphore, initial_count := 0) {
-	s._handle = win32.CreateSemaphoreW(nil, i32(initial_count), 1<<31-1, nil)
-}
-
-semaphore_destroy :: proc(s: ^Semaphore) {
-	win32.CloseHandle(s._handle)
-}
-
-semaphore_post :: proc(s: ^Semaphore, count := 1) {
-	win32.ReleaseSemaphore(s._handle, i32(count), nil)
-}
-
-semaphore_wait_for :: proc(s: ^Semaphore) {
-	// NOTE(tetra, 2019-10-30): wait_for_single_object decrements the count before it returns.
-	result := win32.WaitForSingleObject(s._handle, win32.INFINITE)
-	assert(result != win32.WAIT_FAILED)
-}
-
-
-Mutex :: struct {
-	_critical_section: win32.CRITICAL_SECTION,
-}
-
-
-mutex_init :: proc(m: ^Mutex, spin_count := 0) {
-	win32.InitializeCriticalSectionAndSpinCount(&m._critical_section, u32(spin_count))
-}
-
-mutex_destroy :: proc(m: ^Mutex) {
-	win32.DeleteCriticalSection(&m._critical_section)
-}
-
-mutex_lock :: proc(m: ^Mutex) {
-	win32.EnterCriticalSection(&m._critical_section)
-}
-
-mutex_try_lock :: proc(m: ^Mutex) -> bool {
-	return bool(win32.TryEnterCriticalSection(&m._critical_section))
-}
-
-mutex_unlock :: proc(m: ^Mutex) {
-	win32.LeaveCriticalSection(&m._critical_section)
-}
-
-Blocking_Mutex :: struct {
-	_handle: win32.SRWLOCK,
-}
-
-
-blocking_mutex_init :: proc(m: ^Blocking_Mutex) {
-	win32.InitializeSRWLock(&m._handle)
-}
-
-blocking_mutex_destroy :: proc(m: ^Blocking_Mutex) {
-	//
-}
-
-blocking_mutex_lock :: proc(m: ^Blocking_Mutex) {
-	win32.AcquireSRWLockExclusive(&m._handle)
-}
-
-blocking_mutex_try_lock :: proc(m: ^Blocking_Mutex) -> bool {
-	return bool(win32.TryAcquireSRWLockExclusive(&m._handle))
-}
-
-blocking_mutex_unlock :: proc(m: ^Blocking_Mutex) {
-	win32.ReleaseSRWLockExclusive(&m._handle)
-}
-
-
-// Blocks until signalled.
-// When signalled, awakens exactly one waiting thread.
-Condition :: struct {
-	_handle: win32.CONDITION_VARIABLE,
-
-	mutex: Condition_Mutex_Ptr,
-}
-
-
-condition_init :: proc(c: ^Condition, mutex: Condition_Mutex_Ptr) -> bool {
-	assert(mutex != nil)
-	win32.InitializeConditionVariable(&c._handle)
-	c.mutex = mutex
-	return true
-}
-
-condition_destroy :: proc(c: ^Condition) {
-	//
-}
-
-condition_signal :: proc(c: ^Condition) -> bool {
-	if c._handle.ptr == nil {
-		return false
-	}
-	win32.WakeConditionVariable(&c._handle)
-	return true
-}
-
-condition_broadcast :: proc(c: ^Condition) -> bool {
-	if c._handle.ptr == nil {
-		return false
-	}
-	win32.WakeAllConditionVariable(&c._handle)
-	return true
-}
-
-condition_wait_for :: proc(c: ^Condition) -> bool {
-	switch m in &c.mutex {
-	case ^Mutex:
-		return cast(bool)win32.SleepConditionVariableCS(&c._handle, &m._critical_section, win32.INFINITE)
-	case ^Blocking_Mutex:
-		return cast(bool)win32.SleepConditionVariableSRW(&c._handle, &m._handle, win32.INFINITE, 0)
-	}
-	return false
-}
-condition_wait_for_timeout :: proc(c: ^Condition, duration: time.Duration) -> bool {
-	ms := win32.DWORD((max(time.duration_nanoseconds(duration), 0) + 999999)/1000000)
-	switch m in &c.mutex {
-	case ^Mutex:
-		return cast(bool)win32.SleepConditionVariableCS(&c._handle, &m._critical_section, ms)
-	case ^Blocking_Mutex:
-		return cast(bool)win32.SleepConditionVariableSRW(&c._handle, &m._handle, ms, 0)
-	}
-	return false
-}
-
-
-
-
-RW_Lock :: struct {
-	_handle: win32.SRWLOCK,
-}
-
-rw_lock_init :: proc(l: ^RW_Lock) {
-	l._handle = win32.SRWLOCK_INIT
-}
-rw_lock_destroy :: proc(l: ^RW_Lock) {
-	//
-}
-rw_lock_read :: proc(l: ^RW_Lock) {
-	win32.AcquireSRWLockShared(&l._handle)
-}
-rw_lock_try_read :: proc(l: ^RW_Lock) -> bool {
-	return bool(win32.TryAcquireSRWLockShared(&l._handle))
-}
-rw_lock_write :: proc(l: ^RW_Lock) {
-	win32.AcquireSRWLockExclusive(&l._handle)
-}
-rw_lock_try_write :: proc(l: ^RW_Lock) -> bool {
-	return bool(win32.TryAcquireSRWLockExclusive(&l._handle))
-}
-rw_lock_read_unlock :: proc(l: ^RW_Lock) {
-	win32.ReleaseSRWLockShared(&l._handle)
-}
-rw_lock_write_unlock :: proc(l: ^RW_Lock) {
-	win32.ReleaseSRWLockExclusive(&l._handle)
-}
-
-
-thread_yield :: proc() {
-	win32.SwitchToThread()
-}
-
diff --git a/core/sync/wait_group.odin b/core/sync/wait_group.odin
deleted file mode 100644
index 63d882ed1..000000000
--- a/core/sync/wait_group.odin
+++ /dev/null
@@ -1,58 +0,0 @@
-package sync
-
-import "core:intrinsics"
-
-Wait_Group :: struct {
-	counter: int,
-	mutex:   Blocking_Mutex,
-	cond:    Condition,
-}
-
-wait_group_init :: proc(wg: ^Wait_Group) {
-	wg.counter = 0
-	blocking_mutex_init(&wg.mutex)
-	condition_init(&wg.cond, &wg.mutex)
-}
-
-
-wait_group_destroy :: proc(wg: ^Wait_Group) {
-	condition_destroy(&wg.cond)
-	blocking_mutex_destroy(&wg.mutex)
-}
-
-wait_group_add :: proc(wg: ^Wait_Group, delta: int) {
-	if delta == 0 {
-		return
-	}
-
-	blocking_mutex_lock(&wg.mutex)
-	defer blocking_mutex_unlock(&wg.mutex)
-
-	intrinsics.atomic_add(&wg.counter, delta)
-	if wg.counter < 0 {
-		panic("sync.Wait_Group negative counter")
-	}
-	if wg.counter == 0 {
-		condition_broadcast(&wg.cond)
-		if wg.counter != 0 {
-			panic("sync.Wait_Group misuse: sync.wait_group_add called concurrently with sync.wait_group_wait")
-		}
-	}
-}
-
-wait_group_done :: proc(wg: ^Wait_Group) {
-	wait_group_add(wg, -1)
-}
-
-wait_group_wait :: proc(wg: ^Wait_Group) {
-	blocking_mutex_lock(&wg.mutex)
-	defer blocking_mutex_unlock(&wg.mutex)
-
-	if wg.counter != 0 {
-		condition_wait_for(&wg.cond)
-		if wg.counter != 0 {
-			panic("sync.Wait_Group misuse: sync.wait_group_add called concurrently with sync.wait_group_wait")
-		}
-	}
-}
-
diff --git a/core/testing/runner_windows.odin b/core/testing/runner_windows.odin
index 560f23956..e812c410a 100644
--- a/core/testing/runner_windows.odin
+++ b/core/testing/runner_windows.odin
@@ -21,7 +21,7 @@ sema_wait :: proc "contextless" (s: ^Sema) {
 			win32.WaitOnAddress(&s.count, &original_count, size_of(original_count), win32.INFINITE)
 			original_count = s.count
 		}
-		if original_count == intrinsics.atomic_cxchg(&s.count, original_count-1, original_count) {
+		if original_count == intrinsics.atomic_compare_exchange_strong(&s.count, original_count-1, original_count) {
 			return
 		}
 	}
@@ -46,7 +46,7 @@ sema_wait_with_timeout :: proc "contextless" (s: ^Sema, duration: time.Duration)
 			}
 			original_count = s.count
 		}
-		if original_count == intrinsics.atomic_cxchg(&s.count, original_count-1, original_count) {
+		if original_count == intrinsics.atomic_compare_exchange_strong(&s.count, original_count-1, original_count) {
 			return true
 		}
 	}
diff --git a/core/thread/thread_pool.odin b/core/thread/thread_pool.odin
index 37ee4fa98..3f782cf73 100644
--- a/core/thread/thread_pool.odin
+++ b/core/thread/thread_pool.odin
@@ -1,67 +1,75 @@
 package thread
 
+/*
+	thread.Pool
+	Copyright 2022 eisbehr
+	Made available under Odin's BSD-3 license.
+*/
+
 import "core:intrinsics"
 import "core:sync"
 import "core:mem"
 
-Task_Status :: enum i32 {
-	Ready,
-	Busy,
-	Waiting,
-	Term,
-}
-
-Task_Proc :: #type proc(task: ^Task)
+Task_Proc :: #type proc(task: Task)
 
 Task :: struct {
-	procedure: Task_Proc,
-	data: rawptr,
+	procedure:  Task_Proc,
+	data:       rawptr,
 	user_index: int,
+	allocator:  mem.Allocator,
 }
 
