Get rid of if constexpr

Raghuveer Devulapalli · Raghuveer Devulapalli · commit bab65af523ef · 2023-05-02T14:33:17.000-07:00
diff --git a/src/avx512-64bit-common.h b/src/avx512-64bit-common.h
@@ -136,18 +136,18 @@ struct ymm_vector<float> {
     {
         return _mm256_set1_ps(v);
     }
-    template <uint8_t mask>
+    template <uint8_t mask, bool = (mask == 0b01010101)>
     static zmm_t shuffle(zmm_t zmm)
     {
         /* Hack!: have to make shuffles within 128-bit lanes work for both
          * 32-bit and 64-bit */
-        if constexpr (mask == 0b01010101) {
-            return _mm256_shuffle_ps(zmm, zmm, 0b10110001);
-        }
-        else {
-            /* Not used, so far */
-            return _mm256_shuffle_ps(zmm, zmm, mask);
-        }
+        return _mm256_shuffle_ps(zmm, zmm, 0b10110001);
+        //if constexpr (mask == 0b01010101) {
+        //}
+        //else {
+        //    /* Not used, so far */
+        //    return _mm256_shuffle_ps(zmm, zmm, mask);
+        //}
     }
     static void storeu(void *mem, zmm_t x)
     {
@@ -271,18 +271,12 @@ struct ymm_vector<uint32_t> {
     {
         return _mm256_set1_epi32(v);
     }
-    template <uint8_t mask>
+    template <uint8_t mask, bool = (mask == 0b01010101)>
     static zmm_t shuffle(zmm_t zmm)
     {
         /* Hack!: have to make shuffles within 128-bit lanes work for both
          * 32-bit and 64-bit */
-        if constexpr (mask == 0b01010101) {
-            return _mm256_shuffle_epi32(zmm, 0b10110001);
-        }
-        else {
-            /* Not used, so far */
-            return _mm256_shuffle_epi32(zmm, mask);
-        }
+        return _mm256_shuffle_epi32(zmm, 0b10110001);
     }
     static void storeu(void *mem, zmm_t x)
     {
@@ -406,18 +400,12 @@ struct ymm_vector<int32_t> {
     {
         return _mm256_set1_epi32(v);
     }
-    template <uint8_t mask>
+    template <uint8_t mask, bool = (mask == 0b01010101)>
     static zmm_t shuffle(zmm_t zmm)
     {
         /* Hack!: have to make shuffles within 128-bit lanes work for both
          * 32-bit and 64-bit */
-        if constexpr (mask == 0b01010101) {
-            return _mm256_shuffle_epi32(zmm, 0b10110001);
-        }
-        else {
-            /* Not used, so far */
-            return _mm256_shuffle_epi32(zmm, mask);
-        }
+        return _mm256_shuffle_epi32(zmm, 0b10110001);
     }
     static void storeu(void *mem, zmm_t x)
     {

Original file line number	Diff line number	Diff line change
`@@ -136,18 +136,18 @@ struct ymm_vector<float> {`
`136`	`136`	`{`
`137`	`137`	`return _mm256_set1_ps(v);`
`138`	`138`	`}`
`139`		`- template <uint8_t mask>`
	`139`	`+ template <uint8_t mask, bool = (mask == 0b01010101)>`
`140`	`140`	`static zmm_t shuffle(zmm_t zmm)`
`141`	`141`	`{`
`142`	`142`	`/* Hack!: have to make shuffles within 128-bit lanes work for both`
`143`	`143`	`* 32-bit and 64-bit */`
`144`		`- if constexpr (mask == 0b01010101) {`
`145`		`- return _mm256_shuffle_ps(zmm, zmm, 0b10110001);`
`146`		`- }`
`147`		`- else {`
`148`		`- /* Not used, so far */`
`149`		`- return _mm256_shuffle_ps(zmm, zmm, mask);`
`150`		`- }`
	`144`	`+ return _mm256_shuffle_ps(zmm, zmm, 0b10110001);`
	`145`	`+ //if constexpr (mask == 0b01010101) {`
	`146`	`+ //}`
	`147`	`+ //else {`
	`148`	`+ // /* Not used, so far */`
	`149`	`+ // return _mm256_shuffle_ps(zmm, zmm, mask);`
	`150`	`+ //}`
`151`	`151`	`}`
`152`	`152`	`static void storeu(void *mem, zmm_t x)`
`153`	`153`	`{`
`@@ -271,18 +271,12 @@ struct ymm_vector<uint32_t> {`
`271`	`271`	`{`
`272`	`272`	`return _mm256_set1_epi32(v);`
`273`	`273`	`}`
`274`		`- template <uint8_t mask>`
	`274`	`+ template <uint8_t mask, bool = (mask == 0b01010101)>`
`275`	`275`	`static zmm_t shuffle(zmm_t zmm)`
`276`	`276`	`{`
`277`	`277`	`/* Hack!: have to make shuffles within 128-bit lanes work for both`
`278`	`278`	`* 32-bit and 64-bit */`
`279`		`- if constexpr (mask == 0b01010101) {`
`280`		`- return _mm256_shuffle_epi32(zmm, 0b10110001);`
`281`		`- }`
`282`		`- else {`
`283`		`- /* Not used, so far */`
`284`		`- return _mm256_shuffle_epi32(zmm, mask);`
`285`		`- }`
	`279`	`+ return _mm256_shuffle_epi32(zmm, 0b10110001);`
`286`	`280`	`}`
`287`	`281`	`static void storeu(void *mem, zmm_t x)`
`288`	`282`	`{`
`@@ -406,18 +400,12 @@ struct ymm_vector<int32_t> {`
`406`	`400`	`{`
`407`	`401`	`return _mm256_set1_epi32(v);`
`408`	`402`	`}`
`409`		`- template <uint8_t mask>`
	`403`	`+ template <uint8_t mask, bool = (mask == 0b01010101)>`
`410`	`404`	`static zmm_t shuffle(zmm_t zmm)`
`411`	`405`	`{`
`412`	`406`	`/* Hack!: have to make shuffles within 128-bit lanes work for both`
`413`	`407`	`* 32-bit and 64-bit */`
`414`		`- if constexpr (mask == 0b01010101) {`
`415`		`- return _mm256_shuffle_epi32(zmm, 0b10110001);`
`416`		`- }`
`417`		`- else {`
`418`		`- /* Not used, so far */`
`419`		`- return _mm256_shuffle_epi32(zmm, mask);`
`420`		`- }`
	`408`	`+ return _mm256_shuffle_epi32(zmm, 0b10110001);`
`421`	`409`	`}`
`422`	`410`	`static void storeu(void *mem, zmm_t x)`
`423`	`411`	`{`