(svn r26247) -Fix [FS#5854, FS#5855]: Possible out of bounds reads with the sse blitters (MJP)

6 files changed, 164 insertions, 208 deletions

diff --git a/src/blitter/32bpp_anim_sse4.cpp b/src/blitter/32bpp_anim_sse4.cpp
index a200d8d8c..5bc63bbb0 100644
--- a/src/blitter/32bpp_anim_sse4.cpp
+++ b/src/blitter/32bpp_anim_sse4.cpp
@@ -80,10 +80,11 @@ inline void Blitter_32bppSSE4_Anim::Draw(const Blitter::BlitterParams *bp, ZoomL
 					}
 
 					case RM_WITH_SKIP: {
-						uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
-						__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-						__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
 						for (uint x = (uint) effective_width/2; x != 0; x--) {
+							uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
+							__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+							__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
+
 							/* Remap colours. */
 							const byte m0 = mvX2;
 							if (m0 >= PALETTE_ANIM_START) {
@@ -125,42 +126,32 @@ inline void Blitter_32bppSSE4_Anim::Draw(const Blitter::BlitterParams *bp, ZoomL
 bmno_alpha_blend:
 							ALPHA_BLEND_2(pack_low_cm);
 bmno_full_opacity:
-							srcABCD = _mm_blend_epi16(srcABCD, dstABCD, 0xF0);
-
-							src_mv += 2;
-							mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
-							src += 2;
-							anim += 2;
-							dstABCD = _mm_loadu_si128((__m128i*) (dst+2));
-							_mm_storeu_si128((__m128i *) dst, srcABCD);
-							srcABCD = _mm_loadu_si128((const __m128i*) src);
-							dst += 2;
-							continue;
-
+							_mm_storel_epi64((__m128i *) dst, srcABCD);
 bmno_full_transparency:
 							src_mv += 2;
-							mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
-							dst += 2;
 							src += 2;
 							anim += 2;
-							dstABCD = _mm_loadu_si128((__m128i*) dst);
-							srcABCD = _mm_loadu_si128((const __m128i*) src);
+							dst += 2;
 						}
 
 						if (bt_last == BT_ODD) {
 							if (src->a == 0) {
 							} else if (src->a == 255) {
-								*anim = (uint16) mvX2;
-								*dst = ((byte) mvX2 >= PALETTE_ANIM_START) ? AdjustBrightness(LookupColourInPalette((byte) mvX2), (byte) (mvX2 >> 8)) : *src;
+								*anim = *(const uint16*) src_mv;
+								*dst = (src_mv->m >= PALETTE_ANIM_START) ? AdjustBrightness(LookupColourInPalette(src_mv->m), src_mv->v) : *src;
 							} else {
 								*anim = 0;
-								if ((byte) mvX2 >= PALETTE_ANIM_START) {
-									ALIGN(16) Colour colour = AdjustBrightness(LookupColourInPalette((byte) mvX2), (byte) (mvX2 >> 8));
+								__m128i srcABCD;
+								__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
+								if (src_mv->m >= PALETTE_ANIM_START) {
+									Colour colour = AdjustBrightness(LookupColourInPalette(src_mv->m), src_mv->v);
 									colour.a = src->a;
-									srcABCD = _mm_load_si128((__m128i*) &colour);
+									srcABCD = _mm_cvtsi32_si128(colour.data);
+								} else {
+									srcABCD = _mm_cvtsi32_si128(src->data);
 								}
 								ALPHA_BLEND_2(pack_low_cm);
-								(*dst).data = EXTR32(srcABCD, 0);
+								dst->data = _mm_cvtsi128_si32(srcABCD);
 							}
 						}
 						break;
@@ -181,18 +172,18 @@ bmno_full_transparency:
 						const int width_diff = si->sprite_width - bp->width;
 						effective_width = bp->width - (int) src_rgba_line[0].data;
 						const int delta_diff = (int) src_rgba_line[1].data - width_diff;
-						const int nd = effective_width - delta_diff;
-						effective_width = delta_diff > 0 ? nd : effective_width;
+						const int new_width = effective_width - delta_diff;
+						effective_width = delta_diff > 0 ? new_width : effective_width;
 						if (effective_width <= 0) break;
 						/* FALLTHROUGH */
 					}
 