-Task_Id :: distinct i32
-INVALID_TASK_ID :: Task_Id(-1)
-
-
+// Do not access the pool's members directly while the pool threads are running,
+// since they use different kinds of locking and mutual exclusion devices.
+// Careless access can and will lead to nasty bugs. Once initialized, the
+// pool's memory address is not allowed to change until it is destroyed.
 Pool :: struct {
-	allocator:             mem.Allocator,
-	mutex:                 sync.Mutex,
-	sem_available:         sync.Semaphore,
-	processing_task_count: int, // atomic
-	is_running:            bool,
+	allocator:     mem.Allocator,
+	mutex:         sync.Mutex,
+	sem_available: sync.Sema,
+
+	// the following values are atomic
+	num_waiting:       int,
+	num_in_processing: int,
+	num_outstanding:   int, // num_waiting + num_in_processing
+	num_done:          int,
+	// end of atomics
+
+	is_running: bool,
 
 	threads: []^Thread,
 
-	tasks: [dynamic]Task,
+	tasks:      [dynamic]Task,
+	tasks_done: [dynamic]Task,
 }
 
-pool_init :: proc(pool: ^Pool, thread_count: int, allocator := context.allocator) {
-	worker_thread_internal :: proc(t: ^Thread) {
-		pool := (^Pool)(t.data)
-
-		for pool.is_running {
-			sync.semaphore_wait_for(&pool.sem_available)
-
-			if task, ok := pool_try_and_pop_task(pool); ok {
-				pool_do_work(pool, &task)
-			}
-		}
-
-		sync.semaphore_post(&pool.sem_available, 1)
-	}
-
-
+// Once initialized, the pool's memory address is not allowed to change until
+// it is destroyed. If thread_count < 1, thread count 1 will be used.
+//
+// The thread pool requires an allocator which it either owns, or which is thread safe.
+pool_init :: proc(pool: ^Pool, thread_count: int, allocator: mem.Allocator) {
 	context.allocator = allocator
 	pool.allocator = allocator
-	pool.tasks = make([dynamic]Task)
-	pool.threads = make([]^Thread, thread_count)
+	pool.tasks      = make([dynamic]Task)
+	pool.tasks_done = make([dynamic]Task)
+	pool.threads    = make([]^Thread, max(thread_count, 1))
 
-	sync.mutex_init(&pool.mutex)
-	sync.semaphore_init(&pool.sem_available)
 	pool.is_running = true
 
 	for _, i in pool.threads {
-		t := create(worker_thread_internal)
+		t := create(proc(t: ^Thread) {
+			pool := (^Pool)(t.data)
+
+			for intrinsics.atomic_load(&pool.is_running) {
+				sync.wait(&pool.sem_available)
+
+				if task, ok := pool_pop_waiting(pool); ok {
+					pool_do_work(pool, task)
+				}
+			}
+
+			sync.post(&pool.sem_available, 1)
+		})
 		t.user_index = i
 		t.data = pool
 		pool.threads[i] = t
@@ -70,15 +78,13 @@ pool_init :: proc(pool: ^Pool, thread_count: int, allocator := context.allocator
 
 pool_destroy :: proc(pool: ^Pool) {
 	delete(pool.tasks)
+	delete(pool.tasks_done)
 
-	for thread in &pool.threads {
-		destroy(thread)
+	for t in &pool.threads {
+		destroy(t)
 	}
 
 	delete(pool.threads, pool.allocator)
-
-	sync.mutex_destroy(&pool.mutex)
-	sync.semaphore_destroy(&pool.sem_available)
 }
 
 pool_start :: proc(pool: ^Pool) {
@@ -87,10 +93,12 @@ pool_start :: proc(pool: ^Pool) {
 	}
 }
 
+// Finish tasks that have already started processing, then shut down all pool
+// threads. Might leave over waiting tasks, any memory allocated for the
+// user data of those tasks will not be freed.
 pool_join :: proc(pool: ^Pool) {
-	pool.is_running = false
-
-	sync.semaphore_post(&pool.sem_available, len(pool.threads))
+	intrinsics.atomic_store(&pool.is_running, false)
+	sync.post(&pool.sem_available, len(pool.threads))
 
 	yield()
 
@@ -99,53 +107,112 @@ pool_join :: proc(pool: ^Pool) {
 	}
 }
 
-pool_add_task :: proc(pool: ^Pool, procedure: Task_Proc, data: rawptr, user_index: int = 0) {
-	sync.mutex_lock(&pool.mutex)
-	defer sync.mutex_unlock(&pool.mutex)
+// Add a task to the thread pool.
+//
+// Tasks can be added from any thread, not just the thread that created
+// the thread pool. You can even add tasks from inside other tasks.
+//
+// Each task also needs an allocator which it either owns, or which is thread
+// safe. By default, allocations in the task are disabled by use of the
+// nil_allocator.
+pool_add_task :: proc(pool: ^Pool, procedure: Task_Proc, data: rawptr, user_index: int = 0, allocator := context.allocator) {
+	sync.guard(&pool.mutex)
 
-	task: Task
-	task.procedure = procedure
-	task.data = data
-	task.user_index = user_index
-
-	append(&pool.tasks, task)
-	sync.semaphore_post(&pool.sem_available, 1)
+	append(&pool.tasks, Task{
+		procedure  = procedure,
+		data       = data,
+		user_index = user_index,
+		allocator  = allocator,
+	})
+	intrinsics.atomic_add(&pool.num_waiting, 1)
+	intrinsics.atomic_add(&pool.num_outstanding, 1)
+	sync.post(&pool.sem_available, 1)
 }
 
-pool_try_and_pop_task :: proc(pool: ^Pool) -> (task: Task, got_task: bool = false) {
-	if sync.mutex_try_lock(&pool.mutex) {
-		if len(pool.tasks) != 0 {
-			intrinsics.atomic_add(&pool.processing_task_count, 1)
-			task = pop_front(&pool.tasks)
-			got_task = true
-		}
-		sync.mutex_unlock(&pool.mutex)
+// Number of tasks waiting to be processed. Only informational, mostly for
+// debugging. Don't rely on this value being consistent with other num_*
+// values.
+pool_num_waiting :: #force_inline proc(pool: ^Pool) -> int {
+	return intrinsics.atomic_load(&pool.num_waiting)
+}
+
+// Number of tasks currently being processed. Only informational, mostly for
+// debugging. Don't rely on this value being consistent with other num_*
+// values.
+pool_num_in_processing :: #force_inline proc(pool: ^Pool) -> int {
+	return intrinsics.atomic_load(&pool.num_in_processing)
+}
+
+// Outstanding tasks are all tasks that are not done, that is, tasks that are
+// waiting, as well as tasks that are currently being processed. Only
+// informational, mostly for debugging. Don't rely on this value being
+// consistent with other num_* values.
+pool_num_outstanding :: #force_inline proc(pool: ^Pool) -> int {
+	return intrinsics.atomic_load(&pool.num_outstanding)
+}
+
+// Number of tasks which are done processing. Only informational, mostly for
+// debugging. Don't rely on this value being consistent with other num_*
+// values.
+pool_num_done :: #force_inline proc(pool: ^Pool) -> int {
+	return intrinsics.atomic_load(&pool.num_done)
+}
+
+// If tasks are only being added from one thread, and this procedure is being
+// called from that same thread, it will reliably tell if the thread pool is
+// empty or not. Empty in this case means there are no tasks waiting, being
+// processed, or _done_.
+pool_is_empty :: #force_inline proc(pool: ^Pool) -> bool {
+	return pool_num_outstanding(pool) == 0 && pool_num_done(pool) == 0
+}
+
+// Mostly for internal use.
+pool_pop_waiting :: proc(pool: ^Pool) -> (task: Task, got_task: bool) {
+	sync.guard(&pool.mutex)
+
+	if len(pool.tasks) != 0 {
+		intrinsics.atomic_sub(&pool.num_waiting, 1)
+		intrinsics.atomic_add(&pool.num_in_processing, 1)
+		task = pop_front(&pool.tasks)
+		got_task = true
 	}
+
 	return
 }
 
+// Use this to take out finished tasks.
+pool_pop_done :: proc(pool: ^Pool) -> (task: Task, got_task: bool) {
+	sync.guard(&pool.mutex)
 
-pool_do_work :: proc(pool: ^Pool, task: ^Task) {
-	task.procedure(task)
-	intrinsics.atomic_sub(&pool.processing_task_count, 1)
-}
-
-
-pool_wait_and_process :: proc(pool: ^Pool) {
-	for len(pool.tasks) != 0 || intrinsics.atomic_load(&pool.processing_task_count) != 0 {
-		if task, ok := pool_try_and_pop_task(pool); ok {
-			pool_do_work(pool, &task)
-		}
-
-		// Safety kick
-		if len(pool.tasks) != 0 && intrinsics.atomic_load(&pool.processing_task_count) == 0 {
-			sync.mutex_lock(&pool.mutex)
-			sync.semaphore_post(&pool.sem_available, len(pool.tasks))
-			sync.mutex_unlock(&pool.mutex)
-		}
-
-		yield()
+	if len(pool.tasks_done) != 0 {
+		task = pop_front(&pool.tasks_done)
+		got_task = true
+		intrinsics.atomic_sub(&pool.num_done, 1)
 	}
 
+	return
+}
+
+// Mostly for internal use.
+pool_do_work :: proc(pool: ^Pool, task: Task) {
+	{
+		context.allocator = task.allocator
+		task.procedure(task)
+	}
+
+	sync.guard(&pool.mutex)
+
+	append(&pool.tasks_done, task)
+	intrinsics.atomic_add(&pool.num_done, 1)
+	intrinsics.atomic_sub(&pool.num_outstanding, 1)
+	intrinsics.atomic_sub(&pool.num_in_processing, 1)
+}
+
+// Process the rest of the tasks, also use this thread for processing, then join
+// all the pool threads.
+pool_finish :: proc(pool: ^Pool) {
+	for task in pool_pop_waiting(pool) {
+		pool_do_work(pool, task)
+	}
 	pool_join(pool)
 }
diff --git a/core/thread/thread_unix.odin b/core/thread/thread_unix.odin
index fbf89f122..8452df112 100644
--- a/core/thread/thread_unix.odin
+++ b/core/thread/thread_unix.odin
@@ -4,7 +4,7 @@ package thread
 
 import "core:runtime"
 import "core:intrinsics"
-import sync "core:sync/sync2"
+import "core:sync"
 import "core:sys/unix"
 
