freedreno/a2xx: formats update

For render formats, update fd2_pipe2color to only work with HW supported
render formats, and remove the format whitelist is_format_supported. This
patch enables float render formats (which work).

For vertex/texture formats, use a generic function which translates using
the bitsize of the channels. Since we fake support for some vertex formats,
check for these in is_format_supported to avoid enabling them as sampler
formats.

Signed-off-by: Jonathan Marek <[email protected]>
Reviewed-by: Rob Clark <[email protected]>

4 files changed, 103 insertions, 247 deletions

diff --git a/src/gallium/drivers/freedreno/a2xx/fd2_gmem.c b/src/gallium/drivers/freedreno/a2xx/fd2_gmem.c
index 82dc535a3fc..6a78db5377e 100644
--- a/src/gallium/drivers/freedreno/a2xx/fd2_gmem.c
+++ b/src/gallium/drivers/freedreno/a2xx/fd2_gmem.c
@@ -52,7 +52,6 @@ static uint32_t fmt2swap(enum pipe_format format)
 	case PIPE_FORMAT_B5G5R5X1_UNORM:
 	case PIPE_FORMAT_B4G4R4A4_UNORM:
 	case PIPE_FORMAT_B4G4R4X4_UNORM:
-	/* TODO probably some more.. */
 		return 1;
 	default:
 		return 0;
diff --git a/src/gallium/drivers/freedreno/a2xx/fd2_screen.c b/src/gallium/drivers/freedreno/a2xx/fd2_screen.c
index cd57af11ac5..42c0c024e61 100644
--- a/src/gallium/drivers/freedreno/a2xx/fd2_screen.c
+++ b/src/gallium/drivers/freedreno/a2xx/fd2_screen.c
@@ -53,27 +53,20 @@ fd2_screen_is_format_supported(struct pipe_screen *pscreen,
 	if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
 		return false;
 
-	/* TODO figure out how to render to other formats.. */
 	if ((usage & PIPE_BIND_RENDER_TARGET) &&
-			((format != PIPE_FORMAT_B5G6R5_UNORM) &&
-			 (format != PIPE_FORMAT_B5G5R5A1_UNORM) &&
-			 (format != PIPE_FORMAT_B5G5R5X1_UNORM) &&
-			 (format != PIPE_FORMAT_B4G4R4A4_UNORM) &&
-			 (format != PIPE_FORMAT_B4G4R4X4_UNORM) &&
-			 (format != PIPE_FORMAT_B8G8R8A8_UNORM) &&
-			 (format != PIPE_FORMAT_B8G8R8X8_UNORM) &&
-			 (format != PIPE_FORMAT_R8G8B8A8_UNORM) &&
-			 (format != PIPE_FORMAT_R8G8B8X8_UNORM))) {
-		DBG("not supported render target: format=%s, target=%d, sample_count=%d, usage=%x",
-				util_format_name(format), target, sample_count, usage);
-		return false;
+	    fd2_pipe2color(format) != (enum a2xx_colorformatx)~0) {
+		retval |= PIPE_BIND_RENDER_TARGET;
 	}
 
-	if ((usage & (PIPE_BIND_SAMPLER_VIEW |
-				PIPE_BIND_VERTEX_BUFFER)) &&
-			(fd2_pipe2surface(format) != (enum a2xx_sq_surfaceformat)~0)) {
-		retval |= usage & (PIPE_BIND_SAMPLER_VIEW |
-				PIPE_BIND_VERTEX_BUFFER);
+	if ((usage & (PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_VERTEX_BUFFER)) &&
+	    !util_format_is_srgb(format)) {
+		enum a2xx_sq_surfaceformat fmt = fd2_pipe2surface(format);
+		unsigned block_size = util_format_get_blocksize(format);
+		if (fmt != ~0)
+			retval |= usage & PIPE_BIND_VERTEX_BUFFER;
+		if (fmt != ~0 && block_size != 3 && block_size != 6 &&
+		    (block_size != 12 || format == PIPE_FORMAT_R32G32B32_FLOAT))
+			retval |= usage & PIPE_BIND_SAMPLER_VIEW;
 	}
 
 	if ((usage & (PIPE_BIND_RENDER_TARGET |
diff --git a/src/gallium/drivers/freedreno/a2xx/fd2_texture.c b/src/gallium/drivers/freedreno/a2xx/fd2_texture.c
index 3aca767f285..7aa26ca5673 100644
--- a/src/gallium/drivers/freedreno/a2xx/fd2_texture.c
+++ b/src/gallium/drivers/freedreno/a2xx/fd2_texture.c
@@ -180,7 +180,16 @@ fd2_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
 	so->base.reference.count = 1;
 	so->base.context = pctx;
 