 					case RM_WITH_SKIP: {
-						uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
-						__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-						__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
-
 						for (uint x = (uint) effective_width / 2; x != 0; x--) {
+							uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
+							__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+							__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
+
 							/* Remap colours. */
 							const uint m0 = (byte) mvX2;
 							const uint r0 = remap[m0];
@@ -250,53 +241,40 @@ bmno_full_transparency:
 bmcr_alpha_blend:
 							ALPHA_BLEND_2(pack_low_cm);
 bmcr_full_opacity:
-							srcABCD = _mm_blend_epi16(srcABCD, dstABCD, 0xF0);
-
-							src += 2;
-							src_mv += 2;
-							anim += 2;
-							mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
-							dstABCD = _mm_loadu_si128((__m128i*) (dst+2));
-							_mm_storeu_si128((__m128i *) dst, srcABCD);
-							srcABCD = _mm_loadu_si128((const __m128i*) src);
-							dst += 2;
-							continue;
-
+							_mm_storel_epi64((__m128i *) dst, srcABCD);
 bmcr_full_transparency:
 							src_mv += 2;
-							mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
 							dst += 2;
 							src += 2;
 							anim += 2;
-							dstABCD = _mm_loadu_si128((__m128i*) dst);
-							srcABCD = _mm_loadu_si128((const __m128i*) src);
 						}
 
 						if (effective_width & 1) {
 							/* In case the m-channel is zero, do not remap this pixel in any way. */
-							if (src->a == 0) {
-							} else if ((byte) mvX2 != 0) {
-								const uint r = remap[(byte) mvX2];
-								*anim = (src->a == 255) ? (r | ((uint16) mvX2 & 0xFF00)) : 0;
+							__m128i srcABCD;
+							if (src->a == 0) break;
+							if (src_mv->m) {
+								const uint r = remap[src_mv->m];
+								*anim = (src->a == 255) ? r | ((uint16) src_mv->v << 8 ) : 0;
 								if (r != 0) {
-									Colour remapped_colour = AdjustBrightness(LookupColourInPalette(r), (byte) (mvX2 >> 8));
+									Colour remapped_colour = AdjustBrightness(this->LookupColourInPalette(r), src_mv->v);
 									if (src->a == 255) {
 										*dst = remapped_colour;
 									} else {
 										remapped_colour.a = src->a;
-										INSR32(remapped_colour.data, srcABCD, 0);
+										srcABCD = _mm_cvtsi32_si128(remapped_colour.data);
 										goto bmcr_alpha_blend_single;
 									}
 								}
 							} else {
 								*anim = 0;
-								if (src->a == 255) {
-									*dst = *src;
-								} else {
+								srcABCD = _mm_cvtsi32_si128(src->data);
+								if (src->a < 255) {
 bmcr_alpha_blend_single:
+									__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
 									ALPHA_BLEND_2(pack_low_cm);
-									(*dst).data = EXTR32(srcABCD, 0);
 								}
+								dst->data = _mm_cvtsi128_si32(srcABCD);
 							}
 						}
 						break;
@@ -309,29 +287,27 @@ bmcr_alpha_blend_single:
 