 Thread_State :: enum u8 {
diff --git a/core/thread/thread_windows.odin b/core/thread/thread_windows.odin
index 428e241d0..68382444c 100644
--- a/core/thread/thread_windows.odin
+++ b/core/thread/thread_windows.odin
@@ -3,13 +3,21 @@
 package thread
 
 import "core:runtime"
-import sync "core:sync/sync2"
+import "core:intrinsics"
+import "core:sync"
 import win32 "core:sys/windows"
 
+Thread_State :: enum u8 {
+	Started,
+	Joined,
+	Done,
+}
+
 Thread_Os_Specific :: struct {
 	win32_thread:    win32.HANDLE,
 	win32_thread_id: win32.DWORD,
-	done: bool, // see note in `is_done`
+	mutex:           sync.Mutex,
+	flags:           bit_set[Thread_State; u8],
 }
 
 _thread_priority_map := [Thread_Priority]i32{
@@ -26,15 +34,16 @@ _create :: proc(procedure: Thread_Proc, priority := Thread_Priority.Normal) -> ^
 		context = t.init_context.? or_else runtime.default_context()
 		
 		t.id = sync.current_thread_id()
+
 		t.procedure(t)
 
+		intrinsics.atomic_store(&t.flags, t.flags + {.Done})
+
 		if t.init_context == nil {
 			if context.temp_allocator.data == &runtime.global_default_temp_allocator_data {
 				runtime.default_temp_allocator_destroy(auto_cast context.temp_allocator.data)
 			}
 		}
-
-		sync.atomic_store(&t.done, true)
 		return 0
 	}
 
@@ -61,23 +70,31 @@ _create :: proc(procedure: Thread_Proc, priority := Thread_Priority.Normal) -> ^
 	return thread
 }
 
-_start :: proc(thread: ^Thread) {
-	win32.ResumeThread(thread.win32_thread)
+_start :: proc(t: ^Thread) {
+	sync.guard(&t.mutex)
+	t.flags += {.Started}
+	win32.ResumeThread(t.win32_thread)
 }
 
-_is_done :: proc(using thread: ^Thread) -> bool {
+_is_done :: proc(t: ^Thread) -> bool {
 	// NOTE(tetra, 2019-10-31): Apparently using wait_for_single_object and
 	// checking if it didn't time out immediately, is not good enough,
 	// so we do it this way instead.
-	return sync.atomic_load(&done)
+	return .Done in sync.atomic_load(&t.flags)
 }
 
-_join :: proc(using thread: ^Thread) {
-	if win32_thread != win32.INVALID_HANDLE {
-		win32.WaitForSingleObject(win32_thread, win32.INFINITE)
-		win32.CloseHandle(win32_thread)
-		win32_thread = win32.INVALID_HANDLE
+_join :: proc(t: ^Thread) {
+	sync.guard(&t.mutex)
+
+	if .Joined in t.flags || t.win32_thread == win32.INVALID_HANDLE {
+		return
 	}
+
+	win32.WaitForSingleObject(t.win32_thread, win32.INFINITE)
+	win32.CloseHandle(t.win32_thread)
+	t.win32_thread = win32.INVALID_HANDLE
+
+	t.flags += {.Joined}
 }
 
 _join_multiple :: proc(threads: ..^Thread) {
diff --git a/examples/all/all_main.odin b/examples/all/all_main.odin
index dafd29ab3..6a039e4dd 100644
--- a/examples/all/all_main.odin
+++ b/examples/all/all_main.odin
@@ -96,7 +96,6 @@ import sort           "core:sort"
 import strconv        "core:strconv"
 import strings        "core:strings"
 import sync           "core:sync"
-import sync2          "core:sync/sync2"
 import testing        "core:testing"
 import scanner        "core:text/scanner"
 import thread         "core:thread"
@@ -187,7 +186,6 @@ _ :: sort
 _ :: strconv
 _ :: strings
 _ :: sync
-_ :: sync2
 _ :: testing
 _ :: scanner
 _ :: thread
diff --git a/examples/demo/demo.odin b/examples/demo/demo.odin
index ee4a2dd2b..b3ce3ade9 100644
--- a/examples/demo/demo.odin
+++ b/examples/demo/demo.odin
@@ -1147,7 +1147,7 @@ threading_example :: proc() {
 
 	{ // Thread Pool
 		fmt.println("\n## Thread Pool")
-		task_proc :: proc(t: ^thread.Task) {
+		task_proc :: proc(t: thread.Task) {
 			index := t.user_index % len(prefix_table)
 			for iteration in 1..=5 {
 				fmt.printf("Worker Task %d is on iteration %d\n", t.user_index, iteration)
@@ -1157,7 +1157,7 @@ threading_example :: proc() {
 		}
 
 		pool: thread.Pool
-		thread.pool_init(pool=&pool, thread_count=3)
+		thread.pool_init(pool=&pool, thread_count=3, allocator=context.allocator)
 		defer thread.pool_destroy(&pool)
 
 
@@ -1166,7 +1166,7 @@ threading_example :: proc() {
 		}
 
 		thread.pool_start(&pool)
-		thread.pool_wait_and_process(&pool)
+		thread.pool_finish(&pool)
 	}
 }
 
diff --git a/src/check_builtin.cpp b/src/check_builtin.cpp
index e480704e3..8b8814176 100644
--- a/src/check_builtin.cpp
+++ b/src/check_builtin.cpp
@@ -379,6 +379,35 @@ bool check_builtin_objc_procedure(CheckerContext *c, Operand *operand, Ast *call
 	}
 }
 
+bool check_atomic_memory_order_argument(CheckerContext *c, Ast *expr, String const &builtin_name, OdinAtomicMemoryOrder *memory_order_, char const *extra_message = nullptr) {
+	Operand x = {};
+	check_expr_with_type_hint(c, &x, expr, t_atomic_memory_order);
+	if (x.mode == Addressing_Invalid) {
+		return false;
+	}
+	if (!are_types_identical(x.type, t_atomic_memory_order) || x.mode != Addressing_Constant)  {
+		gbString str = type_to_string(x.type);
+		if (extra_message) {
+			error(x.expr, "Expected a constant Atomic_Memory_Order value for the %s of '%.*s', got %s", extra_message, LIT(builtin_name), str);
+		} else {
+			error(x.expr, "Expected a constant Atomic_Memory_Order value for '%.*s', got %s", LIT(builtin_name), str);
+		}
+		gb_string_free(str);
+		return false;
+	}
+	i64 value = exact_value_to_i64(x.value);
+	if (value < 0 || value >= OdinAtomicMemoryOrder_COUNT) {
+		error(x.expr, "Illegal Atomic_Memory_Order value, got %lld", cast(long long)value);
+		return false;
+	}
+	if (memory_order_) {
+		*memory_order_ = cast(OdinAtomicMemoryOrder)value;
+	}
+
+	return true;
+
+}
+
 bool check_builtin_procedure(CheckerContext *c, Operand *operand, Ast *call, i32 id, Type *type_hint) {
 	ast_node(ce, CallExpr, call);
 	if (ce->inlining != ProcInlining_none) {
@@ -423,6 +452,11 @@ bool check_builtin_procedure(CheckerContext *c, Operand *operand, Ast *call, i32
 		// NOTE(bill): The first arg may be a Type, this will be checked case by case
 		break;
 
+	case BuiltinProc_atomic_thread_fence:
+	case BuiltinProc_atomic_signal_fence:
+		// NOTE(bill): first type will require a type hint
+		break;
+
 	case BuiltinProc_DIRECTIVE: {
 		ast_node(bd, BasicDirective, ce->proc);
 		String name = bd->name.string;
@@ -3198,11 +3232,27 @@ bool check_builtin_procedure(CheckerContext *c, Operand *operand, Ast *call, i32
 		break;
 
 
-	case BuiltinProc_atomic_fence:
-	case BuiltinProc_atomic_fence_acq:
-	case BuiltinProc_atomic_fence_rel:
-	case BuiltinProc_atomic_fence_acqrel:
-		operand->mode = Addressing_NoValue;
+
+	case BuiltinProc_atomic_thread_fence:
+	case BuiltinProc_atomic_signal_fence:
+		{
+			OdinAtomicMemoryOrder memory_order = {};
+			if (!check_atomic_memory_order_argument(c, ce->args[0], builtin_name, &memory_order)) {
+				return false;
+			}
+			switch (memory_order) {
+			case OdinAtomicMemoryOrder_acquire:
+			case OdinAtomicMemoryOrder_release:
+			case OdinAtomicMemoryOrder_acq_rel:
+			case OdinAtomicMemoryOrder_seq_cst:
+				break;
+			default:
+				error(ce->args[0], "Illegal memory ordering for '%.*s', got .%s", LIT(builtin_name), OdinAtomicMemoryOrder_strings[memory_order]);
+				break;
+			}
+
+			operand->mode = Addressing_NoValue;
+		}
 		break;
 
 	case BuiltinProc_volatile_store:
@@ -3210,9 +3260,6 @@ bool check_builtin_procedure(CheckerContext *c, Operand *operand, Ast *call, i32
 	case BuiltinProc_unaligned_store:
 		/*fallthrough*/
 	case BuiltinProc_atomic_store:
-	case BuiltinProc_atomic_store_rel:
-	case BuiltinProc_atomic_store_relaxed:
-	case BuiltinProc_atomic_store_unordered:
 		{
 			Type *elem = nullptr;
 			if (!is_type_normal_pointer(operand->type, &elem)) {
@@ -3228,14 +3275,40 @@ bool check_builtin_procedure(CheckerContext *c, Operand *operand, Ast *call, i32
 			break;
 		}
 