+	enum sq_tex_sign sign = SQ_TEX_SIGN_UNSIGNED;
+	if (util_format_is_snorm(cso->format))
+		sign = SQ_TEX_SIGN_SIGNED;
+	/* note: SQ_TEX_SIGN_GAMMA same as SQ_TEX_SIGN_UNSIGNED (a200) */
+
 	so->tex0 =
+		A2XX_SQ_TEX_0_SIGN_X(sign) |
+		A2XX_SQ_TEX_0_SIGN_Y(sign) |
+		A2XX_SQ_TEX_0_SIGN_Z(sign) |
+		A2XX_SQ_TEX_0_SIGN_W(sign) |
 		A2XX_SQ_TEX_0_PITCH(rsc->slices[0].pitch) |
 		COND(rsc->tile_mode, A2XX_SQ_TEX_0_TILED);
 	so->tex1 =
diff --git a/src/gallium/drivers/freedreno/a2xx/fd2_util.c b/src/gallium/drivers/freedreno/a2xx/fd2_util.c
index 5957a6a4df5..af570701e54 100644
--- a/src/gallium/drivers/freedreno/a2xx/fd2_util.c
+++ b/src/gallium/drivers/freedreno/a2xx/fd2_util.c
@@ -32,181 +32,81 @@
 enum a2xx_sq_surfaceformat
 fd2_pipe2surface(enum pipe_format format)
 {
-	switch (format) {
-	/* 8-bit buffers. */
-	case PIPE_FORMAT_A8_UNORM:
-	case PIPE_FORMAT_A8_SNORM:
-	case PIPE_FORMAT_A8_UINT:
-	case PIPE_FORMAT_A8_SINT:
-	case PIPE_FORMAT_I8_UNORM:
-	case PIPE_FORMAT_I8_SNORM:
-	case PIPE_FORMAT_I8_UINT:
-	case PIPE_FORMAT_I8_SINT:
-	case PIPE_FORMAT_L8_UNORM:
-	case PIPE_FORMAT_L8_SNORM:
-	case PIPE_FORMAT_L8_UINT:
-	case PIPE_FORMAT_L8_SINT:
-	case PIPE_FORMAT_L8_SRGB:
-	case PIPE_FORMAT_R8_UNORM:
-	case PIPE_FORMAT_R8_SNORM:
-	case PIPE_FORMAT_R8_UINT:
-	case PIPE_FORMAT_R8_SINT:
-		return FMT_8;
-
-	/* 16-bit buffers. */
-	case PIPE_FORMAT_B5G6R5_UNORM:
-		return FMT_5_6_5;
-	case PIPE_FORMAT_B5G5R5A1_UNORM:
-	case PIPE_FORMAT_B5G5R5X1_UNORM:
-		return FMT_1_5_5_5;
-	case PIPE_FORMAT_B4G4R4A4_UNORM:
-	case PIPE_FORMAT_B4G4R4X4_UNORM:
-		return FMT_4_4_4_4;
-	case PIPE_FORMAT_Z16_UNORM:
-		return FMT_16;
-	case PIPE_FORMAT_L8A8_UNORM:
-	case PIPE_FORMAT_L8A8_SNORM:
-	case PIPE_FORMAT_L8A8_UINT:
-	case PIPE_FORMAT_L8A8_SINT:
-	case PIPE_FORMAT_L8A8_SRGB:
-	case PIPE_FORMAT_R8G8_UNORM:
-	case PIPE_FORMAT_R8G8_SNORM:
-	case PIPE_FORMAT_R8G8_UINT:
-	case PIPE_FORMAT_R8G8_SINT:
-		return FMT_8_8;
-	case PIPE_FORMAT_R16_UNORM:
-	case PIPE_FORMAT_R16_SNORM:
-	case PIPE_FORMAT_R16_UINT:
-	case PIPE_FORMAT_R16_SINT:
-	case PIPE_FORMAT_A16_UNORM:
-	case PIPE_FORMAT_A16_SNORM:
-	case PIPE_FORMAT_A16_UINT:
-	case PIPE_FORMAT_A16_SINT:
-	case PIPE_FORMAT_L16_UNORM:
-	case PIPE_FORMAT_L16_SNORM:
-	case PIPE_FORMAT_L16_UINT:
-	case PIPE_FORMAT_L16_SINT:
-	case PIPE_FORMAT_I16_UNORM:
-	case PIPE_FORMAT_I16_SNORM:
-	case PIPE_FORMAT_I16_UINT:
-	case PIPE_FORMAT_I16_SINT:
-		return FMT_16;
-	case PIPE_FORMAT_R16_FLOAT:
-	case PIPE_FORMAT_A16_FLOAT:
-	case PIPE_FORMAT_L16_FLOAT:
-	case PIPE_FORMAT_I16_FLOAT:
-		return FMT_16_FLOAT;
-
-	/* 32-bit buffers. */
-	case PIPE_FORMAT_A8B8G8R8_SRGB:
-	case PIPE_FORMAT_A8B8G8R8_UNORM:
-	case PIPE_FORMAT_A8R8G8B8_UNORM:
-	case PIPE_FORMAT_B8G8R8A8_SRGB:
-	case PIPE_FORMAT_B8G8R8A8_UNORM:
-	case PIPE_FORMAT_B8G8R8X8_UNORM:
-	case PIPE_FORMAT_R8G8B8A8_SNORM:
-	case PIPE_FORMAT_R8G8B8A8_UNORM:
-	case PIPE_FORMAT_R8G8B8X8_UNORM:
-	case PIPE_FORMAT_R8SG8SB8UX8U_NORM:
-	case PIPE_FORMAT_X8B8G8R8_UNORM:
-	case PIPE_FORMAT_X8R8G8B8_UNORM:
-	case PIPE_FORMAT_R8G8B8_UNORM:
-	case PIPE_FORMAT_R8G8B8A8_SINT:
-	case PIPE_FORMAT_R8G8B8A8_UINT:
-		return FMT_8_8_8_8;
-	case PIPE_FORMAT_R10G10B10A2_UNORM:
-	case PIPE_FORMAT_R10G10B10X2_SNORM:
-	case PIPE_FORMAT_B10G10R10A2_UNORM:
-	case PIPE_FORMAT_B10G10R10A2_UINT:
