From a58dbf34ca88656739a8f8e5f4259e760365c9d0 Mon Sep 17 00:00:00 2001 From: Brian Date: Sat, 8 Nov 2008 10:29:23 -0700 Subject: gallium: implement SSE codegen for TGSI_OPCODE_NRM/NRM4 --- src/gallium/auxiliary/tgsi/tgsi_sse2.c | 34 +++++++++++++++++++++++++++++++++- 1 file changed, 33 insertions(+), 1 deletion(-) diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c index b8e4ab6d8..3ce2c1c27 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c +++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c @@ -2087,7 +2087,39 @@ emit_instruction( break; case TGSI_OPCODE_NRM: - return 0; + /* fall-through */ + case TGSI_OPCODE_NRM4: + /* 3 or 4-component normalization */ + { + uint dims = (inst->Instruction.Opcode == TGSI_OPCODE_NRM) ? 3 : 4; + /* note: cannot use xmm regs 2/3 here (see emit_rsqrt() above) */ + FETCH( func, *inst, 4, 0, CHAN_X ); /* xmm4 = src[0].x */ + FETCH( func, *inst, 5, 0, CHAN_Y ); /* xmm5 = src[0].y */ + FETCH( func, *inst, 6, 0, CHAN_Z ); /* xmm6 = src[0].z */ + if (dims == 4) { + FETCH( func, *inst, 7, 0, CHAN_W ); /* xmm7 = src[0].w */ + } + emit_MOV( func, 0, 4 ); /* xmm0 = xmm3 */ + emit_mul( func, 0, 4 ); /* xmm0 *= xmm3 */ + emit_MOV( func, 1, 5 ); /* xmm1 = xmm4 */ + emit_mul( func, 1, 5 ); /* xmm1 *= xmm4 */ + emit_add( func, 0, 1 ); /* xmm0 += xmm1 */ + emit_MOV( func, 1, 6 ); /* xmm1 = xmm5 */ + emit_mul( func, 1, 6 ); /* xmm1 *= xmm5 */ + emit_add( func, 0, 1 ); /* xmm0 += xmm1 */ + if (dims == 4) { + emit_MOV( func, 1, 7 ); /* xmm1 = xmm7 */ + emit_mul( func, 1, 7 ); /* xmm1 *= xmm7 */ + emit_add( func, 0, 0 ); /* xmm0 += xmm1 */ + } + emit_rsqrt( func, 1, 0 ); /* xmm1 = 1/sqrt(xmm0) */ + FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { + if (chan_index < dims) { + emit_mul( func, 4+chan_index, 1); /* xmm[4+ch] *= xmm1 */ + STORE( func, *inst, 4+chan_index, 0, chan_index ); + } + } + } break; case TGSI_OPCODE_DIV: -- cgit v1.2.3