+	case BuiltinProc_atomic_store_explicit:
+		{
+			Type *elem = nullptr;
+			if (!is_type_normal_pointer(operand->type, &elem)) {
+				error(operand->expr, "Expected a pointer for '%.*s'", LIT(builtin_name));
+				return false;
+			}
+			Operand x = {};
+			check_expr_with_type_hint(c, &x, ce->args[1], elem);
+			check_assignment(c, &x, elem, builtin_name);
+
+			OdinAtomicMemoryOrder memory_order = {};
+			if (!check_atomic_memory_order_argument(c, ce->args[2], builtin_name, &memory_order)) {
+				return false;
+			}
+			switch (memory_order) {
+			case OdinAtomicMemoryOrder_consume:
+			case OdinAtomicMemoryOrder_acquire:
+			case OdinAtomicMemoryOrder_acq_rel:
+				error(ce->args[2], "Illegal memory order .%s for '%.*s'", OdinAtomicMemoryOrder_strings[memory_order], LIT(builtin_name));
+				break;
+			}
+
+			operand->type = nullptr;
+			operand->mode = Addressing_NoValue;
+			break;
+		}
+
+
 	case BuiltinProc_volatile_load:
 		/*fallthrough*/
 	case BuiltinProc_unaligned_load:
 		/*fallthrough*/
 	case BuiltinProc_atomic_load:
-	case BuiltinProc_atomic_load_acq:
-	case BuiltinProc_atomic_load_relaxed:
-	case BuiltinProc_atomic_load_unordered:
 		{
 			Type *elem = nullptr;
 			if (!is_type_normal_pointer(operand->type, &elem)) {
@@ -3247,41 +3320,38 @@ bool check_builtin_procedure(CheckerContext *c, Operand *operand, Ast *call, i32
 			break;
 		}
 
+	case BuiltinProc_atomic_load_explicit:
+		{
+			Type *elem = nullptr;
+			if (!is_type_normal_pointer(operand->type, &elem)) {
+				error(operand->expr, "Expected a pointer for '%.*s'", LIT(builtin_name));
+				return false;
+			}
+
+			OdinAtomicMemoryOrder memory_order = {};
+			if (!check_atomic_memory_order_argument(c, ce->args[1], builtin_name, &memory_order)) {
+				return false;
+			}
+
+			switch (memory_order) {
+			case OdinAtomicMemoryOrder_release:
+			case OdinAtomicMemoryOrder_acq_rel:
+				error(ce->args[1], "Illegal memory order .%s for '%.*s'", OdinAtomicMemoryOrder_strings[memory_order], LIT(builtin_name));
+				break;
+			}
+
+			operand->type = elem;
+			operand->mode = Addressing_Value;
+			break;
+		}
+
 	case BuiltinProc_atomic_add:
-	case BuiltinProc_atomic_add_acq:
-	case BuiltinProc_atomic_add_rel:
-	case BuiltinProc_atomic_add_acqrel:
-	case BuiltinProc_atomic_add_relaxed:
 	case BuiltinProc_atomic_sub:
-	case BuiltinProc_atomic_sub_acq:
-	case BuiltinProc_atomic_sub_rel:
-	case BuiltinProc_atomic_sub_acqrel:
-	case BuiltinProc_atomic_sub_relaxed:
 	case BuiltinProc_atomic_and:
-	case BuiltinProc_atomic_and_acq:
-	case BuiltinProc_atomic_and_rel:
-	case BuiltinProc_atomic_and_acqrel:
-	case BuiltinProc_atomic_and_relaxed:
 	case BuiltinProc_atomic_nand:
-	case BuiltinProc_atomic_nand_acq:
-	case BuiltinProc_atomic_nand_rel:
-	case BuiltinProc_atomic_nand_acqrel:
-	case BuiltinProc_atomic_nand_relaxed:
 	case BuiltinProc_atomic_or:
-	case BuiltinProc_atomic_or_acq:
-	case BuiltinProc_atomic_or_rel:
-	case BuiltinProc_atomic_or_acqrel:
-	case BuiltinProc_atomic_or_relaxed:
 	case BuiltinProc_atomic_xor:
-	case BuiltinProc_atomic_xor_acq:
-	case BuiltinProc_atomic_xor_rel:
-	case BuiltinProc_atomic_xor_acqrel:
-	case BuiltinProc_atomic_xor_relaxed:
-	case BuiltinProc_atomic_xchg:
-	case BuiltinProc_atomic_xchg_acq:
-	case BuiltinProc_atomic_xchg_rel:
-	case BuiltinProc_atomic_xchg_acqrel:
-	case BuiltinProc_atomic_xchg_relaxed:
+	case BuiltinProc_atomic_exchange:
 		{
 			Type *elem = nullptr;
 			if (!is_type_normal_pointer(operand->type, &elem)) {
@@ -3297,25 +3367,35 @@ bool check_builtin_procedure(CheckerContext *c, Operand *operand, Ast *call, i32
 			break;
 		}
 
-	case BuiltinProc_atomic_cxchg:
-	case BuiltinProc_atomic_cxchg_acq:
-	case BuiltinProc_atomic_cxchg_rel:
-	case BuiltinProc_atomic_cxchg_acqrel:
-	case BuiltinProc_atomic_cxchg_relaxed:
-	case BuiltinProc_atomic_cxchg_failrelaxed:
-	case BuiltinProc_atomic_cxchg_failacq:
-	case BuiltinProc_atomic_cxchg_acq_failrelaxed:
-	case BuiltinProc_atomic_cxchg_acqrel_failrelaxed:
+	case BuiltinProc_atomic_add_explicit:
+	case BuiltinProc_atomic_sub_explicit:
+	case BuiltinProc_atomic_and_explicit:
+	case BuiltinProc_atomic_nand_explicit:
+	case BuiltinProc_atomic_or_explicit:
+	case BuiltinProc_atomic_xor_explicit:
+	case BuiltinProc_atomic_exchange_explicit:
+		{
+			Type *elem = nullptr;
+			if (!is_type_normal_pointer(operand->type, &elem)) {
+				error(operand->expr, "Expected a pointer for '%.*s'", LIT(builtin_name));
+				return false;
+			}
+			Operand x = {};
+			check_expr_with_type_hint(c, &x, ce->args[1], elem);
+			check_assignment(c, &x, elem, builtin_name);
 
-	case BuiltinProc_atomic_cxchgweak:
-	case BuiltinProc_atomic_cxchgweak_acq:
-	case BuiltinProc_atomic_cxchgweak_rel:
-	case BuiltinProc_atomic_cxchgweak_acqrel:
-	case BuiltinProc_atomic_cxchgweak_relaxed:
-	case BuiltinProc_atomic_cxchgweak_failrelaxed:
-	case BuiltinProc_atomic_cxchgweak_failacq:
-	case BuiltinProc_atomic_cxchgweak_acq_failrelaxed:
-	case BuiltinProc_atomic_cxchgweak_acqrel_failrelaxed:
+
+			if (!check_atomic_memory_order_argument(c, ce->args[2], builtin_name, nullptr)) {
+				return false;
+			}
+
+			operand->type = elem;
+			operand->mode = Addressing_Value;
+			break;
+		}
+
+	case BuiltinProc_atomic_compare_exchange_strong:
+	case BuiltinProc_atomic_compare_exchange_weak:
 		{
 			Type *elem = nullptr;
 			if (!is_type_normal_pointer(operand->type, &elem)) {
@@ -3333,7 +3413,92 @@ bool check_builtin_procedure(CheckerContext *c, Operand *operand, Ast *call, i32
 			operand->type = elem;
 			break;
 		}
-		break;
+
+	case BuiltinProc_atomic_compare_exchange_strong_explicit:
+	case BuiltinProc_atomic_compare_exchange_weak_explicit:
+		{
+			Type *elem = nullptr;
+			if (!is_type_normal_pointer(operand->type, &elem)) {
+				error(operand->expr, "Expected a pointer for '%.*s'", LIT(builtin_name));
+				return false;
+			}
+			Operand x = {};
+			Operand y = {};
+			check_expr_with_type_hint(c, &x, ce->args[1], elem);
+			check_expr_with_type_hint(c, &y, ce->args[2], elem);
+			check_assignment(c, &x, elem, builtin_name);
+			check_assignment(c, &y, elem, builtin_name);
+
+			OdinAtomicMemoryOrder success_memory_order = {};
+			OdinAtomicMemoryOrder failure_memory_order = {};
+			if (!check_atomic_memory_order_argument(c, ce->args[3], builtin_name, &success_memory_order, "success ordering")) {
+				return false;
+			}
+			if (!check_atomic_memory_order_argument(c, ce->args[4], builtin_name, &failure_memory_order, "failure ordering")) {
+				return false;
+			}
+
+			bool invalid_combination = false;
+
+			switch (success_memory_order) {
+			case OdinAtomicMemoryOrder_relaxed:
+			case OdinAtomicMemoryOrder_release:
+				if (failure_memory_order != OdinAtomicMemoryOrder_relaxed) {
+					invalid_combination = true;
+				}
+				break;
+			case OdinAtomicMemoryOrder_consume:
+				switch (failure_memory_order) {
+				case OdinAtomicMemoryOrder_relaxed:
+				case OdinAtomicMemoryOrder_consume:
+					break;
+				default:
+					invalid_combination = true;
+					break;
+				}
+				break;
+			case OdinAtomicMemoryOrder_acquire:
+			case OdinAtomicMemoryOrder_acq_rel:
+				switch (failure_memory_order) {
+				case OdinAtomicMemoryOrder_relaxed:
+				case OdinAtomicMemoryOrder_consume:
+				case OdinAtomicMemoryOrder_acquire:
+					break;
+				default:
+					invalid_combination = true;
+					break;
+				}
+				break;
+			case OdinAtomicMemoryOrder_seq_cst:
+				switch (failure_memory_order) {
+				case OdinAtomicMemoryOrder_relaxed:
+				case OdinAtomicMemoryOrder_consume:
+				case OdinAtomicMemoryOrder_acquire:
+				case OdinAtomicMemoryOrder_seq_cst:
+					break;
+				default:
+					invalid_combination = true;
+					break;
+				}
+				break;
+			default:
+				invalid_combination = true;
+				break;
+			}
+
+
+			if (invalid_combination) {
+				error(ce->args[3], "Illegal memory order pairing for '%.*s', success = .%s, failure = .%s",
+					LIT(builtin_name),
+					OdinAtomicMemoryOrder_strings[success_memory_order],
+					OdinAtomicMemoryOrder_strings[failure_memory_order]
+				);
+			}
+
+			operand->mode = Addressing_OptionalOk;
+			operand->type = elem;
+			break;
+		}
 
 	case BuiltinProc_fixed_point_mul:
 	case BuiltinProc_fixed_point_div:
diff --git a/src/checker.cpp b/src/checker.cpp
index 53bba654d..1bb786ea1 100644
--- a/src/checker.cpp
+++ b/src/checker.cpp
@@ -829,15 +829,16 @@ struct GlobalEnumValue {
 	i64 value;
 };
 