 			case BM_TRANSPARENT: {
 				/* Make the current colour a bit more black, so it looks like this image is transparent. */
-				__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-				__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
 				for (uint x = (uint) bp->width / 2; x > 0; x--) {
+					__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+					__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
 					__m128i srcAB = _mm_unpacklo_epi8(srcABCD, _mm_setzero_si128());
 					__m128i dstAB = _mm_unpacklo_epi8(dstABCD, _mm_setzero_si128());
-					__m128i dstCD = _mm_unpackhi_epi8(dstABCD, _mm_setzero_si128());
 					__m128i alphaAB = _mm_shuffle_epi8(srcAB, a_cm);
 					alphaAB = _mm_srli_epi16(alphaAB, 2); // Reduce to 64 levels of shades so the max value fits in 16 bits.
 					__m128i nom = _mm_sub_epi16(tr_nom_base, alphaAB);
 					dstAB = _mm_mullo_epi16(dstAB, nom);
 					dstAB = _mm_srli_epi16(dstAB, 8);
-					dstAB = _mm_packus_epi16(dstAB, dstCD);
-					Colour *old_dst = dst;
+					dstAB = _mm_packus_epi16(dstAB, dstAB);
+					_mm_storel_epi64((__m128i *) dst, dstAB);
 					src += 2;
 					dst += 2;
 					anim += 2;
-					dstABCD = _mm_loadu_si128((__m128i*) dst);
-					_mm_storeu_si128((__m128i *) old_dst, dstAB);
-					srcABCD = _mm_loadu_si128((const __m128i*) src);
 					if (src[-2].a) anim[-2] = 0;
 					if (src[-1].a) anim[-1] = 0;
 				}
 				if (bp->width & 1) {
+					__m128i srcABCD = _mm_cvtsi32_si128(src->data);
+					__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
 					__m128i srcAB = _mm_unpacklo_epi8(srcABCD, _mm_setzero_si128());
 					__m128i dstAB = _mm_unpacklo_epi8(dstABCD, _mm_setzero_si128());
 					__m128i alphaAB = _mm_shuffle_epi8(srcAB, a_cm);
@@ -340,7 +316,7 @@ bmcr_alpha_blend_single:
 					dstAB = _mm_mullo_epi16(dstAB, nom);
 					dstAB = _mm_srli_epi16(dstAB, 8);
 					dstAB = _mm_packus_epi16(dstAB, dstAB);
-					(*dst).data = EXTR32(dstAB, 0);
+					dst->data = _mm_cvtsi128_si32(dstAB);
 					if (src[0].a) anim[0] = 0;
 				}
 				break;
diff --git a/src/blitter/32bpp_sse2.cpp b/src/blitter/32bpp_sse2.cpp
index 137c91417..c5e7e70d0 100644
--- a/src/blitter/32bpp_sse2.cpp
+++ b/src/blitter/32bpp_sse2.cpp
@@ -70,18 +70,18 @@ inline void Blitter_32bppSSE2::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 
 					case RM_WITH_SKIP: {
 						for (uint x = (uint) effective_width / 2; x > 0; x--) {
-							__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-							__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
+							__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+							__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
 							ALPHA_BLEND_2();
-							*(uint64*) dst = EXTR64(srcABCD, 0);
+							_mm_storel_epi64((__m128i*) dst, srcABCD);
 							src += 2;
 							dst += 2;
 						}
 						if (bt_last == BT_ODD) {
-							__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-							__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
+							__m128i srcABCD = _mm_cvtsi32_si128(src->data);
+							__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
 							ALPHA_BLEND_2();
-							(*dst).data = EXTR32(srcABCD, 0);
+							dst->data = _mm_cvtsi128_si32(srcABCD);
 						}
 						break;
 					}
@@ -99,8 +99,8 @@ inline void Blitter_32bppSSE2::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 						const int width_diff = si->sprite_width - bp->width;
 						effective_width = bp->width - (int) src_rgba_line[0].data;
 						const int delta_diff = (int) src_rgba_line[1].data - width_diff;
-						const int nd = effective_width - delta_diff;
-						effective_width = delta_diff > 0 ? nd : effective_width;
+						const int new_width = effective_width - delta_diff;
+						effective_width = delta_diff > 0 ? new_width : effective_width;
 						if (effective_width <= 0) break;
 						/* FALLTHROUGH */
 					}
@@ -108,30 +108,28 @@ inline void Blitter_32bppSSE2::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 					case RM_WITH_SKIP: {
 						const byte *remap = bp->remap;
 						for (uint x = (uint) effective_width; x != 0; x--) {
-							/* In case the m-channel is zero, do not remap this pixel in any way */
-							if (src_mv->m == 0) {
-								if (src->a < 255) {
-									__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-									__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
-									ALPHA_BLEND_2();
-									(*dst).data = EXTR32(srcABCD, 0);
-								} else {
-									*dst = src->data;
-								}
-							} else {
+							/* In case the m-channel is zero, do not remap this pixel in any way. */
+							__m128i srcABCD;
+							if (src_mv->m) {
 								const uint r = remap[src_mv->m];
 								if (r != 0) {
 									Colour remapped_colour = AdjustBrightness(this->LookupColourInPalette(r), src_mv->v);
-									if (src->a < 255) {
-										__m128i srcABCD;
-										__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
-										remapped_colour.a = src->a;
-										INSR32(remapped_colour.data, srcABCD, 0);
-										ALPHA_BLEND_2();
-										(*dst).data = EXTR32(srcABCD, 0);
-									} else
+									if (src->a == 255) {
 										*dst = remapped_colour;
+									} else {
+										remapped_colour.a = src->a;
+										srcABCD = _mm_cvtsi32_si128(remapped_colour.data);
+										goto bmcr_alpha_blend_single;
+									}
+								}
+							} else {
+								srcABCD = _mm_cvtsi32_si128(src->data);
+								if (src->a < 255) {
+bmcr_alpha_blend_single:
+									__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
+									ALPHA_BLEND_2();
 								}
+								dst->data = _mm_cvtsi128_si32(srcABCD);
 							}
 							src_mv++;
 							dst++;
@@ -149,27 +147,25 @@ inline void Blitter_32bppSSE2::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 				/* Make the current colour a bit more black, so it looks like this image is transparent.
 				 * rgb = rgb * ((256/4) * 4 - (alpha/4)) / ((256/4) * 4)
 				 */
-				__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-				__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
 				for (uint x = (uint) bp->width / 2; x > 0; x--) {
+					__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+					__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
 					__m128i srcAB = _mm_unpacklo_epi8(srcABCD, _mm_setzero_si128());
 					__m128i dstAB = _mm_unpacklo_epi8(dstABCD, _mm_setzero_si128());
-					__m128i dstCD = _mm_unpackhi_epi8(dstABCD, _mm_setzero_si128());
 					__m128i alphaAB = _mm_shufflelo_epi16(srcAB, 0x3F);
 					alphaAB = _mm_shufflehi_epi16(alphaAB, 0x3F);
 					alphaAB = _mm_srli_epi16(alphaAB, 2); // Reduce to 64 levels of shades so the max value fits in 16 bits.
 					__m128i nom = _mm_sub_epi16(tr_nom_base, alphaAB);
 					dstAB = _mm_mullo_epi16(dstAB, nom);
 					dstAB = _mm_srli_epi16(dstAB, 8);
-					dstAB = _mm_packus_epi16(dstAB, dstCD);
-					Colour *old_dst = dst;
+					dstAB = _mm_packus_epi16(dstAB, dstAB);
+					_mm_storel_epi64((__m128i *) dst, dstAB);
 					src += 2;
 					dst += 2;
-					dstABCD = _mm_loadu_si128((__m128i*) dst);
-					_mm_storeu_si128((__m128i *) old_dst, dstAB);
-					srcABCD = _mm_loadu_si128((const __m128i*) src);
 				}
 				if (bp->width & 1) {
+					__m128i srcABCD = _mm_cvtsi32_si128(src->data);
+					__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
 					__m128i srcAB = _mm_unpacklo_epi8(srcABCD, _mm_setzero_si128());
 					__m128i dstAB = _mm_unpacklo_epi8(dstABCD, _mm_setzero_si128());
 					__m128i alphaAB = _mm_shufflelo_epi16(srcAB, 0x3F);
@@ -179,7 +175,7 @@ inline void Blitter_32bppSSE2::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 					dstAB = _mm_mullo_epi16(dstAB, nom);
 					dstAB = _mm_srli_epi16(dstAB, 8);
 					dstAB = _mm_packus_epi16(dstAB, dstAB);
-					(*dst).data = EXTR32(dstAB, 0);
+					dst->data = _mm_cvtsi128_si32(dstAB);
 				}
 				break;
 			}
@@ -345,7 +341,7 @@ inline Colour Blitter_32bppSSE2::AdjustBrightness(Colour colour, uint8 brightnes
 IGNORE_UNINITIALIZED_WARNING_START
 /* static */ Colour Blitter_32bppSSE2::ReallyAdjustBrightness(Colour colour, uint8 brightness)
 {
-	ALIGN(16) uint64 c16 = colour.b | (uint64) colour.g << 16 | (uint64) colour.r << 32;
+	uint64 c16 = colour.b | (uint64) colour.g << 16 | (uint64) colour.r << 32;
 	c16 *= brightness;
 	uint64 c16_ob = c16; // Helps out of order execution.
 	c16 /= DEFAULT_BRIGHTNESS;
@@ -357,12 +353,20 @@ IGNORE_UNINITIALIZED_WARNING_START
 
