diff mbox series

[6/8] target/ppc: Optimize emulation of vclzw instruction

Message ID 1559816130-17113-7-git-send-email-stefan.brankovic@rt-rk.com (mailing list archive)
State New, archived
Headers show
Series Optimize emulation of ten Altivec instructions: lvsl, | expand

Commit Message

Stefan Brankovic June 6, 2019, 10:15 a.m. UTC
Optimize Altivec instruction vclzw (Vector Count Leading Zeros Word).
This instruction counts the number of leading zeros of each word element
in source register and places result in the appropriate word element of
destination register.

We perform counting in two iterations of for loop(one for each
doubleword element of source register vB). First thing we do in
loop is placing appropriate doubleword element of vB in variable
avr. Then we perform counting using tcg-s count leading zeros
function. Since it counts leading zeros on 64 bit lenght, we have to
move ith word element to highest 32 bits of variable tmp, or it with
mask(so we get all ones in lowest 32 bits), then perform
tcg_gen_clzi_i64 and move it's result in appropriate word element of
variable result. In the end of each loop iteration we save variable
result to appropriate doubleword element of destination register vD.

Signed-off-by: Stefan Brankovic <stefan.brankovic@rt-rk.com>
---
 target/ppc/translate/vmx-impl.inc.c | 57 ++++++++++++++++++++++++++++++++++++-
 1 file changed, 56 insertions(+), 1 deletion(-)

Comments

Richard Henderson June 6, 2019, 6:34 p.m. UTC | #1
On 6/6/19 5:15 AM, Stefan Brankovic wrote:
> +    for (i = 0; i < 2; i++) {
> +        if (i == 0) {
> +            /* Get high doubleword element of vB in avr. */
> +            get_avr64(avr, VB, true);
> +        } else {
> +            /* Get low doubleword element of vB in avr. */
> +            get_avr64(avr, VB, false);
> +        }

Better as simply get_avr64(avr, VB, i);

> +        /*
> +         * Perform count for every word element using tcg_gen_clzi_i64.
> +         * Since it counts leading zeros on 64 bit lenght, we have to move
> +         * ith word element to highest 32 bits of tmp, or it with mask(so we get
> +         * all ones in lowest 32 bits), then perform tcg_gen_clzi_i64 and move
> +         * it's result in appropriate word element of result.
> +         */
> +        tcg_gen_shli_i64(tmp, avr, 32);
> +        tcg_gen_or_i64(tmp, tmp, mask);
> +        tcg_gen_clzi_i64(result, tmp, 64);
> +
> +        tcg_gen_or_i64(tmp, avr, mask);
> +        tcg_gen_clzi_i64(tmp, tmp, 64);

s/64/32.

> +        tcg_gen_deposit_i64(result, result, tmp, 32, 32);

That said, it's probably better to treat this as 4 words, not 2 doublewords.

	for (i = 0; i < 4; i++) {
	    tcg_gen_ld_i32(tmp, cpu_env, avr_full_offset(VB) + i * 4);
	    tcg_gen_clzi_i32(tmp, tmp, 32);
	    tcg_gen_st_i32(tmp, cpu_env, avr_full_offset(VT) + i * 4);
	}


r~
Stefan Brankovic June 17, 2019, 11:50 a.m. UTC | #2
On 6.6.19. 20:34, Richard Henderson wrote:
> On 6/6/19 5:15 AM, Stefan Brankovic wrote:
>> +    for (i = 0; i < 2; i++) {
>> +        if (i == 0) {
>> +            /* Get high doubleword element of vB in avr. */
>> +            get_avr64(avr, VB, true);
>> +        } else {
>> +            /* Get low doubleword element of vB in avr. */
>> +            get_avr64(avr, VB, false);
>> +        }
> Better as simply get_avr64(avr, VB, i);
Definitely shorter way to do this.
>
>> +        /*
>> +         * Perform count for every word element using tcg_gen_clzi_i64.
>> +         * Since it counts leading zeros on 64 bit lenght, we have to move
>> +         * ith word element to highest 32 bits of tmp, or it with mask(so we get
>> +         * all ones in lowest 32 bits), then perform tcg_gen_clzi_i64 and move
>> +         * it's result in appropriate word element of result.
>> +         */
>> +        tcg_gen_shli_i64(tmp, avr, 32);
>> +        tcg_gen_or_i64(tmp, tmp, mask);
>> +        tcg_gen_clzi_i64(result, tmp, 64);
>> +
>> +        tcg_gen_or_i64(tmp, avr, mask);
>> +        tcg_gen_clzi_i64(tmp, tmp, 64);
> s/64/32.
>
>> +        tcg_gen_deposit_i64(result, result, tmp, 32, 32);
> That said, it's probably better to treat this as 4 words, not 2 doublewords.
>
> 	for (i = 0; i < 4; i++) {
> 	    tcg_gen_ld_i32(tmp, cpu_env, avr_full_offset(VB) + i * 4);
> 	    tcg_gen_clzi_i32(tmp, tmp, 32);
> 	    tcg_gen_st_i32(tmp, cpu_env, avr_full_offset(VT) + i * 4);
> 	}
>
I will use this way in v2.