-Slice<Entity *> add_global_enum_type(String const &type_name, GlobalEnumValue *values, isize value_count) {
+Slice<Entity *> add_global_enum_type(String const &type_name, GlobalEnumValue *values, isize value_count, Type **enum_type_ = nullptr) {
 	Scope *scope = create_scope(nullptr, builtin_pkg->scope);
-	Entity *e = alloc_entity_type_name(scope, make_token_ident(type_name), nullptr, EntityState_Resolved);
+	Entity *entity = alloc_entity_type_name(scope, make_token_ident(type_name), nullptr, EntityState_Resolved);
 
 	Type *enum_type = alloc_type_enum();
-	Type *named_type = alloc_type_named(type_name, enum_type, e);
+	Type *named_type = alloc_type_named(type_name, enum_type, entity);
 	set_base_type(named_type, enum_type);
 	enum_type->Enum.base_type = t_int;
 	enum_type->Enum.scope = scope;
+	entity->type = named_type;
 
 	auto fields = array_make<Entity *>(permanent_allocator(), value_count);
 	for (isize i = 0; i < value_count; i++) {
@@ -858,6 +859,9 @@ Slice<Entity *> add_global_enum_type(String const &type_name, GlobalEnumValue *v
 	enum_type->Enum.min_value = &enum_type->Enum.fields[enum_type->Enum.min_value_index]->Constant.value;
 	enum_type->Enum.max_value = &enum_type->Enum.fields[enum_type->Enum.max_value_index]->Constant.value;
 
+
+	if (enum_type_) *enum_type_ = named_type;
+
 	return slice_from_array(fields);
 }
 void add_global_enum_constant(Slice<Entity *> const &fields, char const *name, i64 value) {
@@ -986,6 +990,21 @@ void init_universal(void) {
 		add_global_enum_constant(fields, "ODIN_ERROR_POS_STYLE", build_context.ODIN_ERROR_POS_STYLE);
 	}
 
+	{
+		GlobalEnumValue values[OdinAtomicMemoryOrder_COUNT] = {
+			{OdinAtomicMemoryOrder_strings[OdinAtomicMemoryOrder_relaxed], OdinAtomicMemoryOrder_relaxed},
+			{OdinAtomicMemoryOrder_strings[OdinAtomicMemoryOrder_consume], OdinAtomicMemoryOrder_consume},
+			{OdinAtomicMemoryOrder_strings[OdinAtomicMemoryOrder_acquire], OdinAtomicMemoryOrder_acquire},
+			{OdinAtomicMemoryOrder_strings[OdinAtomicMemoryOrder_release], OdinAtomicMemoryOrder_release},
+			{OdinAtomicMemoryOrder_strings[OdinAtomicMemoryOrder_acq_rel], OdinAtomicMemoryOrder_acq_rel},
+			{OdinAtomicMemoryOrder_strings[OdinAtomicMemoryOrder_seq_cst], OdinAtomicMemoryOrder_seq_cst},
+		};
+
+		add_global_enum_type(str_lit("Atomic_Memory_Order"), values, gb_count_of(values), &t_atomic_memory_order);
+		GB_ASSERT(t_atomic_memory_order->kind == Type_Named);
+		scope_insert(intrinsics_pkg->scope, t_atomic_memory_order->Named.type_name);
+	}
+
 
 	add_global_bool_constant("ODIN_DEBUG",                    bc->ODIN_DEBUG);
 	add_global_bool_constant("ODIN_DISABLE_ASSERT",           bc->ODIN_DISABLE_ASSERT);
diff --git a/src/checker_builtin_procs.hpp b/src/checker_builtin_procs.hpp
index c647d5a01..c6b0afee0 100644
--- a/src/checker_builtin_procs.hpp
+++ b/src/checker_builtin_procs.hpp
@@ -86,77 +86,30 @@ enum BuiltinProcId {
 	BuiltinProc_prefetch_write_instruction,
 	BuiltinProc_prefetch_write_data,
 
-	BuiltinProc_atomic_fence,
-	BuiltinProc_atomic_fence_acq,
-	BuiltinProc_atomic_fence_rel,
-	BuiltinProc_atomic_fence_acqrel,
-
+	BuiltinProc_atomic_thread_fence,
+	BuiltinProc_atomic_signal_fence,
 	BuiltinProc_atomic_store,
-	BuiltinProc_atomic_store_rel,
-	BuiltinProc_atomic_store_relaxed,
-	BuiltinProc_atomic_store_unordered,
-
+	BuiltinProc_atomic_store_explicit,
 	BuiltinProc_atomic_load,
-	BuiltinProc_atomic_load_acq,
-	BuiltinProc_atomic_load_relaxed,
-	BuiltinProc_atomic_load_unordered,
-
+	BuiltinProc_atomic_load_explicit,
 	BuiltinProc_atomic_add,
-	BuiltinProc_atomic_add_acq,
-	BuiltinProc_atomic_add_rel,
-	BuiltinProc_atomic_add_acqrel,
-	BuiltinProc_atomic_add_relaxed,
+	BuiltinProc_atomic_add_explicit,
 	BuiltinProc_atomic_sub,
-	BuiltinProc_atomic_sub_acq,
-	BuiltinProc_atomic_sub_rel,
-	BuiltinProc_atomic_sub_acqrel,
-	BuiltinProc_atomic_sub_relaxed,
+	BuiltinProc_atomic_sub_explicit,
 	BuiltinProc_atomic_and,
-	BuiltinProc_atomic_and_acq,
-	BuiltinProc_atomic_and_rel,
-	BuiltinProc_atomic_and_acqrel,
-	BuiltinProc_atomic_and_relaxed,
+	BuiltinProc_atomic_and_explicit,
 	BuiltinProc_atomic_nand,
-	BuiltinProc_atomic_nand_acq,
-	BuiltinProc_atomic_nand_rel,
-	BuiltinProc_atomic_nand_acqrel,
-	BuiltinProc_atomic_nand_relaxed,
+	BuiltinProc_atomic_nand_explicit,
 	BuiltinProc_atomic_or,
-	BuiltinProc_atomic_or_acq,
-	BuiltinProc_atomic_or_rel,
-	BuiltinProc_atomic_or_acqrel,
-	BuiltinProc_atomic_or_relaxed,
+	BuiltinProc_atomic_or_explicit,
 	BuiltinProc_atomic_xor,
-	BuiltinProc_atomic_xor_acq,
-	BuiltinProc_atomic_xor_rel,
-	BuiltinProc_atomic_xor_acqrel,
-	BuiltinProc_atomic_xor_relaxed,
-
-	BuiltinProc_atomic_xchg,
-	BuiltinProc_atomic_xchg_acq,
-	BuiltinProc_atomic_xchg_rel,
-	BuiltinProc_atomic_xchg_acqrel,
-	BuiltinProc_atomic_xchg_relaxed,
-
-	BuiltinProc_atomic_cxchg,
-	BuiltinProc_atomic_cxchg_acq,
-	BuiltinProc_atomic_cxchg_rel,
-	BuiltinProc_atomic_cxchg_acqrel,
-	BuiltinProc_atomic_cxchg_relaxed,
-	BuiltinProc_atomic_cxchg_failrelaxed,
-	BuiltinProc_atomic_cxchg_failacq,
-	BuiltinProc_atomic_cxchg_acq_failrelaxed,
-	BuiltinProc_atomic_cxchg_acqrel_failrelaxed,
-
-	BuiltinProc_atomic_cxchgweak,
-	BuiltinProc_atomic_cxchgweak_acq,
-	BuiltinProc_atomic_cxchgweak_rel,
-	BuiltinProc_atomic_cxchgweak_acqrel,
-	BuiltinProc_atomic_cxchgweak_relaxed,
-	BuiltinProc_atomic_cxchgweak_failrelaxed,
-	BuiltinProc_atomic_cxchgweak_failacq,
-	BuiltinProc_atomic_cxchgweak_acq_failrelaxed,
-	BuiltinProc_atomic_cxchgweak_acqrel_failrelaxed,
+	BuiltinProc_atomic_xor_explicit,
+	BuiltinProc_atomic_exchange,
+	BuiltinProc_atomic_exchange_explicit,
+	BuiltinProc_atomic_compare_exchange_strong,
+	BuiltinProc_atomic_compare_exchange_strong_explicit,
+	BuiltinProc_atomic_compare_exchange_weak,
+	BuiltinProc_atomic_compare_exchange_weak_explicit,
 
 	BuiltinProc_fixed_point_mul,
 	BuiltinProc_fixed_point_div,
@@ -352,78 +305,30 @@ gb_global BuiltinProc builtin_procs[BuiltinProc_COUNT] = {
 	{STR_LIT("prefetch_write_instruction"), 2, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
 	{STR_LIT("prefetch_write_data"),        2, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
 
