[Mesa-dev] [PATCH 4/4] tgsi/lowering: improved lowering for LRP
Rob Clark
robdclark at gmail.com
Wed Apr 13 22:42:42 UTC 2016
From: Russell King <rmk at arm.linux.org.uk>
Provide an improved lowering for LRP, which can be implemented in two
MAD instructions with a bit of rearranging of the equation, rather
than the literal implementation of two multiplies, an add and a
subtract.
Signed-off-by: Russell King <rmk at arm.linux.org.uk>
Reviewed-by: Rob Clark <robdclark at gmail.com>
---
src/gallium/auxiliary/tgsi/tgsi_lowering.c | 55 +++++++++++-------------------
1 file changed, 20 insertions(+), 35 deletions(-)
diff --git a/src/gallium/auxiliary/tgsi/tgsi_lowering.c b/src/gallium/auxiliary/tgsi/tgsi_lowering.c
index 2a99158..d54f4b9 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_lowering.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_lowering.c
@@ -388,14 +388,15 @@ transform_scs(struct tgsi_transform_context *tctx,
* dst.z = src0.z \times src1.z + (1.0 - src0.z) \times src2.z
* dst.w = src0.w \times src1.w + (1.0 - src0.w) \times src2.w
*
- * ; needs: 2 tmp, imm{1.0}
- * MUL tmpA, src0, src1
- * SUB tmpB, imm{1.0}, src0
- * MUL tmpB, tmpB, src2
- * ADD dst, tmpA, tmpB
+ * This becomes: src0 \times src1 + src2 - src0 \times src2, which
+ * can then become: src0 \times src1 - (src0 \times src2 - src2)
+ *
+ * ; needs: 1 tmp
+ * MAD tmpA, src0, src2, -src2
+ * MAD dst, src0, src1, -tmpA
*/
-#define LRP_GROW (NINST(2) + NINST(2) + NINST(2) + NINST(2) - OINST(3))
-#define LRP_TMP 2
+#define LRP_GROW (NINST(3) + NINST(3) - OINST(3))
+#define LRP_TMP 1
static void
transform_lrp(struct tgsi_transform_context *tctx,
struct tgsi_full_instruction *inst)
@@ -408,44 +409,28 @@ transform_lrp(struct tgsi_transform_context *tctx,
struct tgsi_full_instruction new_inst;
if (dst->Register.WriteMask & TGSI_WRITEMASK_XYZW) {
- /* MUL tmpA, src0, src1 */
+ /* MAD tmpA, src0, src2, -src2 */
new_inst = tgsi_default_full_instruction();
- new_inst.Instruction.Opcode = TGSI_OPCODE_MUL;
+ new_inst.Instruction.Opcode = TGSI_OPCODE_MAD;
new_inst.Instruction.NumDstRegs = 1;
reg_dst(&new_inst.Dst[0], &ctx->tmp[A].dst, TGSI_WRITEMASK_XYZW);
- new_inst.Instruction.NumSrcRegs = 2;
+ new_inst.Instruction.NumSrcRegs = 3;
reg_src(&new_inst.Src[0], src0, SWIZ(X, Y, Z, W));
- reg_src(&new_inst.Src[1], src1, SWIZ(X, Y, Z, W));
- tctx->emit_instruction(tctx, &new_inst);
-
- /* SUB tmpB, imm{1.0}, src0 */
- new_inst = tgsi_default_full_instruction();
- new_inst.Instruction.Opcode = TGSI_OPCODE_SUB;
- new_inst.Instruction.NumDstRegs = 1;
- reg_dst(&new_inst.Dst[0], &ctx->tmp[B].dst, TGSI_WRITEMASK_XYZW);
- new_inst.Instruction.NumSrcRegs = 2;
- reg_src(&new_inst.Src[0], &ctx->imm, SWIZ(Y, Y, Y, Y));
- reg_src(&new_inst.Src[1], src0, SWIZ(X, Y, Z, W));
- tctx->emit_instruction(tctx, &new_inst);
-
- /* MUL tmpB, tmpB, src2 */
- new_inst = tgsi_default_full_instruction();
- new_inst.Instruction.Opcode = TGSI_OPCODE_MUL;
- new_inst.Instruction.NumDstRegs = 1;
- reg_dst(&new_inst.Dst[0], &ctx->tmp[B].dst, TGSI_WRITEMASK_XYZW);
- new_inst.Instruction.NumSrcRegs = 2;
- reg_src(&new_inst.Src[0], &ctx->tmp[B].src, SWIZ(X, Y, Z, W));
reg_src(&new_inst.Src[1], src2, SWIZ(X, Y, Z, W));
+ reg_src(&new_inst.Src[2], src2, SWIZ(X, Y, Z, W));
+ new_inst.Src[2].Register.Negate = !new_inst.Src[2].Register.Negate;
tctx->emit_instruction(tctx, &new_inst);
- /* ADD dst, tmpA, tmpB */
+ /* MAD dst, src0, src1, -tmpA */
new_inst = tgsi_default_full_instruction();
- new_inst.Instruction.Opcode = TGSI_OPCODE_ADD;
+ new_inst.Instruction.Opcode = TGSI_OPCODE_MAD;
new_inst.Instruction.NumDstRegs = 1;
reg_dst(&new_inst.Dst[0], dst, TGSI_WRITEMASK_XYZW);
- new_inst.Instruction.NumSrcRegs = 2;
- reg_src(&new_inst.Src[0], &ctx->tmp[A].src, SWIZ(X, Y, Z, W));
- reg_src(&new_inst.Src[1], &ctx->tmp[B].src, SWIZ(X, Y, Z, W));
+ new_inst.Instruction.NumSrcRegs = 3;
+ reg_src(&new_inst.Src[0], src0, SWIZ(X, Y, Z, W));
+ reg_src(&new_inst.Src[1], src1, SWIZ(X, Y, Z, W));
+ reg_src(&new_inst.Src[2], &ctx->tmp[A].src, SWIZ(X, Y, Z, W));
+ new_inst.Src[2].Register.Negate = true;
tctx->emit_instruction(tctx, &new_inst);
}
}
--
2.5.5
More information about the mesa-dev
mailing list