 	const uint32 alpha32 = colour.data & 0xFF000000;
 	__m128i ret;
+#ifdef _SQ64
+	ret = _mm_cvtsi64_si128(c16);
+#else
 	INSR64(c16, ret, 0);
+#endif
 	if (ob != 0) {
 		/* Reduce overbright strength. */
 		ob /= 2;
 		__m128i ob128;
+#ifdef _SQ64
+		ob128 = _mm_cvtsi64_si128(ob | ob << 16 | ob << 32);
+#else
 		INSR64(ob | ob << 16 | ob << 32, ob128, 0);
+#endif
 		__m128i white = OVERBRIGHT_VALUE_MASK;
 		__m128i c128 = ret;
 		ret = _mm_subs_epu16(white, c128); /* PSUBUSW,   (255 - rgb) */
@@ -372,7 +376,7 @@ IGNORE_UNINITIALIZED_WARNING_START
 	}
 
 	ret = _mm_packus_epi16(ret, ret);      /* PACKUSWB, saturate and pack. */
-	return alpha32 | EXTR32(ret, 0);
+	return alpha32 | _mm_cvtsi128_si32(ret);
 }
 IGNORE_UNINITIALIZED_WARNING_STOP
 
diff --git a/src/blitter/32bpp_sse4.cpp b/src/blitter/32bpp_sse4.cpp
index 357a587d9..c00dab0d7 100644
--- a/src/blitter/32bpp_sse4.cpp
+++ b/src/blitter/32bpp_sse4.cpp
@@ -76,24 +76,19 @@ inline void Blitter_32bppSSE4::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 					}
 