-	{STR_LIT("atomic_fence"),        0, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_fence_acq"),    0, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_fence_rel"),    0, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_fence_acqrel"), 0, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
-
-	{STR_LIT("atomic_store"),           2, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_store_rel"),       2, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_store_relaxed"),   2, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_store_unordered"), 2, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
-
-	{STR_LIT("atomic_load"),            1, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_load_acq"),        1, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_load_relaxed"),    1, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_load_unordered"),  1, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-
-	{STR_LIT("atomic_add"),             2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_add_acq"),         2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_add_rel"),         2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_add_acqrel"),      2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_add_relaxed"),     2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_sub"),             2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_sub_acq"),         2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_sub_rel"),         2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_sub_acqrel"),      2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_sub_relaxed"),     2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_and"),             2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_and_acq"),         2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_and_rel"),         2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_and_acqrel"),      2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_and_relaxed"),     2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_nand"),            2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_nand_acq"),        2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_nand_rel"),        2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_nand_acqrel"),     2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_nand_relaxed"),    2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_or"),              2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_or_acq"),          2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_or_rel"),          2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_or_acqrel"),       2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_or_relaxed"),      2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_xor"),             2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_xor_acq"),         2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_xor_rel"),         2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_xor_acqrel"),      2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_xor_relaxed"),     2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-
-	{STR_LIT("atomic_xchg"),            2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_xchg_acq"),        2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_xchg_rel"),        2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_xchg_acqrel"),     2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_xchg_relaxed"),    2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-
-	{STR_LIT("atomic_cxchg"),                    3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchg_acq"),                3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchg_rel"),                3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchg_acqrel"),             3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchg_relaxed"),            3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchg_failrelaxed"),        3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchg_failacq"),            3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchg_acq_failrelaxed"),    3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchg_acqrel_failrelaxed"), 3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-
-	{STR_LIT("atomic_cxchgweak"),                    3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchgweak_acq"),                3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchgweak_rel"),                3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchgweak_acqrel"),             3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchgweak_relaxed"),            3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchgweak_failrelaxed"),        3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchgweak_failacq"),            3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchgweak_acq_failrelaxed"),    3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-	{STR_LIT("atomic_cxchgweak_acqrel_failrelaxed"), 3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
-
+	{STR_LIT("atomic_thread_fence"),                     1, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_signal_fence"),                     1, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_store"),                            2, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_store_explicit"),                   3, false, Expr_Stmt, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_load"),                             1, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_load_explicit"),                    2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_add"),                              2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_add_explicit"),                     3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_sub"),                              2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_sub_explicit"),                     3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_and"),                              2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_and_explicit"),                     3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_nand"),                             2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_nand_explicit"),                    3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_or"),                               2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_or_explicit"),                      3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_xor"),                              2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_xor_explicit"),                     3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_exchange"),                         2, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_exchange_explicit"),                3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_compare_exchange_strong"),          3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_compare_exchange_strong_explicit"), 5, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_compare_exchange_weak"),            3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
+	{STR_LIT("atomic_compare_exchange_weak_explicit"),   5, false, Expr_Expr, BuiltinProcPkg_intrinsics},
 
 	{STR_LIT("fixed_point_mul"), 3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
 	{STR_LIT("fixed_point_div"), 3, false, Expr_Expr, BuiltinProcPkg_intrinsics},
diff --git a/src/llvm_backend_proc.cpp b/src/llvm_backend_proc.cpp
index d7f3d6c45..4de2af9d8 100644
--- a/src/llvm_backend_proc.cpp
+++ b/src/llvm_backend_proc.cpp
@@ -1606,36 +1606,26 @@ lbValue lb_build_builtin_proc(lbProcedure *p, Ast *expr, TypeAndValue const &tv,
 		}
 
 
-
-	case BuiltinProc_atomic_fence:
-		LLVMBuildFence(p->builder, LLVMAtomicOrderingSequentiallyConsistent, false, "");
+	// TODO(bill): Which is correct?
+	case BuiltinProc_atomic_thread_fence:
+		LLVMBuildFence(p->builder, llvm_atomic_ordering_from_odin(ce->args[0]), false, "");
 		return {};
-	case BuiltinProc_atomic_fence_acq:
-		LLVMBuildFence(p->builder, LLVMAtomicOrderingAcquire, false, "");
-		return {};
-	case BuiltinProc_atomic_fence_rel:
-		LLVMBuildFence(p->builder, LLVMAtomicOrderingRelease, false, "");
-		return {};
-	case BuiltinProc_atomic_fence_acqrel:
-		LLVMBuildFence(p->builder, LLVMAtomicOrderingAcquireRelease, false, "");
+	case BuiltinProc_atomic_signal_fence:
+		LLVMBuildFence(p->builder, llvm_atomic_ordering_from_odin(ce->args[0]), true, "");
 		return {};
 
 	case BuiltinProc_volatile_store:
 	case BuiltinProc_atomic_store:
-	case BuiltinProc_atomic_store_rel:
-	case BuiltinProc_atomic_store_relaxed:
-	case BuiltinProc_atomic_store_unordered: {
+	case BuiltinProc_atomic_store_explicit: {
 		lbValue dst = lb_build_expr(p, ce->args[0]);
 		lbValue val = lb_build_expr(p, ce->args[1]);
 		val = lb_emit_conv(p, val, type_deref(dst.type));
 
 		LLVMValueRef instr = LLVMBuildStore(p->builder, val.value, dst.value);
 		switch (id) {
-		case BuiltinProc_volatile_store:         LLVMSetVolatile(instr, true);                                     break;
-		case BuiltinProc_atomic_store:           LLVMSetOrdering(instr, LLVMAtomicOrderingSequentiallyConsistent); break;
-		case BuiltinProc_atomic_store_rel:       LLVMSetOrdering(instr, LLVMAtomicOrderingRelease);                break;
-		case BuiltinProc_atomic_store_relaxed:   LLVMSetOrdering(instr, LLVMAtomicOrderingMonotonic);              break;
-		case BuiltinProc_atomic_store_unordered: LLVMSetOrdering(instr, LLVMAtomicOrderingUnordered);              break;
+		case BuiltinProc_volatile_store:        LLVMSetVolatile(instr, true);                                        break;
+		case BuiltinProc_atomic_store:          LLVMSetOrdering(instr, LLVMAtomicOrderingSequentiallyConsistent);    break;
+		case BuiltinProc_atomic_store_explicit: LLVMSetOrdering(instr, llvm_atomic_ordering_from_odin(ce->args[2])); break;
 		}
 
 		LLVMSetAlignment(instr, cast(unsigned)type_align_of(type_deref(dst.type)));
@@ -1645,18 +1635,14 @@ lbValue lb_build_builtin_proc(lbProcedure *p, Ast *expr, TypeAndValue const &tv,
 
 	case BuiltinProc_volatile_load:
 	case BuiltinProc_atomic_load:
-	case BuiltinProc_atomic_load_acq:
-	case BuiltinProc_atomic_load_relaxed:
-	case BuiltinProc_atomic_load_unordered: {
+	case BuiltinProc_atomic_load_explicit: {
 		lbValue dst = lb_build_expr(p, ce->args[0]);
 
 		LLVMValueRef instr = LLVMBuildLoad(p->builder, dst.value, "");
 		switch (id) {
-		case BuiltinProc_volatile_load:         LLVMSetVolatile(instr, true);                                     break;
-		case BuiltinProc_atomic_load:           LLVMSetOrdering(instr, LLVMAtomicOrderingSequentiallyConsistent); break;
-		case BuiltinProc_atomic_load_acq:       LLVMSetOrdering(instr, LLVMAtomicOrderingAcquire);                break;
-		case BuiltinProc_atomic_load_relaxed:   LLVMSetOrdering(instr, LLVMAtomicOrderingMonotonic);              break;
-		case BuiltinProc_atomic_load_unordered: LLVMSetOrdering(instr, LLVMAtomicOrderingUnordered);              break;
+		case BuiltinProc_volatile_load:        LLVMSetVolatile(instr, true);                                        break;
+		case BuiltinProc_atomic_load:          LLVMSetOrdering(instr, LLVMAtomicOrderingSequentiallyConsistent);    break;
+		case BuiltinProc_atomic_load_explicit: LLVMSetOrdering(instr, llvm_atomic_ordering_from_odin(ce->args[1])); break;
 		}
 		LLVMSetAlignment(instr, cast(unsigned)type_align_of(type_deref(dst.type)));
 
@@ -1686,40 +1672,19 @@ lbValue lb_build_builtin_proc(lbProcedure *p, Ast *expr, TypeAndValue const &tv,
 		}
 
 	case BuiltinProc_atomic_add:
-	case BuiltinProc_atomic_add_acq:
-	case BuiltinProc_atomic_add_rel:
-	case BuiltinProc_atomic_add_acqrel:
-	case BuiltinProc_atomic_add_relaxed:
 	case BuiltinProc_atomic_sub:
-	case BuiltinProc_atomic_sub_acq:
-	case BuiltinProc_atomic_sub_rel:
-	case BuiltinProc_atomic_sub_acqrel:
-	case BuiltinProc_atomic_sub_relaxed:
 	case BuiltinProc_atomic_and:
-	case BuiltinProc_atomic_and_acq:
-	case BuiltinProc_atomic_and_rel:
-	case BuiltinProc_atomic_and_acqrel:
-	case BuiltinProc_atomic_and_relaxed:
 	case BuiltinProc_atomic_nand:
-	case BuiltinProc_atomic_nand_acq:
-	case BuiltinProc_atomic_nand_rel:
-	case BuiltinProc_atomic_nand_acqrel:
-	case BuiltinProc_atomic_nand_relaxed:
 	case BuiltinProc_atomic_or:
-	case BuiltinProc_atomic_or_acq:
-	case BuiltinProc_atomic_or_rel:
-	case BuiltinProc_atomic_or_acqrel:
-	case BuiltinProc_atomic_or_relaxed:
 	case BuiltinProc_atomic_xor:
-	case BuiltinProc_atomic_xor_acq:
-	case BuiltinProc_atomic_xor_rel:
-	case BuiltinProc_atomic_xor_acqrel:
-	case BuiltinProc_atomic_xor_relaxed:
-	case BuiltinProc_atomic_xchg:
-	case BuiltinProc_atomic_xchg_acq:
-	case BuiltinProc_atomic_xchg_rel:
-	case BuiltinProc_atomic_xchg_acqrel:
-	case BuiltinProc_atomic_xchg_relaxed: {
+	case BuiltinProc_atomic_exchange:
+	case BuiltinProc_atomic_add_explicit:
+	case BuiltinProc_atomic_sub_explicit:
+	case BuiltinProc_atomic_and_explicit:
+	case BuiltinProc_atomic_nand_explicit:
+	case BuiltinProc_atomic_or_explicit:
+	case BuiltinProc_atomic_xor_explicit:
+	case BuiltinProc_atomic_exchange_explicit: {
 		lbValue dst = lb_build_expr(p, ce->args[0]);
 		lbValue val = lb_build_expr(p, ce->args[1]);
 		val = lb_emit_conv(p, val, type_deref(dst.type));
@@ -1728,41 +1693,20 @@ lbValue lb_build_builtin_proc(lbProcedure *p, Ast *expr, TypeAndValue const &tv,
 		LLVMAtomicOrdering ordering = {};
 
