If-conversion RFC 4 - x86 only (for now), use cmovXX

Replacement of tiny conditional jump graphlets with
conditional move instructions.

Currently enabled only for x86. Arm64 support using cselXX
will be essentially identical.

Adds (internal) frontend sel0/sel1 ops with flag-specific
backend xselXX following jnz implementation pattern.

Testing: standard QBE, cproc, harec, hare, roland

1 files changed, 238 insertions, 0 deletions

diff --git a/test/ifc.ssa b/test/ifc.ssa
new file mode 100644
index 0000000..29f4457
--- /dev/null
+++ b/test/ifc.ssa
@@ -0,0 +1,238 @@
+export
+function l $ifc1(l %v0, l %v1, w %c) {
+@start
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+export
+function l $ifc2(l %v0, l %v1, w %p) {
+@start
+	%c =w cnew %p, 42
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+export
+function l $ifc3(l %v0, l %v1, w %p) {
+@start
+	%c =w cugtw %p, 42
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+export
+function l $ifclts(s %s0, s %s1, l %v0, l %v1) {
+@start
+	%c =w clts %s0, %s1
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+export
+function l $ifcles(s %s0, s %s1, l %v0, l %v1) {
+@start
+	%c =w cles %s0, %s1
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+export
+function l $ifcgts(s %s0, s %s1, l %v0, l %v1) {
+@start
+	%c =w cgts %s0, %s1
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+export
+function l $ifcges(s %s0, s %s1, l %v0, l %v1) {
+@start
+	%c =w cges %s0, %s1
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+export
+function l $ifceqs(s %s0, s %s1, l %v0, l %v1) {
+@start
+	%c =w ceqs %s0, %s1
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+export
+function l $ifcnes(s %s0, s %s1, l %v0, l %v1) {
+@start
+	%c =w cnes %s0, %s1
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+export
+function l $ifcos(s %s0, s %s1, l %v0, l %v1) {
+@start
+	%c =w cos %s0, %s1
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+export
+function l $ifcuos(s %s0, s %s1, l %v0, l %v1) {
+@start
+	%c =w cuos %s0, %s1
+	jnz %c, @true, @false
+@true
+	%v =l copy %v1
+	jmp @end
+@false
+	%v =l copy %v0
+	jmp @end
+@end
+	ret %v
+}
+
+# >>> driver
+# extern long ifc1(long, long, int);
+# extern long ifc2(long, long, int);
+# extern long ifc3(long, long, int);
+# extern long ifclts(float, float, long, long);
+# extern long ifcles(float, float, long, long);
+# extern long ifcgts(float, float, long, long);
+# extern long ifcges(float, float, long, long);
+# extern long ifceqs(float, float, long, long);
+# extern long ifcnes(float, float, long, long);
+# extern long ifcos(float, float, long, long);
+# extern long ifcuos(float, float, long, long);
+# int main() {
+# 	return
+# 		ifc1(7, 5, 0) != 7
+# 		|| ifc1(7, 5, 1) != 5
+# 		|| ifc1(7, 5, 33) != 5
+# 		|| ifc2(7, 5, 42) != 7
+# 		|| ifc2(7, 5, 41) != 5
+# 		|| ifc2(7, 5, 43) != 5
+# 		|| ifc3(7, 5, 42) != 7
+# 		|| ifc3(7, 5, 41) != 7
+# 		|| ifc3(7, 5, 43) != 5
+# 		|| ifclts(5.0f, 6.0f, 7, 5) != 5
+# 		|| ifclts(5.0f, 5.0f, 7, 5) != 7
+# 		|| ifclts(5.0f, 4.0f, 7, 5) != 7
+# 		|| ifclts(5.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifclts(0.0f/0.0f, 5.0f, 7, 5) != 7
+# 		|| ifclts(0.0f/0.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifcles(5.0f, 6.0f, 7, 5) != 5
+# 		|| ifcles(5.0f, 5.0f, 7, 5) != 5
+# 		|| ifcles(5.0f, 4.0f, 7, 5) != 7
+# 		|| ifcles(5.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifcles(0.0f/0.0f, 5.0f, 7, 5) != 7
+# 		|| ifcles(0.0f/0.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifcgts(5.0f, 6.0f, 7, 5) != 7
+# 		|| ifcgts(5.0f, 5.0f, 7, 5) != 7
+# 		|| ifcgts(5.0f, 4.0f, 7, 5) != 5
+# 		|| ifcgts(5.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifcgts(0.0f/0.0f, 5.0f, 7, 5) != 7
+# 		|| ifcgts(0.0f/0.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifcges(5.0f, 6.0f, 7, 5) != 7
+# 		|| ifcges(5.0f, 5.0f, 7, 5) != 5
+# 		|| ifcges(5.0f, 4.0f, 7, 5) != 5
+# 		|| ifcges(5.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifcges(0.0f/0.0f, 5.0f, 7, 5) != 7
+# 		|| ifcges(0.0f/0.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifceqs(5.0f, 6.0f, 7, 5) != 7
+# 		|| ifceqs(5.0f, 5.0f, 7, 5) != 5
+# 		|| ifceqs(5.0f, 4.0f, 7, 5) != 7
+# 		|| ifceqs(5.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifceqs(0.0f/0.0f, 5.0f, 7, 5) != 7
+# 		|| ifceqs(0.0f/0.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifcnes(5.0f, 6.0f, 7, 5) != 5
+# 		|| ifcnes(5.0f, 5.0f, 7, 5) != 7
+# 		|| ifcnes(5.0f, 4.0f, 7, 5) != 5
+# 		|| ifcnes(5.0f, 0.0f/0.0f, 7, 5) != 5
+# 		|| ifcnes(0.0f/0.0f, 5.0f, 7, 5) != 5
+# 		|| ifcnes(0.0f/0.0f, 0.0f/0.0f, 7, 5) != 5
+# 		|| ifcos(5.0f, 6.0f, 7, 5) != 5
+# 		|| ifcos(5.0f, 5.0f, 7, 5) != 5
+# 		|| ifcos(5.0f, 4.0f, 7, 5) != 5
+# 		|| ifcos(5.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifcos(0.0f/0.0f, 5.0f, 7, 5) != 7
+# 		|| ifcos(0.0f/0.0f, 0.0f/0.0f, 7, 5) != 7
+# 		|| ifcuos(5.0f, 6.0f, 7, 5) != 7
+# 		|| ifcuos(5.0f, 5.0f, 7, 5) != 7
+# 		|| ifcuos(5.0f, 4.0f, 7, 5) != 7
+# 		|| ifcuos(5.0f, 0.0f/0.0f, 7, 5) != 5
+# 		|| ifcuos(0.0f/0.0f, 5.0f, 7, 5) != 5
+# 		|| ifcuos(0.0f/0.0f, 0.0f/0.0f, 7, 5) != 5
+# 		;
+# }
+# <<<