 					case RM_WITH_SKIP: {
-						__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-						__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
 						for (uint x = (uint) effective_width / 2; x > 0; x--) {
+							__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+							__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
 							ALPHA_BLEND_2(pack_low_cm);
-							srcABCD = _mm_blend_epi16(srcABCD, dstABCD, 0xF0);
-							Colour *old_dst = dst;
+							_mm_storel_epi64((__m128i*) dst, srcABCD);
 							src += 2;
 							dst += 2;
-							/* It is VERY important to read next data before it gets invalidated in cpu cache.
-							 * And PEXTR latency is a real problem here.
-							 */
-							dstABCD = _mm_loadu_si128((__m128i*) dst);
-							_mm_storeu_si128((__m128i *) old_dst, srcABCD);
-							srcABCD = _mm_loadu_si128((const __m128i*) src);
 						}
 						if (bt_last == BT_ODD) {
+							__m128i srcABCD = _mm_cvtsi32_si128(src->data);
+							__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
 							ALPHA_BLEND_2(pack_low_cm);
-							*dst = (Colour) EXTR32(srcABCD, 0);
+							dst->data = _mm_cvtsi128_si32(srcABCD);
 						}
 						break;
 					}
@@ -112,18 +107,18 @@ inline void Blitter_32bppSSE4::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 						const int width_diff = si->sprite_width - bp->width;
 						effective_width = bp->width - (int) src_rgba_line[0].data;
 						const int delta_diff = (int) src_rgba_line[1].data - width_diff;
-						const int nd = effective_width - delta_diff;
-						effective_width = delta_diff > 0 ? nd : effective_width;
+						const int new_width = effective_width - delta_diff;
+						effective_width = delta_diff > 0 ? new_width : effective_width;
 						if (effective_width <= 0) break;
 						/* FALLTHROUGH */
 					}
 
 					case RM_WITH_SKIP: {
-						__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-						__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
-						uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
-
 						for (uint x = (uint) effective_width / 2; x > 0; x--) {
+							__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+							__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
+							uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
+
 							/* Remap colours. */
 							if (mvX2 & 0x00FF00FF) {
 								/* Written so the compiler uses CMOV. */
@@ -152,38 +147,35 @@ inline void Blitter_32bppSSE4::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 
 							/* Blend colours. */
 							ALPHA_BLEND_2(pack_low_cm);
-							srcABCD = _mm_blend_epi16(srcABCD, dstABCD, 0xF0);
-							Colour *old_dst = dst;
+							_mm_storel_epi64((__m128i *) dst, srcABCD);
 							dst += 2;
 							src += 2;
 							src_mv += 2;
-							dstABCD = _mm_loadu_si128((__m128i*) dst);
-							_mm_storeu_si128((__m128i *) old_dst, srcABCD);
-							mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
-							srcABCD = _mm_loadu_si128((const __m128i*) src);
 						}
 
 						if (effective_width & 1) {
 							/* In case the m-channel is zero, do not remap this pixel in any way. */
-							if ((byte) mvX2 == 0) {
-								if (src->a < 255) {
-									ALPHA_BLEND_2(pack_low_cm);
-									(*dst).data = EXTR32(srcABCD, 0);
-								} else
-									*dst = *src;
-							} else {
-								const uint r = remap[(byte) mvX2];
+							__m128i srcABCD;
+							if (src_mv->m) {
+								const uint r = remap[src_mv->m];
 								if (r != 0) {
-									Colour remapped_colour = AdjustBrightness(this->LookupColourInPalette(r), (byte) (mvX2 >> 8));
+									Colour remapped_colour = AdjustBrightness(this->LookupColourInPalette(r), src_mv->v);
 									if (src->a == 255) {
 										*dst = remapped_colour;
 									} else {
 										remapped_colour.a = src->a;
-										INSR32(remapped_colour.data, srcABCD, 0);
-										ALPHA_BLEND_2(pack_low_cm);
-										(*dst).data = EXTR32(srcABCD, 0);
+										srcABCD = _mm_cvtsi32_si128(remapped_colour.data);
+										goto bmcr_alpha_blend_single;
 									}
 								}
+							} else {
+								srcABCD = _mm_cvtsi32_si128(src->data);
+								if (src->a < 255) {
+bmcr_alpha_blend_single:
+									__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
+									ALPHA_BLEND_2(pack_low_cm);
+								}
+								dst->data = _mm_cvtsi128_si32(srcABCD);
 							}
 						}
 						break;
@@ -199,26 +191,24 @@ inline void Blitter_32bppSSE4::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 				/* Make the current colour a bit more black, so it looks like this image is transparent.
 				 * rgb = rgb * ((256/4) * 4 - (alpha/4)) / ((256/4) * 4)
 				 */
-				__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-				__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
 				for (uint x = (uint) bp->width / 2; x > 0; x--) {
+					__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+					__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
 					__m128i srcAB = _mm_unpacklo_epi8(srcABCD, _mm_setzero_si128());
 					__m128i dstAB = _mm_unpacklo_epi8(dstABCD, _mm_setzero_si128());
-					__m128i dstCD = _mm_unpackhi_epi8(dstABCD, _mm_setzero_si128());
 					__m128i alphaAB = _mm_shuffle_epi8(srcAB, a_cm);
 					alphaAB = _mm_srli_epi16(alphaAB, 2); // Reduce to 64 levels of shades so the max value fits in 16 bits.
 					__m128i nom = _mm_sub_epi16(tr_nom_base, alphaAB);
 					dstAB = _mm_mullo_epi16(dstAB, nom);
 					dstAB = _mm_srli_epi16(dstAB, 8);
-					dstAB = _mm_packus_epi16(dstAB, dstCD);
-					Colour *old_dst = dst;
+					dstAB = _mm_packus_epi16(dstAB, dstAB);
+					_mm_storel_epi64((__m128i *) dst, dstAB);
 					src += 2;
 					dst += 2;
-					dstABCD = _mm_loadu_si128((__m128i*) dst);
-					_mm_storeu_si128((__m128i *) old_dst, dstAB);
-					srcABCD = _mm_loadu_si128((const __m128i*) src);
 				}
 				if (bp->width & 1) {
+					__m128i srcABCD = _mm_cvtsi32_si128(src->data);
+					__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
 					__m128i srcAB = _mm_unpacklo_epi8(srcABCD, _mm_setzero_si128());
 					__m128i dstAB = _mm_unpacklo_epi8(dstABCD, _mm_setzero_si128());
 					__m128i alphaAB = _mm_shuffle_epi8(srcAB, a_cm);
@@ -227,7 +217,7 @@ inline void Blitter_32bppSSE4::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 					dstAB = _mm_mullo_epi16(dstAB, nom);
 					dstAB = _mm_srli_epi16(dstAB, 8);
 					dstAB = _mm_packus_epi16(dstAB, dstAB);
-					(*dst).data = EXTR32(dstAB, 0);
+					dst->data = _mm_cvtsi128_si32(dstAB);
 				}
 