 		switch (id) {
-		case BuiltinProc_atomic_add:          op = LLVMAtomicRMWBinOpAdd;  ordering = LLVMAtomicOrderingSequentiallyConsistent; break;
-		case BuiltinProc_atomic_add_acq:      op = LLVMAtomicRMWBinOpAdd;  ordering = LLVMAtomicOrderingAcquire; break;
-		case BuiltinProc_atomic_add_rel:      op = LLVMAtomicRMWBinOpAdd;  ordering = LLVMAtomicOrderingRelease; break;
-		case BuiltinProc_atomic_add_acqrel:   op = LLVMAtomicRMWBinOpAdd;  ordering = LLVMAtomicOrderingAcquireRelease; break;
-		case BuiltinProc_atomic_add_relaxed:  op = LLVMAtomicRMWBinOpAdd;  ordering = LLVMAtomicOrderingMonotonic; break;
-		case BuiltinProc_atomic_sub:          op = LLVMAtomicRMWBinOpSub;  ordering = LLVMAtomicOrderingSequentiallyConsistent; break;
-		case BuiltinProc_atomic_sub_acq:      op = LLVMAtomicRMWBinOpSub;  ordering = LLVMAtomicOrderingAcquire; break;
-		case BuiltinProc_atomic_sub_rel:      op = LLVMAtomicRMWBinOpSub;  ordering = LLVMAtomicOrderingRelease; break;
-		case BuiltinProc_atomic_sub_acqrel:   op = LLVMAtomicRMWBinOpSub;  ordering = LLVMAtomicOrderingAcquireRelease; break;
-		case BuiltinProc_atomic_sub_relaxed:  op = LLVMAtomicRMWBinOpSub;  ordering = LLVMAtomicOrderingMonotonic; break;
-		case BuiltinProc_atomic_and:          op = LLVMAtomicRMWBinOpAnd;  ordering = LLVMAtomicOrderingSequentiallyConsistent; break;
-		case BuiltinProc_atomic_and_acq:      op = LLVMAtomicRMWBinOpAnd;  ordering = LLVMAtomicOrderingAcquire; break;
-		case BuiltinProc_atomic_and_rel:      op = LLVMAtomicRMWBinOpAnd;  ordering = LLVMAtomicOrderingRelease; break;
-		case BuiltinProc_atomic_and_acqrel:   op = LLVMAtomicRMWBinOpAnd;  ordering = LLVMAtomicOrderingAcquireRelease; break;
-		case BuiltinProc_atomic_and_relaxed:  op = LLVMAtomicRMWBinOpAnd;  ordering = LLVMAtomicOrderingMonotonic; break;
-		case BuiltinProc_atomic_nand:         op = LLVMAtomicRMWBinOpNand; ordering = LLVMAtomicOrderingSequentiallyConsistent; break;
-		case BuiltinProc_atomic_nand_acq:     op = LLVMAtomicRMWBinOpNand; ordering = LLVMAtomicOrderingAcquire; break;
-		case BuiltinProc_atomic_nand_rel:     op = LLVMAtomicRMWBinOpNand; ordering = LLVMAtomicOrderingRelease; break;
-		case BuiltinProc_atomic_nand_acqrel:  op = LLVMAtomicRMWBinOpNand; ordering = LLVMAtomicOrderingAcquireRelease; break;
-		case BuiltinProc_atomic_nand_relaxed: op = LLVMAtomicRMWBinOpNand; ordering = LLVMAtomicOrderingMonotonic; break;
-		case BuiltinProc_atomic_or:           op = LLVMAtomicRMWBinOpOr;   ordering = LLVMAtomicOrderingSequentiallyConsistent; break;
-		case BuiltinProc_atomic_or_acq:       op = LLVMAtomicRMWBinOpOr;   ordering = LLVMAtomicOrderingAcquire; break;
-		case BuiltinProc_atomic_or_rel:       op = LLVMAtomicRMWBinOpOr;   ordering = LLVMAtomicOrderingRelease; break;
-		case BuiltinProc_atomic_or_acqrel:    op = LLVMAtomicRMWBinOpOr;   ordering = LLVMAtomicOrderingAcquireRelease; break;
-		case BuiltinProc_atomic_or_relaxed:   op = LLVMAtomicRMWBinOpOr;   ordering = LLVMAtomicOrderingMonotonic; break;
-		case BuiltinProc_atomic_xor:          op = LLVMAtomicRMWBinOpXor;  ordering = LLVMAtomicOrderingSequentiallyConsistent; break;
-		case BuiltinProc_atomic_xor_acq:      op = LLVMAtomicRMWBinOpXor;  ordering = LLVMAtomicOrderingAcquire; break;
-		case BuiltinProc_atomic_xor_rel:      op = LLVMAtomicRMWBinOpXor;  ordering = LLVMAtomicOrderingRelease; break;
-		case BuiltinProc_atomic_xor_acqrel:   op = LLVMAtomicRMWBinOpXor;  ordering = LLVMAtomicOrderingAcquireRelease; break;
-		case BuiltinProc_atomic_xor_relaxed:  op = LLVMAtomicRMWBinOpXor;  ordering = LLVMAtomicOrderingMonotonic; break;
-		case BuiltinProc_atomic_xchg:         op = LLVMAtomicRMWBinOpXchg; ordering = LLVMAtomicOrderingSequentiallyConsistent; break;
-		case BuiltinProc_atomic_xchg_acq:     op = LLVMAtomicRMWBinOpXchg; ordering = LLVMAtomicOrderingAcquire; break;
-		case BuiltinProc_atomic_xchg_rel:     op = LLVMAtomicRMWBinOpXchg; ordering = LLVMAtomicOrderingRelease; break;
-		case BuiltinProc_atomic_xchg_acqrel:  op = LLVMAtomicRMWBinOpXchg; ordering = LLVMAtomicOrderingAcquireRelease; break;
-		case BuiltinProc_atomic_xchg_relaxed: op = LLVMAtomicRMWBinOpXchg; ordering = LLVMAtomicOrderingMonotonic; break;
+		case BuiltinProc_atomic_add:               op = LLVMAtomicRMWBinOpAdd;  ordering = LLVMAtomicOrderingSequentiallyConsistent;    break;
+		case BuiltinProc_atomic_sub:               op = LLVMAtomicRMWBinOpSub;  ordering = LLVMAtomicOrderingSequentiallyConsistent;    break;
+		case BuiltinProc_atomic_and:               op = LLVMAtomicRMWBinOpAnd;  ordering = LLVMAtomicOrderingSequentiallyConsistent;    break;
+		case BuiltinProc_atomic_nand:              op = LLVMAtomicRMWBinOpNand; ordering = LLVMAtomicOrderingSequentiallyConsistent;    break;
+		case BuiltinProc_atomic_or:                op = LLVMAtomicRMWBinOpOr;   ordering = LLVMAtomicOrderingSequentiallyConsistent;    break;
+		case BuiltinProc_atomic_xor:               op = LLVMAtomicRMWBinOpXor;  ordering = LLVMAtomicOrderingSequentiallyConsistent;    break;
+		case BuiltinProc_atomic_exchange:          op = LLVMAtomicRMWBinOpXchg; ordering = LLVMAtomicOrderingSequentiallyConsistent;    break;
+		case BuiltinProc_atomic_add_explicit:      op = LLVMAtomicRMWBinOpAdd;  ordering = llvm_atomic_ordering_from_odin(ce->args[2]); break;
+		case BuiltinProc_atomic_sub_explicit:      op = LLVMAtomicRMWBinOpSub;  ordering = llvm_atomic_ordering_from_odin(ce->args[2]); break;
+		case BuiltinProc_atomic_and_explicit:      op = LLVMAtomicRMWBinOpAnd;  ordering = llvm_atomic_ordering_from_odin(ce->args[2]); break;
+		case BuiltinProc_atomic_nand_explicit:     op = LLVMAtomicRMWBinOpNand; ordering = llvm_atomic_ordering_from_odin(ce->args[2]); break;
+		case BuiltinProc_atomic_or_explicit:       op = LLVMAtomicRMWBinOpOr;   ordering = llvm_atomic_ordering_from_odin(ce->args[2]); break;
+		case BuiltinProc_atomic_xor_explicit:      op = LLVMAtomicRMWBinOpXor;  ordering = llvm_atomic_ordering_from_odin(ce->args[2]); break;
+		case BuiltinProc_atomic_exchange_explicit: op = LLVMAtomicRMWBinOpXchg; ordering = llvm_atomic_ordering_from_odin(ce->args[2]); break;
 		}
 
 		lbValue res = {};
@@ -1771,24 +1715,10 @@ lbValue lb_build_builtin_proc(lbProcedure *p, Ast *expr, TypeAndValue const &tv,
 		return res;
 	}
 