-	case PIPE_FORMAT_R10SG10SB10SA2U_NORM:
-		return FMT_2_10_10_10;
-	case PIPE_FORMAT_Z24X8_UNORM:
-	case PIPE_FORMAT_Z24_UNORM_S8_UINT:
-		return FMT_24_8;
-	case PIPE_FORMAT_R32_UINT:
-	case PIPE_FORMAT_R32_SINT:
-	case PIPE_FORMAT_A32_UINT:
-	case PIPE_FORMAT_A32_SINT:
-	case PIPE_FORMAT_L32_UINT:
-	case PIPE_FORMAT_L32_SINT:
-	case PIPE_FORMAT_I32_UINT:
-	case PIPE_FORMAT_I32_SINT:
-		return FMT_32;
-	case PIPE_FORMAT_R32_FLOAT:
-	case PIPE_FORMAT_A32_FLOAT:
-	case PIPE_FORMAT_L32_FLOAT:
-	case PIPE_FORMAT_I32_FLOAT:
-	case PIPE_FORMAT_Z32_FLOAT:
-		return FMT_32_FLOAT;
-	case PIPE_FORMAT_R16G16_FLOAT:
-	case PIPE_FORMAT_L16A16_FLOAT:
-		return FMT_16_16_FLOAT;
-	case PIPE_FORMAT_R16G16_UNORM:
-	case PIPE_FORMAT_R16G16_SNORM:
-	case PIPE_FORMAT_R16G16_UINT:
-	case PIPE_FORMAT_R16G16_SINT:
-	case PIPE_FORMAT_L16A16_UNORM:
-	case PIPE_FORMAT_L16A16_SNORM:
-	case PIPE_FORMAT_L16A16_UINT:
-	case PIPE_FORMAT_L16A16_SINT:
-		return FMT_16_16;
-
-	/* 64-bit buffers. */
-	case PIPE_FORMAT_R16G16B16A16_UINT:
-	case PIPE_FORMAT_R16G16B16A16_SINT:
-	case PIPE_FORMAT_R16G16B16A16_UNORM:
-	case PIPE_FORMAT_R16G16B16A16_SNORM:
-		return FMT_16_16_16_16;
-	case PIPE_FORMAT_R16G16B16A16_FLOAT:
-		return FMT_16_16_16_16_FLOAT;
-	case PIPE_FORMAT_R32G32_FLOAT:
-	case PIPE_FORMAT_L32A32_FLOAT:
-		return FMT_32_32_FLOAT;
-	case PIPE_FORMAT_R32G32_SINT:
-	case PIPE_FORMAT_R32G32_UINT:
-	case PIPE_FORMAT_L32A32_UINT:
-	case PIPE_FORMAT_L32A32_SINT:
-		return FMT_32_32;
-
-	/* 96-bit buffers. */
-	case PIPE_FORMAT_R32G32B32_FLOAT:
-		return FMT_32_32_32_FLOAT;
-
-	/* 128-bit buffers. */
-	case PIPE_FORMAT_R32G32B32A32_SNORM:
-	case PIPE_FORMAT_R32G32B32A32_UNORM:
-	case PIPE_FORMAT_R32G32B32A32_SINT:
-	case PIPE_FORMAT_R32G32B32A32_UINT:
-		return FMT_32_32_32_32;
-	case PIPE_FORMAT_R32G32B32A32_FLOAT:
-		return FMT_32_32_32_32_FLOAT;
-
-	/* Compressed textures. */
-	case PIPE_FORMAT_ETC1_RGB8:
-		return FMT_ETC1_RGB;
-	case PIPE_FORMAT_DXT1_RGB:
-	case PIPE_FORMAT_DXT1_RGBA:
-		return FMT_DXT1;
-	case PIPE_FORMAT_DXT3_RGBA:
-		return FMT_DXT2_3;
-	case PIPE_FORMAT_DXT5_RGBA:
-		return FMT_DXT4_5;
-	case PIPE_FORMAT_ATC_RGB:
-		return FMT_ATI_TC_555_565_RGB;
-	case PIPE_FORMAT_ATC_RGBA_EXPLICIT:
-		return FMT_ATI_TC_555_565_RGBA;
-	case PIPE_FORMAT_ATC_RGBA_INTERPOLATED:
-		return FMT_ATI_TC_555_565_RGBA_INTERP;
-
-	/* YUV buffers. */
-	case PIPE_FORMAT_UYVY:
-		return FMT_Cr_Y1_Cb_Y0;
-	case PIPE_FORMAT_YUYV:
-		return FMT_Y1_Cr_Y0_Cb;
+	const struct util_format_description *desc = util_format_description(format);
+
+	if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN) {
+		switch (format) {
+		/* Compressed textures. */
+		case PIPE_FORMAT_ETC1_RGB8:
+			return FMT_ETC1_RGB;
+		case PIPE_FORMAT_DXT1_RGB:
+		case PIPE_FORMAT_DXT1_RGBA:
+			return FMT_DXT1;
+		case PIPE_FORMAT_DXT3_RGBA:
+			return FMT_DXT2_3;
+		case PIPE_FORMAT_DXT5_RGBA:
+			return FMT_DXT4_5;
+		case PIPE_FORMAT_ATC_RGB:
+			return FMT_ATI_TC_555_565_RGB;
+		case PIPE_FORMAT_ATC_RGBA_EXPLICIT:
+			return FMT_ATI_TC_555_565_RGBA;
+		case PIPE_FORMAT_ATC_RGBA_INTERPOLATED:
+			return FMT_ATI_TC_555_565_RGBA_INTERP;
+		/* YUV buffers. */
+		case PIPE_FORMAT_UYVY:
+			return FMT_Y1_Cr_Y0_Cb;
+		case PIPE_FORMAT_YUYV:
+			return FMT_Cr_Y1_Cb_Y0;
+		default:
+			return ~0;
+		}
+	}
 