 				break;
@@ -290,7 +280,7 @@ inline Colour Blitter_32bppSSE4::AdjustBrightness(Colour colour, uint8 brightnes
 IGNORE_UNINITIALIZED_WARNING_START
 /* static */ Colour Blitter_32bppSSE4::ReallyAdjustBrightness(Colour colour, uint8 brightness)
 {
-	ALIGN(16) uint64 c16 = colour.b | (uint64) colour.g << 16 | (uint64) colour.r << 32;
+	uint64 c16 = colour.b | (uint64) colour.g << 16 | (uint64) colour.r << 32;
 	c16 *= brightness;
 	uint64 c16_ob = c16; // Helps out of order execution.
 	c16 /= DEFAULT_BRIGHTNESS;
@@ -317,7 +307,7 @@ IGNORE_UNINITIALIZED_WARNING_START
 	}
 
 	ret = _mm_packus_epi16(ret, ret);      /* PACKUSWB, saturate and pack. */
-	return alpha32 | EXTR32(ret, 0);
+	return alpha32 | _mm_cvtsi128_si32(ret);
 }
 IGNORE_UNINITIALIZED_WARNING_STOP
 
diff --git a/src/blitter/32bpp_ssse3.cpp b/src/blitter/32bpp_ssse3.cpp
index 531262162..ebfc3ce6d 100644
--- a/src/blitter/32bpp_ssse3.cpp
+++ b/src/blitter/32bpp_ssse3.cpp
@@ -50,7 +50,7 @@ inline void Blitter_32bppSSSE3::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 
 	/* Load these variables into register before loop. */
 	const __m128i a_cm        = ALPHA_CONTROL_MASK;
-	const __m128i pack_hi_cm  = PACK_HIGH_CONTROL_MASK;
+	const __m128i pack_low_cm = PACK_LOW_CONTROL_MASK;
 	const __m128i briAB_cm    = BRIGHTNESS_LOW_CONTROL_MASK;
 	const __m128i div_cleaner = BRIGHTNESS_DIV_CLEANER;
 	const __m128i ob_check    = OVERBRIGHT_PRESENCE_MASK;
@@ -79,27 +79,19 @@ inline void Blitter_32bppSSSE3::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 					}
 