-	case BuiltinProc_atomic_cxchg:
-	case BuiltinProc_atomic_cxchg_acq:
-	case BuiltinProc_atomic_cxchg_rel:
-	case BuiltinProc_atomic_cxchg_acqrel:
-	case BuiltinProc_atomic_cxchg_relaxed:
-	case BuiltinProc_atomic_cxchg_failrelaxed:
-	case BuiltinProc_atomic_cxchg_failacq:
-	case BuiltinProc_atomic_cxchg_acq_failrelaxed:
-	case BuiltinProc_atomic_cxchg_acqrel_failrelaxed:
-	case BuiltinProc_atomic_cxchgweak:
-	case BuiltinProc_atomic_cxchgweak_acq:
-	case BuiltinProc_atomic_cxchgweak_rel:
-	case BuiltinProc_atomic_cxchgweak_acqrel:
-	case BuiltinProc_atomic_cxchgweak_relaxed:
-	case BuiltinProc_atomic_cxchgweak_failrelaxed:
-	case BuiltinProc_atomic_cxchgweak_failacq:
-	case BuiltinProc_atomic_cxchgweak_acq_failrelaxed:
-	case BuiltinProc_atomic_cxchgweak_acqrel_failrelaxed: {
+	case BuiltinProc_atomic_compare_exchange_strong:
+	case BuiltinProc_atomic_compare_exchange_weak:
+	case BuiltinProc_atomic_compare_exchange_strong_explicit:
+	case BuiltinProc_atomic_compare_exchange_weak_explicit: {
 		lbValue address = lb_build_expr(p, ce->args[0]);
 		Type *elem = type_deref(address.type);
 		lbValue old_value = lb_build_expr(p, ce->args[1]);
@@ -1801,28 +1731,14 @@ lbValue lb_build_builtin_proc(lbProcedure *p, Ast *expr, TypeAndValue const &tv,
 		LLVMBool weak = false;
 
 		switch (id) {
-		case BuiltinProc_atomic_cxchg:                        success_ordering = LLVMAtomicOrderingSequentiallyConsistent; failure_ordering = LLVMAtomicOrderingSequentiallyConsistent; weak = false; break;
-		case BuiltinProc_atomic_cxchg_acq:                    success_ordering = LLVMAtomicOrderingAcquire;                failure_ordering = LLVMAtomicOrderingMonotonic;              weak = false; break;
-		case BuiltinProc_atomic_cxchg_rel:                    success_ordering = LLVMAtomicOrderingRelease;                failure_ordering = LLVMAtomicOrderingMonotonic;              weak = false; break;
-		case BuiltinProc_atomic_cxchg_acqrel:                 success_ordering = LLVMAtomicOrderingAcquireRelease;         failure_ordering = LLVMAtomicOrderingMonotonic;              weak = false; break;
-		case BuiltinProc_atomic_cxchg_relaxed:                success_ordering = LLVMAtomicOrderingMonotonic;              failure_ordering = LLVMAtomicOrderingMonotonic;              weak = false; break;
-		case BuiltinProc_atomic_cxchg_failrelaxed:            success_ordering = LLVMAtomicOrderingSequentiallyConsistent; failure_ordering = LLVMAtomicOrderingMonotonic;              weak = false; break;
-		case BuiltinProc_atomic_cxchg_failacq:                success_ordering = LLVMAtomicOrderingSequentiallyConsistent; failure_ordering = LLVMAtomicOrderingAcquire;                weak = false; break;
-		case BuiltinProc_atomic_cxchg_acq_failrelaxed:        success_ordering = LLVMAtomicOrderingAcquire;                failure_ordering = LLVMAtomicOrderingMonotonic;              weak = false; break;
-		case BuiltinProc_atomic_cxchg_acqrel_failrelaxed:     success_ordering = LLVMAtomicOrderingAcquireRelease;         failure_ordering = LLVMAtomicOrderingMonotonic;              weak = false; break;
-		case BuiltinProc_atomic_cxchgweak:                    success_ordering = LLVMAtomicOrderingSequentiallyConsistent; failure_ordering = LLVMAtomicOrderingSequentiallyConsistent; weak = false; break;
-		case BuiltinProc_atomic_cxchgweak_acq:                success_ordering = LLVMAtomicOrderingAcquire;                failure_ordering = LLVMAtomicOrderingMonotonic;              weak = true;  break;
-		case BuiltinProc_atomic_cxchgweak_rel:                success_ordering = LLVMAtomicOrderingRelease;                failure_ordering = LLVMAtomicOrderingMonotonic;              weak = true;  break;
-		case BuiltinProc_atomic_cxchgweak_acqrel:             success_ordering = LLVMAtomicOrderingAcquireRelease;         failure_ordering = LLVMAtomicOrderingMonotonic;              weak = true;  break;
-		case BuiltinProc_atomic_cxchgweak_relaxed:            success_ordering = LLVMAtomicOrderingMonotonic;              failure_ordering = LLVMAtomicOrderingMonotonic;              weak = true;  break;
-		case BuiltinProc_atomic_cxchgweak_failrelaxed:        success_ordering = LLVMAtomicOrderingSequentiallyConsistent; failure_ordering = LLVMAtomicOrderingMonotonic;              weak = true;  break;
-		case BuiltinProc_atomic_cxchgweak_failacq:            success_ordering = LLVMAtomicOrderingSequentiallyConsistent; failure_ordering = LLVMAtomicOrderingAcquire;                weak = true;  break;
-		case BuiltinProc_atomic_cxchgweak_acq_failrelaxed:    success_ordering = LLVMAtomicOrderingAcquire;                failure_ordering = LLVMAtomicOrderingMonotonic;              weak = true;  break;
-		case BuiltinProc_atomic_cxchgweak_acqrel_failrelaxed: success_ordering = LLVMAtomicOrderingAcquireRelease;         failure_ordering = LLVMAtomicOrderingMonotonic;              weak = true;  break;
+		case BuiltinProc_atomic_compare_exchange_strong:          success_ordering = LLVMAtomicOrderingSequentiallyConsistent;    failure_ordering = LLVMAtomicOrderingSequentiallyConsistent;    weak = false; break;
+		case BuiltinProc_atomic_compare_exchange_weak:            success_ordering = LLVMAtomicOrderingSequentiallyConsistent;    failure_ordering = LLVMAtomicOrderingSequentiallyConsistent;    weak = true;  break;
+		case BuiltinProc_atomic_compare_exchange_strong_explicit: success_ordering = llvm_atomic_ordering_from_odin(ce->args[3]); failure_ordering = llvm_atomic_ordering_from_odin(ce->args[4]); weak = false; break;
+		case BuiltinProc_atomic_compare_exchange_weak_explicit:   success_ordering = llvm_atomic_ordering_from_odin(ce->args[3]); failure_ordering = llvm_atomic_ordering_from_odin(ce->args[4]); weak = true;  break;
 		}
 
 		// TODO(bill): Figure out how to make it weak
-		LLVMBool single_threaded = weak;
+		LLVMBool single_threaded = false;
 
 		LLVMValueRef value = LLVMBuildAtomicCmpXchg(
 			p->builder, address.value,
@@ -1831,6 +1747,7 @@ lbValue lb_build_builtin_proc(lbProcedure *p, Ast *expr, TypeAndValue const &tv,
 			failure_ordering,
 			single_threaded
 		);
+		LLVMSetWeak(value, weak);
 
 		if (tv.type->kind == Type_Tuple) {
 			Type *fix_typed = alloc_type_tuple();
diff --git a/src/llvm_backend_utility.cpp b/src/llvm_backend_utility.cpp
index 399d1632d..037171637 100644
--- a/src/llvm_backend_utility.cpp
+++ b/src/llvm_backend_utility.cpp
@@ -2005,3 +2005,25 @@ lbValue lb_handle_objc_send(lbProcedure *p, Ast *expr) {
 }
 
 
+
+
+LLVMAtomicOrdering llvm_atomic_ordering_from_odin(ExactValue const &value) {
+	GB_ASSERT(value.kind == ExactValue_Integer);
+	i64 v = exact_value_to_i64(value);
+	switch (v) {
+	case OdinAtomicMemoryOrder_relaxed: return LLVMAtomicOrderingUnordered;
+	case OdinAtomicMemoryOrder_consume: return LLVMAtomicOrderingMonotonic;
+	case OdinAtomicMemoryOrder_acquire: return LLVMAtomicOrderingAcquire;
+	case OdinAtomicMemoryOrder_release: return LLVMAtomicOrderingRelease;
+	case OdinAtomicMemoryOrder_acq_rel: return LLVMAtomicOrderingAcquireRelease;
+	case OdinAtomicMemoryOrder_seq_cst: return LLVMAtomicOrderingSequentiallyConsistent;
+	}
+	GB_PANIC("Unknown atomic ordering");
+	return LLVMAtomicOrderingSequentiallyConsistent;
+}
+
+
+LLVMAtomicOrdering llvm_atomic_ordering_from_odin(Ast *expr) {
+	ExactValue value = type_and_value_of_expr(expr).value;
+	return llvm_atomic_ordering_from_odin(value);
+}
diff --git a/src/types.cpp b/src/types.cpp
index b231218a2..e10dae1ed 100644
--- a/src/types.cpp
+++ b/src/types.cpp
@@ -692,6 +692,28 @@ gb_global Type *t_objc_id    = nullptr;
 gb_global Type *t_objc_SEL   = nullptr;
 gb_global Type *t_objc_Class = nullptr;
 
+enum OdinAtomicMemoryOrder : i32 {
+	OdinAtomicMemoryOrder_relaxed = 0, // unordered
+	OdinAtomicMemoryOrder_consume = 1, // monotonic
+	OdinAtomicMemoryOrder_acquire = 2,
+	OdinAtomicMemoryOrder_release = 3,
+	OdinAtomicMemoryOrder_acq_rel = 4,
+	OdinAtomicMemoryOrder_seq_cst = 5,
+	OdinAtomicMemoryOrder_COUNT,
+};
+
+char const *OdinAtomicMemoryOrder_strings[OdinAtomicMemoryOrder_COUNT] = {
+	"Relaxed",
+	"Consume",
+	"Acquire",
+	"Release",
+	"Acq_Rel",
+	"Seq_Cst",
+};
+
+gb_global Type *t_atomic_memory_order = nullptr;
+
+
 
 
 gb_global RecursiveMutex g_type_mutex;