From: José Fonseca Date: Fri, 2 Jul 2010 17:42:49 +0000 (+0100) Subject: gallivm: Do 4ubyte AoS texture filtering for any format that can be expressed. X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=d981bde38472d8d3bb74dab67eccd7c82915a566;p=mesa.git gallivm: Do 4ubyte AoS texture filtering for any format that can be expressed. Except if it has only one channel, as it would take the same number of instructions. --- diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.c b/src/gallium/auxiliary/gallivm/lp_bld_sample.c index 6e556f9d4e6..800d1211136 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_sample.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.c @@ -127,20 +127,51 @@ lp_sampler_static_state(struct lp_sampler_static_state *state, /** * Compute the offset of a pixel block. * - * x, y, z, y_stride, z_stride are vectors, and they refer to pixel blocks, as - * per format description, and not individual pixels. + * x, y, z, y_stride, z_stride are vectors, and they refer to pixels. + * + * Returns the relative offset and i,j sub-block coordinates */ -LLVMValueRef +void lp_build_sample_offset(struct lp_build_context *bld, const struct util_format_description *format_desc, LLVMValueRef x, LLVMValueRef y, LLVMValueRef z, LLVMValueRef y_stride, - LLVMValueRef z_stride) + LLVMValueRef z_stride, + LLVMValueRef *out_offset, + LLVMValueRef *out_i, + LLVMValueRef *out_j) { LLVMValueRef x_stride; LLVMValueRef offset; + LLVMValueRef i; + LLVMValueRef j; + + /* + * Describe the coordinates in terms of pixel blocks. + * + * TODO: pixel blocks are power of two. LLVM should convert rem/div to + * bit arithmetic. Verify this. + */ + + if (format_desc->block.width == 1) { + i = bld->zero; + } + else { + LLVMValueRef block_width = lp_build_const_int_vec(bld->type, format_desc->block.width); + i = LLVMBuildURem(bld->builder, x, block_width, ""); + x = LLVMBuildUDiv(bld->builder, x, block_width, ""); + } + + if (format_desc->block.height == 1) { + j = bld->zero; + } + else { + LLVMValueRef block_height = lp_build_const_int_vec(bld->type, format_desc->block.height); + j = LLVMBuildURem(bld->builder, y, block_height, ""); + y = LLVMBuildUDiv(bld->builder, y, block_height, ""); + } x_stride = lp_build_const_vec(bld->type, format_desc->block.bits/8); offset = lp_build_mul(bld, x, x_stride); @@ -155,5 +186,7 @@ lp_build_sample_offset(struct lp_build_context *bld, offset = lp_build_add(bld, offset, z_offset); } - return offset; + *out_offset = offset; + *out_i = i; + *out_j = j; } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.h b/src/gallium/auxiliary/gallivm/lp_bld_sample.h index 64eb15a680a..5b8f478094b 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_sample.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.h @@ -146,14 +146,17 @@ lp_sampler_static_state(struct lp_sampler_static_state *state, const struct pipe_sampler_state *sampler); -LLVMValueRef +void lp_build_sample_offset(struct lp_build_context *bld, const struct util_format_description *format_desc, LLVMValueRef x, LLVMValueRef y, LLVMValueRef z, LLVMValueRef y_stride, - LLVMValueRef z_stride); + LLVMValueRef z_stride, + LLVMValueRef *out_offset, + LLVMValueRef *out_i, + LLVMValueRef *out_j); void diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c index 8cca3f639a5..b8c1a7234bc 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c @@ -265,35 +265,11 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld, } } - /* - * Describe the coordinates in terms of pixel blocks. - * - * TODO: pixel blocks are power of two. LLVM should convert rem/div to - * bit arithmetic. Verify this. - */ - - if (bld->format_desc->block.width == 1) { - i = bld->uint_coord_bld.zero; - } - else { - LLVMValueRef block_width = lp_build_const_int_vec(bld->uint_coord_bld.type, bld->format_desc->block.width); - i = LLVMBuildURem(bld->builder, x, block_width, ""); - x = LLVMBuildUDiv(bld->builder, x, block_width, ""); - } - - if (bld->format_desc->block.height == 1) { - j = bld->uint_coord_bld.zero; - } - else { - LLVMValueRef block_height = lp_build_const_int_vec(bld->uint_coord_bld.type, bld->format_desc->block.height); - j = LLVMBuildURem(bld->builder, y, block_height, ""); - y = LLVMBuildUDiv(bld->builder, y, block_height, ""); - } - /* convert x,y,z coords to linear offset from start of texture, in bytes */ - offset = lp_build_sample_offset(&bld->uint_coord_bld, - bld->format_desc, - x, y, z, y_stride, z_stride); + lp_build_sample_offset(&bld->uint_coord_bld, + bld->format_desc, + x, y, z, y_stride, z_stride, + &offset, &i, &j); if (use_border) { /* If we can sample the border color, it means that texcoords may @@ -345,6 +321,9 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld, } +/** + * Fetch the texels as <4n x i8> in AoS form. + */ static LLVMValueRef lp_build_sample_packed(struct lp_build_sample_context *bld, LLVMValueRef x, @@ -352,25 +331,46 @@ lp_build_sample_packed(struct lp_build_sample_context *bld, LLVMValueRef y_stride, LLVMValueRef data_array) { - LLVMValueRef offset; + LLVMValueRef offset, i, j; LLVMValueRef data_ptr; + LLVMValueRef res; - offset = lp_build_sample_offset(&bld->uint_coord_bld, - bld->format_desc, - x, y, NULL, y_stride, NULL); - - assert(bld->format_desc->block.width == 1); - assert(bld->format_desc->block.height == 1); - assert(bld->format_desc->block.bits <= bld->texel_type.width); + /* convert x,y,z coords to linear offset from start of texture, in bytes */ + lp_build_sample_offset(&bld->uint_coord_bld, + bld->format_desc, + x, y, NULL, y_stride, NULL, + &offset, &i, &j); /* get pointer to mipmap level 0 data */ data_ptr = lp_build_get_const_mipmap_level(bld, data_array, 0); - return lp_build_gather(bld->builder, - bld->texel_type.length, - bld->format_desc->block.bits, - bld->texel_type.width, - data_ptr, offset); + if (util_format_is_rgba8_variant(bld->format_desc)) { + /* Just fetch the data directly without swizzling */ + assert(bld->format_desc->block.width == 1); + assert(bld->format_desc->block.height == 1); + assert(bld->format_desc->block.bits <= bld->texel_type.width); + + res = lp_build_gather(bld->builder, + bld->texel_type.length, + bld->format_desc->block.bits, + bld->texel_type.width, + data_ptr, offset); + } + else { + struct lp_type type; + + assert(bld->texel_type.width == 32); + + memset(&type, 0, sizeof type); + type.width = 8; + type.length = bld->texel_type.length*4; + type.norm = TRUE; + + res = lp_build_fetch_rgba_aos(bld->builder, bld->format_desc, type, + data_ptr, offset, i, j); + } + + return res; } @@ -1910,9 +1910,16 @@ lp_build_sample_2d_linear_aos(struct lp_build_sample_context *bld, bld->texel_type, packed, unswizzled); - lp_build_format_swizzle_soa(bld->format_desc, - &bld->texel_bld, - unswizzled, texel_out); + if (util_format_is_rgba8_variant(bld->format_desc)) { + lp_build_format_swizzle_soa(bld->format_desc, + &bld->texel_bld, + unswizzled, texel_out); + } else { + texel_out[0] = unswizzled[0]; + texel_out[1] = unswizzled[1]; + texel_out[2] = unswizzled[2]; + texel_out[3] = unswizzled[3]; + } apply_sampler_swizzle(bld, texel_out); } @@ -2048,7 +2055,8 @@ lp_build_sample_soa(LLVMBuilderRef builder, /* For debug: no-op texture sampling */ lp_build_sample_nop(&bld, texel_out); } - else if (util_format_is_rgba8_variant(bld.format_desc) && + else if (util_format_fits_8unorm(bld.format_desc) && + bld.format_desc->nr_channels > 1 && static_state->target == PIPE_TEXTURE_2D && static_state->min_img_filter == PIPE_TEX_FILTER_LINEAR && static_state->mag_img_filter == PIPE_TEX_FILTER_LINEAR &&