Kind Regards,

Stefan

> r~
diff mbox series

Patch

diff --git a/target/ppc/translate/vmx-impl.inc.c b/target/ppc/translate/vmx-impl.inc.c
index 1c34908..7689739 100644
--- a/target/ppc/translate/vmx-impl.inc.c
+++ b/target/ppc/translate/vmx-impl.inc.c
@@ -878,6 +878,61 @@  static void trans_vgbbd(DisasContext *ctx)
 }
 
 /*
+ * vclzw VRT,VRB - Vector Count Leading Zeros Word
+ *
+ * Counting the number of leading zero bits of each word element in source
+ * register and placing result in appropriate word element of destination
+ * register.
+ */
+static void trans_vclzw(DisasContext *ctx)
+{
+    int VT = rD(ctx->opcode);
+    int VB = rB(ctx->opcode);
+    TCGv_i64 avr = tcg_temp_new_i64();
+    TCGv_i64 result = tcg_temp_new_i64();
+    TCGv_i64 tmp = tcg_temp_new_i64();
+    TCGv_i64 mask = tcg_const_i64(0xffffffffULL);
+    int i;
+
+    for (i = 0; i < 2; i++) {
+        if (i == 0) {
+            /* Get high doubleword element of vB in avr. */
+            get_avr64(avr, VB, true);
+        } else {
+            /* Get low doubleword element of vB in avr. */
+            get_avr64(avr, VB, false);
+        }
+        /*
+         * Perform count for every word element using tcg_gen_clzi_i64.
+         * Since it counts leading zeros on 64 bit lenght, we have to move
+         * ith word element to highest 32 bits of tmp, or it with mask(so we get
+         * all ones in lowest 32 bits), then perform tcg_gen_clzi_i64 and move
+         * it's result in appropriate word element of result.
+         */
+        tcg_gen_shli_i64(tmp, avr, 32);
+        tcg_gen_or_i64(tmp, tmp, mask);
+        tcg_gen_clzi_i64(result, tmp, 64);
+
+        tcg_gen_or_i64(tmp, avr, mask);
+        tcg_gen_clzi_i64(tmp, tmp, 64);
+        tcg_gen_deposit_i64(result, result, tmp, 32, 32);
+
+        if (i == 0) {
+            /* Place result in high doubleword element of vD. */
+            set_avr64(VT, result, true);
+        } else {
+            /* Place result in low doubleword element of vD. */
+            set_avr64(VT, result, false);
+        }
+    }
+
+    tcg_temp_free_i64(avr);
+    tcg_temp_free_i64(result);
+    tcg_temp_free_i64(tmp);
+    tcg_temp_free_i64(mask);
+}
+
+/*
  * vclzd VRT,VRB - Vector Count Leading Zeros Doubleword
  *
  * Counting the number of leading zero bits of each doubleword element in source
@@ -1413,7 +1468,7 @@  GEN_VAFORM_PAIRED(vmaddfp, vnmsubfp, 23)
 
 GEN_VXFORM_NOA(vclzb, 1, 28)
 GEN_VXFORM_NOA(vclzh, 1, 29)
-GEN_VXFORM_NOA(vclzw, 1, 30)
+GEN_VXFORM_TRANS(vclzw, 1, 30)
 GEN_VXFORM_TRANS(vclzd, 1, 31)
 GEN_VXFORM_NOA_2(vnegw, 1, 24, 6)
 GEN_VXFORM_NOA_2(vnegd, 1, 24, 7)