CogVM source as per VMMaker.oscog-eem.3722

Fix the underlying cause of the failure of partitionedMul:with:nBits:wordBits:
on 32-bits (a shift requires an explicit cast to avoid truncation).

Revert hacks.

Author: eliotmiranda

building/win32x86/common/Makefile.msvc.flags

@@ -42,7 +42,6 @@ undefine MSVC
 export TOOLPREFIX
 $(info CC=$(CC))
 AS:=as --32
-XCFLAGS+=-DRGBMul32BitFallBack=1
 
 $(info CC=$(CC))
 

src/plugins/BitBltPlugin/BitBltPlugin.c

@@ -1,10 +1,10 @@
 /* Automatically generated by
-	SmartSyntaxPluginCodeGenerator VMMaker.oscog-eem.3721 uuid: f3895c70-5b57-4684-92f6-72aee920c70e
+	SmartSyntaxPluginCodeGenerator VMMaker.oscog-eem.3722 uuid: 5cb8d220-2e6d-4207-9f4f-2f3a99b4c5e1
 	(Compiler-eem.527)
    from
-	BitBltSimulation VMMaker.oscog-eem.3721 uuid: f3895c70-5b57-4684-92f6-72aee920c70e
+	BitBltSimulation VMMaker.oscog-eem.3722 uuid: 5cb8d220-2e6d-4207-9f4f-2f3a99b4c5e1
  */
-static char __buildInfo[] = "BitBltSimulation VMMaker.oscog-eem.3721 uuid: f3895c70-5b57-4684-92f6-72aee920c70e " __DATE__ ;
+static char __buildInfo[] = "BitBltSimulation VMMaker.oscog-eem.3722 uuid: 5cb8d220-2e6d-4207-9f4f-2f3a99b4c5e1 " __DATE__ ;
 
 
 #include "config.h"
@@ -132,9 +132,6 @@ static unsigned int partitionedAddtonBitscomponentMaskcarryOverflowMask(unsigned
 static unsigned int partitionedANDtonBitsnPartitions(unsigned int word1, unsigned int word2, sqInt nBits, sqInt nParts);
 static unsigned int partitionedMaxwithnBitsnPartitions(unsigned int word1, unsigned int word2, sqInt nBits, sqInt nParts);
 static unsigned int partitionedMinwithnBitsnPartitions(unsigned int word1, unsigned int word2, sqInt nBits, sqInt nParts);
-#if RGBMul32BitFallBack
-static unsigned int partitionedMulwithnBitsnPartitions(unsigned int word1, unsigned int word2, sqInt nBits, sqInt nParts);
-#endif /* RGBMul32BitFallBack */
 static unsigned int partitionedMulwithnBitswordBits(unsigned int word1, unsigned int word2, sqInt nBits, sqInt wordBits);
 static unsigned int partitionedSubfromnBitsnPartitions(unsigned int word1, unsigned int word2, sqInt nBits, sqInt nParts);
 static unsigned int pixClearwith(unsigned int sourceWord, unsigned int destinationWord);
@@ -349,7 +346,7 @@ static int maskTable[33] = {
 0, 1, 3, 0, 15, 31, 0, 0, 255, 0, 0, 0, 0, 0, 0, 0, 65535,
 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, -1
 };
-static const char *moduleName = "BitBltPlugin VMMaker.oscog-eem.3721 " INT_EXT;
+static const char *moduleName = "BitBltPlugin VMMaker.oscog-eem.3722 " INT_EXT;
 static sqInt noHalftone;
 static sqInt noSource;
 static sqInt numGCsOnInvocation;
@@ -4068,52 +4065,6 @@ partitionedMinwithnBitsnPartitions(unsigned int word1, unsigned int word2, sqInt
 }
 
