vc4: Don't bother masking out the low 24 bits for integer multiplies

The hardware just uses the low 24 lines, saving us an AND to drop the high
bits.

total uniforms in shared programs: 13433 -> 13423 (-0.07%)
uniforms in affected programs:     356 -> 346 (-2.81%)
total instructions in shared programs: 40003 -> 39989 (-0.03%)
instructions in affected programs:     910 -> 896 (-1.54%)

diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index 49b94666b886885dda1f7776fd9319d619f4d36f..9e145e54ccd8ac8bfc117332a20f11f62c849f9f 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -366,18 +366,14 @@ tgsi_to_qir_umul(struct vc4_compile *c,
                  struct tgsi_full_instruction *tgsi_inst,
                  enum qop op, struct qreg *src, int i)
 {
-        struct qreg src0_hi = qir_SHR(c, src[0 * 4 + i],
-                                      qir_uniform_ui(c, 24));
-        struct qreg src0_lo = qir_AND(c, src[0 * 4 + i],
-                                      qir_uniform_ui(c, 0xffffff));
-        struct qreg src1_hi = qir_SHR(c, src[1 * 4 + i],
-                                      qir_uniform_ui(c, 24));
-        struct qreg src1_lo = qir_AND(c, src[1 * 4 + i],
-                                      qir_uniform_ui(c, 0xffffff));
-
-        struct qreg hilo = qir_MUL24(c, src0_hi, src1_lo);
-        struct qreg lohi = qir_MUL24(c, src0_lo, src1_hi);
-        struct qreg lolo = qir_MUL24(c, src0_lo, src1_lo);
+        struct qreg src0 = src[0 * 4 + i];
+        struct qreg src0_hi = qir_SHR(c, src0, qir_uniform_ui(c, 24));
+        struct qreg src1 = src[1 * 4 + i];
+        struct qreg src1_hi = qir_SHR(c, src1, qir_uniform_ui(c, 24));
+
+        struct qreg hilo = qir_MUL24(c, src0_hi, src1);
+        struct qreg lohi = qir_MUL24(c, src0, src1_hi);
+        struct qreg lolo = qir_MUL24(c, src0, src1);
 
         return qir_ADD(c, lolo, qir_SHL(c,
                                         qir_ADD(c, hilo, lohi),