-	default:
-		return ~0;
+	uint32_t channel_size = 0;
+	for (unsigned i = 0; i < 4; i++)
+		channel_size |= desc->channel[i].size << i*8;
+
+	/* Note: the 3 channel 24bpp/48bpp/96bpp formats are only for vertex fetch
+	 * we can use the 4 channel format and ignore the 4th component just isn't used
+	 * XXX: is it possible for the extra loaded component to cause a MMU fault?
+	 */
+
+#define CASE(r, g, b, a) case (r | g << 8 | b << 16 | a << 24)
+	if (desc->channel[0].type == UTIL_FORMAT_TYPE_FLOAT) {
+		switch (channel_size) {
+		CASE(16,  0,  0,  0): return FMT_16_FLOAT;
+		CASE(16, 16,  0,  0): return FMT_16_16_FLOAT;
+		CASE(16, 16, 16,  0): return FMT_16_16_16_16_FLOAT; /* Note: only for vertex */
+		CASE(16, 16, 16, 16): return FMT_16_16_16_16_FLOAT;
+		CASE(32,  0,  0,  0): return FMT_32_FLOAT;
+		CASE(32, 32,  0,  0): return FMT_32_32_FLOAT;
+		CASE(32, 32, 32,  0): return FMT_32_32_32_FLOAT;
+		CASE(32, 32, 32, 32): return FMT_32_32_32_32_FLOAT;
+		}
+	} else if (desc->is_unorm || desc->is_snorm) {
+		switch (channel_size) {
+		CASE( 8,  0,  0,  0): return FMT_8;
+		CASE( 8,  8,  0,  0): return FMT_8_8;
+		CASE( 8,  8,  8,  0): return FMT_8_8_8_8; /* Note: only for vertex */
+		CASE( 8,  8,  8,  8): return FMT_8_8_8_8;
+		CASE(16,  0,  0,  0): return FMT_16;
+		CASE(16, 16,  0,  0): return FMT_16_16;
+		CASE(16, 16, 16,  0): return FMT_16_16_16_16; /* Note: only for vertex */
+		CASE(16, 16, 16, 16): return FMT_16_16_16_16;
+		CASE(32,  0,  0,  0): return FMT_32;
+		CASE(32, 32,  0,  0): return FMT_32_32;
+		CASE(32, 32, 32,  0): return FMT_32_32_32_32; /* Note: only for vertex */
+		CASE(32, 32, 32, 32): return FMT_32_32_32_32;
+		CASE( 4,  4,  4,  4): return FMT_4_4_4_4;
+		CASE( 5,  5,  5,  1): return FMT_1_5_5_5;
+		CASE( 5,  6,  5,  0): return FMT_5_6_5;
+		CASE(10, 10, 10,  2): return FMT_2_10_10_10;
+		CASE( 8, 24,  0,  0): return FMT_24_8;
+		}
 	}
+#undef CASE
+
+	return ~0;
 }
 