 
-/*	Multiply word1 with word2 as nParts partitions of nBits each.
-	This is useful for packed pixels, or packed colors.
-	Bug in loop version when non-white background */
-/*	In C, integer multiplication might answer a wrong value if the unsigned
-	values are declared as signed.
-	This problem does not affect this method, because the most significant bit
-	(i.e. the sign bit) will
-	always be zero (jmv)
- */
-
-	/* BitBltSimulation>>#partitionedMul:with:nBits:nPartitions: */
-#if RGBMul32BitFallBack
-static unsigned int
-partitionedMulwithnBitsnPartitions(unsigned int word1, unsigned int word2, sqInt nBits, sqInt nParts)
-{
-	unsigned int dMask;
-	unsigned int product;
-	unsigned int result;
-	unsigned int sMask;
-
-	/* partition mask starts at the right */
-	sMask = maskTable[nBits];
-	dMask = (((usqInt)(sMask) << nBits));
-
-	/* optimized first step */
-	result = ((usqInt)((((((word1 & sMask) + 1) * ((word2 & sMask) + 1)) - 1) & dMask))) >> nBits;
-	if (nParts == 1) {
-		return result;
-	}
-	product = (((((((usqInt)(word1)) >> nBits) & sMask) + 1) * (((((usqInt)(word2)) >> nBits) & sMask) + 1)) - 1) & dMask;
-	result = result | product;
-	if (nParts == 2) {
-		return result;
-	}
-	product = (((((((usqInt)(word1)) >> (2 * nBits)) & sMask) + 1) * (((((usqInt)(word2)) >> (2 * nBits)) & sMask) + 1)) - 1) & dMask;
-	result = result | ((((usqInt)(product) << nBits)));
-	if (nParts == 3) {
-		return result;
-	}
-	product = (((((((usqInt)(word1)) >> (3 * nBits)) & sMask) + 1) * (((((usqInt)(word2)) >> (3 * nBits)) & sMask) + 1)) - 1) & dMask;
-	result = result | ((((usqInt)(product) << (2 * nBits))));
-	return result;
-}
-#endif /* RGBMul32BitFallBack */
-
-
 /*	Multiply each channel of nBits in word1 and word2.
 	We assume that for each channel of nBits, we multiply ratios in interval
 	[0..1], scaled by (1 << nBits - 1).
@@ -4177,8 +4128,10 @@ partitionedMulwithnBitswordBits(unsigned int word1, unsigned int word2, sqInt nB
 	shift = 0;
 	doubleWordMul = 0;
 	for (i = 1; i <= (wordBits / nBits); i += 1) {
+		/* N.B. we must explicitly cast the expression to be shifted to 8 bytes, otherwise Slang will generate a 32-bit shift, truncating the result and hence breaking the code. */
+
 		/* multiply each channel of the two operands */
-		doubleWordMul += (((usqInt)((((((doubleWord1) >> shift) & channelMask) * (((doubleWord2) >> shift) & channelMask)) + half)) << shift));
+		doubleWordMul = ((((unsigned long long) (doubleWordMul + (((((doubleWord1) >> shift) & channelMask) * (((doubleWord2) >> shift) & channelMask)) + half)))) << shift);
 		shift += n2;
 	}
 
@@ -5926,21 +5879,6 @@ rgbMinwith(unsigned int sourceWord, unsigned int destinationWord)
 static unsigned int
 rgbMulwith(unsigned int sourceWord, unsigned int destinationWord)
 {
-
-#  if RGBMul32BitFallBack
-	if (destDepth < 16) {
-		/* Mul each pixel separately */
-		return partitionedMulwithnBitsnPartitions(sourceWord, destinationWord, destDepth, destPPW);
-	}
-	if (destDepth == 16) {
-		/* Mul RGB components of each pixel separately */
-		return (partitionedMulwithnBitsnPartitions(sourceWord, destinationWord, 5, 3)) + ((((usqInt)((partitionedMulwithnBitsnPartitions(((usqInt)(sourceWord)) >> 16, ((usqInt)(destinationWord)) >> 16, 5, 3))) << 16)));
-	}
-	else {
-		/* Mul RGBA components of the pixel separately */
-		return partitionedMulwithnBitsnPartitions(sourceWord, destinationWord, 8, 4);
-	}
-#  else // RGBMul32BitFallBack
 	if (destDepth < 16) {
 		/* Mul each pixel separately */
 		if (destDepth == 1) {
@@ -5956,7 +5894,6 @@ rgbMulwith(unsigned int sourceWord, unsigned int destinationWord)
 		/* Mul RGBA components of the pixel separately */
 		return partitionedMulwithnBitswordBits(sourceWord, destinationWord, 8, 32);
 	}
-#  endif // RGBMul32BitFallBack
 }
 
 	/* BitBltSimulation>>#rgbSub:with: */