Merge pull request #2793 from itzpr3d4t0r/simplify-fill-algorithms

Starbuck5 · web-flow · commit 89f98480116b · 2024-05-21T12:32:50.000-04:00
Simplified SIMD fill algorithms
diff --git a/src_c/simd_surface_fill_avx2.c b/src_c/simd_surface_fill_avx2.c
@@ -85,24 +85,20 @@ _pg_has_avx2()
     }
 
 /* Setup for RUN_16BIT_SHUFFLE_OUT */
-#define SETUP_SHUFFLE                                                         \
-    __m256i shuff_dst, _shuff16_temp, mm256_colorA, mm256_colorB, mm256_zero; \
-    mm256_zero = _mm256_setzero_si256();                                      \
-    mm256_colorA = _mm256_unpacklo_epi8(mm256_color, mm256_zero);             \
-    mm256_colorB = _mm256_unpackhi_epi8(mm256_color, mm256_zero);
+#define SETUP_SHUFFLE                                                      \
+    __m256i shuff_dst, _shuff16_temp, mm256_zero = _mm256_setzero_si256(); \
+    mm256_color = _mm256_unpacklo_epi8(mm256_color, mm256_zero);
 
 #define RUN_16BIT_SHUFFLE_OUT(FILL_CODE)                       \
     /* ==== shuffle pixels out into two registers each, src */ \
     /* and dst set up for 16 bit math, like 0A0R0G0B ==== */   \
     shuff_dst = _mm256_unpacklo_epi8(mm256_dst, mm256_zero);   \
-    mm256_color = mm256_colorA;                                \
                                                                \
     {FILL_CODE}                                                \
                                                                \
     _shuff16_temp = shuff_dst;                                 \
                                                                \
     shuff_dst = _mm256_unpackhi_epi8(mm256_dst, mm256_zero);   \
-    mm256_color = mm256_colorB;                                \
                                                                \
     {FILL_CODE}                                                \
                                                                \
diff --git a/src_c/simd_surface_fill_sse2.c b/src_c/simd_surface_fill_sse2.c
@@ -77,26 +77,23 @@ _pg_HasSSE_NEON()
 
 /* Setup for RUN_16BIT_SHUFFLE_OUT */
 #define SETUP_SHUFFLE                                                   \
-    __m128i shuff_dst, _shuff16_temp, mm128_colorA, mm128_colorB;       \
-    mm128_colorA = _mm_unpacklo_epi8(mm128_color, _mm_setzero_si128()); \
-    mm128_colorB = _mm_unpackhi_epi8(mm128_color, _mm_setzero_si128());
+    __m128i shuff_dst, _shuff16_temp, mm128_zero = _mm_setzero_si128(); \
+    mm128_color = _mm_unpacklo_epi8(mm128_color, mm128_zero);
 
-#define RUN_16BIT_SHUFFLE_OUT(FILL_CODE)                           \
-    /* ==== shuffle pixels out into two registers each, src */     \
-    /* and dst set up for 16 bit math, like 0A0R0G0B ==== */       \
-    shuff_dst = _mm_unpacklo_epi8(mm128_dst, _mm_setzero_si128()); \
-    mm128_color = mm128_colorA;                                    \
-                                                                   \
-    {FILL_CODE}                                                    \
-                                                                   \
-    _shuff16_temp = shuff_dst;                                     \
-                                                                   \
-    shuff_dst = _mm_unpackhi_epi8(mm128_dst, _mm_setzero_si128()); \
-    mm128_color = mm128_colorB;                                    \
-                                                                   \
-    {FILL_CODE}                                                    \
-                                                                   \
-    /* ==== recombine A and B pixels ==== */                       \
+#define RUN_16BIT_SHUFFLE_OUT(FILL_CODE)                       \
+    /* ==== shuffle pixels out into two registers each, src */ \
+    /* and dst set up for 16 bit math, like 0A0R0G0B ==== */   \
+    shuff_dst = _mm_unpacklo_epi8(mm128_dst, mm128_zero);      \
+                                                               \
+    {FILL_CODE}                                                \
+                                                               \
+    _shuff16_temp = shuff_dst;                                 \
+                                                               \
+    shuff_dst = _mm_unpackhi_epi8(mm128_dst, mm128_zero);      \
+                                                               \
+    {FILL_CODE}                                                \
+                                                               \
+    /* ==== recombine A and B pixels ==== */                   \
     mm128_dst = _mm_packus_epi16(_shuff16_temp, shuff_dst);
 
 #define FILLERS(NAME, COLOR_PROCESS_CODE, FILL_CODE)                        \

Original file line number	Diff line number	Diff line change
`@@ -85,24 +85,20 @@ _pg_has_avx2()`
`85`	`85`	`}`
`86`	`86`
`87`	`87`	`/* Setup for RUN_16BIT_SHUFFLE_OUT */`
`88`		`-#define SETUP_SHUFFLE \`
`89`		`- __m256i shuff_dst, _shuff16_temp, mm256_colorA, mm256_colorB, mm256_zero; \`
`90`		`- mm256_zero = _mm256_setzero_si256(); \`
`91`		`- mm256_colorA = _mm256_unpacklo_epi8(mm256_color, mm256_zero); \`
`92`		`- mm256_colorB = _mm256_unpackhi_epi8(mm256_color, mm256_zero);`
	`88`	`+#define SETUP_SHUFFLE \`
	`89`	`+ __m256i shuff_dst, _shuff16_temp, mm256_zero = _mm256_setzero_si256(); \`
	`90`	`+ mm256_color = _mm256_unpacklo_epi8(mm256_color, mm256_zero);`
`93`	`91`
`94`	`92`	`#define RUN_16BIT_SHUFFLE_OUT(FILL_CODE) \`
`95`	`93`	`/* ==== shuffle pixels out into two registers each, src */ \`
`96`	`94`	`/* and dst set up for 16 bit math, like 0A0R0G0B ==== */ \`
`97`	`95`	`shuff_dst = _mm256_unpacklo_epi8(mm256_dst, mm256_zero); \`
`98`		`- mm256_color = mm256_colorA; \`
`99`	`96`	`\`
`100`	`97`	`{FILL_CODE} \`
`101`	`98`	`\`
`102`	`99`	`_shuff16_temp = shuff_dst; \`
`103`	`100`	`\`
`104`	`101`	`shuff_dst = _mm256_unpackhi_epi8(mm256_dst, mm256_zero); \`
`105`		`- mm256_color = mm256_colorB; \`
`106`	`102`	`\`
`107`	`103`	`{FILL_CODE} \`
`108`	`104`	`\`