 					case RM_WITH_SKIP: {
-						__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-						__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
 						for (uint x = (uint) effective_width / 2; x > 0; x--) {
-							ALPHA_BLEND_2(pack_hi_cm);
-							/* With high repack, srcABCD have its 2 blended pixels like: [S0 S1 S2 S3] -> [-- -- BS0 BS1]
-							 * dstABCD shuffled: [D0 D1 D2 D3] -> [D2 D3 D0 D0]
-							 * PALIGNR takes what's in (): [-- -- (BS0 BS1] [D2 D3) D0 D0]
-							 */
-							dstABCD = _mm_shuffle_epi32(dstABCD, 0x0E);
-							srcABCD = _mm_alignr_epi8(dstABCD, srcABCD, 8);
-							Colour *old_dst = dst;
+							__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+							__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
+							ALPHA_BLEND_2(pack_low_cm);
+							_mm_storel_epi64((__m128i*) dst, srcABCD);
 							src += 2;
 							dst += 2;
-							/* It is VERY important to read next data before it gets invalidated in cpu cache. */
-							dstABCD = _mm_loadu_si128((__m128i*) dst);
-							_mm_storeu_si128((__m128i *) old_dst, srcABCD);
-							srcABCD = _mm_loadu_si128((const __m128i*) src);
 						}
 						if (bt_last == BT_ODD) {
-							ALPHA_BLEND_2(pack_hi_cm);
-							(*dst).data = EXTR32(srcABCD, 2);
+							__m128i srcABCD = _mm_cvtsi32_si128(src->data);
+							__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
+							ALPHA_BLEND_2(pack_low_cm);
+							dst->data = _mm_cvtsi128_si32(srcABCD);
 						}
 						break;
 					}
@@ -117,18 +109,18 @@ inline void Blitter_32bppSSSE3::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 						const int width_diff = si->sprite_width - bp->width;
 						effective_width = bp->width - (int) src_rgba_line[0].data;
 						const int delta_diff = (int) src_rgba_line[1].data - width_diff;
-						const int nd = effective_width - delta_diff;
-						effective_width = delta_diff > 0 ? nd : effective_width;
+						const int new_width = effective_width - delta_diff;
+						effective_width = delta_diff > 0 ? new_width : effective_width;
 						if (effective_width <= 0) break;
 						/* FALLTHROUGH */
 					}
 
 					case RM_WITH_SKIP: {
-						__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-						__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
-						uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
-
 						for (uint x = (uint) effective_width / 2; x > 0; x--) {
+							__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+							__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
+							uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
+
 							/* Remap colours. */
 							if (mvX2 & 0x00FF00FF) {
 								/* Written so the compiler uses CMOV. */
@@ -139,7 +131,7 @@ inline void Blitter_32bppSSSE3::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 								Colour c0 = 0; // Use alpha of 0 to keep dst as is.
 								c0 = r0 == 0 ? c0 : c0map;
 								c0 = m0 != 0 ? c0 : src0;
-								INSR32(c0.data, srcABCD, 0);
+								srcABCD = _mm_cvtsi32_si128(c0.data);
 
 								const Colour src1 = src[1];
 								const uint m1 = (byte) (mvX2 >> 16);
@@ -156,40 +148,36 @@ inline void Blitter_32bppSSSE3::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 							}
 
 							/* Blend colours. */
-							ALPHA_BLEND_2(pack_hi_cm);
-							dstABCD = _mm_shuffle_epi32(dstABCD, 0x0E);
-							srcABCD = _mm_alignr_epi8(dstABCD, srcABCD, 8);
-							Colour *old_dst = dst;
+							ALPHA_BLEND_2(pack_low_cm);
+							_mm_storel_epi64((__m128i *) dst, srcABCD);
 							dst += 2;
 							src += 2;
 							src_mv += 2;
-							dstABCD = _mm_loadu_si128((__m128i*) dst);
-							_mm_storeu_si128((__m128i *) old_dst, srcABCD);
-							mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
-							srcABCD = _mm_loadu_si128((const __m128i*) src);
 						}
 
 						if (effective_width & 1) {
-							/* In case the m-channel is zero, do not remap this pixel in any way */
-							if (src_mv->m == 0) {
-								if (src->a < 255) {
-									ALPHA_BLEND_2(pack_hi_cm);
-									(*dst).data = EXTR32(srcABCD, 2);
-								} else {
-									*dst = src->data;
-								}
-							} else {
+							/* In case the m-channel is zero, do not remap this pixel in any way. */
+							__m128i srcABCD;
+							if (src_mv->m) {
 								const uint r = remap[src_mv->m];
 								if (r != 0) {
 									Colour remapped_colour = AdjustBrightness(this->LookupColourInPalette(r), src_mv->v);
-									if (src->a < 255) {
-										remapped_colour.a = src->a;
-										INSR32(remapped_colour.data, srcABCD, 0);
-										ALPHA_BLEND_2(pack_hi_cm);
-										(*dst).data = EXTR32(srcABCD, 2);
-									} else
+									if (src->a == 255) {
 										*dst = remapped_colour;
+									} else {
+										remapped_colour.a = src->a;
+										srcABCD = _mm_cvtsi32_si128(remapped_colour.data);
+										goto bmcr_alpha_blend_single;
+									}
+								}
+							} else {
+								srcABCD = _mm_cvtsi32_si128(src->data);
+								if (src->a < 255) {
+bmcr_alpha_blend_single:
+									__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
+									ALPHA_BLEND_2(pack_low_cm);
 								}
+								dst->data = _mm_cvtsi128_si32(srcABCD);
 							}
 						}
 						break;
@@ -200,30 +188,29 @@ inline void Blitter_32bppSSSE3::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 				src_mv_line += si->sprite_width;
 				break;
 			}
+
 			case BM_TRANSPARENT: {
 				/* Make the current colour a bit more black, so it looks like this image is transparent.
 				 * rgb = rgb * ((256/4) * 4 - (alpha/4)) / ((256/4) * 4)
 				 */
-				__m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
-				__m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
 				for (uint x = (uint) bp->width / 2; x > 0; x--) {
+					__m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
+					__m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
 					__m128i srcAB = _mm_unpacklo_epi8(srcABCD, _mm_setzero_si128());
 					__m128i dstAB = _mm_unpacklo_epi8(dstABCD, _mm_setzero_si128());
-					__m128i dstCD = _mm_unpackhi_epi8(dstABCD, _mm_setzero_si128());
 					__m128i alphaAB = _mm_shuffle_epi8(srcAB, a_cm);
 					alphaAB = _mm_srli_epi16(alphaAB, 2); // Reduce to 64 levels of shades so the max value fits in 16 bits.
 					__m128i nom = _mm_sub_epi16(tr_nom_base, alphaAB);
 					dstAB = _mm_mullo_epi16(dstAB, nom);
 					dstAB = _mm_srli_epi16(dstAB, 8);
-					dstAB = _mm_packus_epi16(dstAB, dstCD);
-					Colour *old_dst = dst;
+					dstAB = _mm_packus_epi16(dstAB, dstAB);
+					_mm_storel_epi64((__m128i *) dst, dstAB);
 					src += 2;
 					dst += 2;
-					dstABCD = _mm_loadu_si128((__m128i*) dst);
-					_mm_storeu_si128((__m128i *) old_dst, dstAB);
-					srcABCD = _mm_loadu_si128((const __m128i*) src);
 				}
 				if (bp->width & 1) {
+					__m128i srcABCD = _mm_cvtsi32_si128(src->data);
+					__m128i dstABCD = _mm_cvtsi32_si128(dst->data);
 					__m128i srcAB = _mm_unpacklo_epi8(srcABCD, _mm_setzero_si128());
 					__m128i dstAB = _mm_unpacklo_epi8(dstABCD, _mm_setzero_si128());
 					__m128i alphaAB = _mm_shuffle_epi8(srcAB, a_cm);
@@ -232,7 +219,7 @@ inline void Blitter_32bppSSSE3::Draw(const Blitter::BlitterParams *bp, ZoomLevel
 					dstAB = _mm_mullo_epi16(dstAB, nom);
 					dstAB = _mm_srli_epi16(dstAB, 8);
 					dstAB = _mm_packus_epi16(dstAB, dstAB);
-					(*dst).data = EXTR32(dstAB, 0);
+					dst->data = _mm_cvtsi128_si32(dstAB);
 				}
 				break;
 			}
diff --git a/src/blitter/32bpp_ssse3.hpp b/src/blitter/32bpp_ssse3.hpp
index cede4185d..97c67657e 100644
--- a/src/blitter/32bpp_ssse3.hpp
+++ b/src/blitter/32bpp_ssse3.hpp
@@ -47,8 +47,7 @@
 	__m128i zero = _mm_setzero_si128(); \
 	__m128i colAB = _mm_unpacklo_epi8(colourX2, zero); \
 	\
-	__m128i briAB; \
-	INSR64(brightnessX2, briAB, 0); \
+	__m128i briAB = _mm_cvtsi32_si128(brightnessX2); \
 	briAB = _mm_shuffle_epi8(briAB, briAB_cm); /* DEFAULT_BRIGHTNESS in 0, 0x00 in 2. */ \
 	colAB = _mm_mullo_epi16(colAB, briAB); \
 	__m128i colAB_ob = _mm_srli_epi16(colAB, 8+7); \
diff --git a/src/video/win32_v.cpp b/src/video/win32_v.cpp
index 02cae68e2..186ac1025 100644
--- a/src/video/win32_v.cpp
+++ b/src/video/win32_v.cpp
@@ -1077,7 +1077,7 @@ static bool AllocateDibSection(int w, int h, bool force)
 	bi->bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
 
 	bi->bmiHeader.biWidth = _wnd.width = w;
-	bi->bmiHeader.biHeight = -(_wnd.height = h+1); // Allocate extra room to prevent out-of-bounds when SSE reads a 16B block at the end of the buffer.
+	bi->bmiHeader.biHeight = -(_wnd.height = h);
 
 	bi->bmiHeader.biPlanes = 1;
 	bi->bmiHeader.biBitCount = BlitterFactory::GetCurrentBlitter()->GetScreenDepth();