 enum a2xx_colorformatx
@@ -214,23 +114,7 @@ fd2_pipe2color(enum pipe_format format)
 {
 	switch (format) {
 	/* 8-bit buffers. */
-	case PIPE_FORMAT_A8_UNORM:
-	case PIPE_FORMAT_A8_SNORM:
-	case PIPE_FORMAT_A8_UINT:
-	case PIPE_FORMAT_A8_SINT:
-	case PIPE_FORMAT_I8_UNORM:
-	case PIPE_FORMAT_I8_SNORM:
-	case PIPE_FORMAT_I8_UINT:
-	case PIPE_FORMAT_I8_SINT:
-	case PIPE_FORMAT_L8_UNORM:
-	case PIPE_FORMAT_L8_SNORM:
-	case PIPE_FORMAT_L8_UINT:
-	case PIPE_FORMAT_L8_SINT:
-	case PIPE_FORMAT_L8_SRGB:
 	case PIPE_FORMAT_R8_UNORM:
-	case PIPE_FORMAT_R8_SNORM:
-	case PIPE_FORMAT_R8_UINT:
-	case PIPE_FORMAT_R8_SINT:
 		return COLORX_8;
 
 	/* 16-bit buffers. */
@@ -242,60 +126,31 @@ fd2_pipe2color(enum pipe_format format)
 	case PIPE_FORMAT_B4G4R4A4_UNORM:
 	case PIPE_FORMAT_B4G4R4X4_UNORM:
 		return COLORX_4_4_4_4;
-	case PIPE_FORMAT_L8A8_UNORM:
-	case PIPE_FORMAT_L8A8_SNORM:
-	case PIPE_FORMAT_L8A8_UINT:
-	case PIPE_FORMAT_L8A8_SINT:
-	case PIPE_FORMAT_L8A8_SRGB:
 	case PIPE_FORMAT_R8G8_UNORM:
-	case PIPE_FORMAT_R8G8_SNORM:
-	case PIPE_FORMAT_R8G8_UINT:
-	case PIPE_FORMAT_R8G8_SINT:
-	case PIPE_FORMAT_Z16_UNORM:
 		return COLORX_8_8;
-	case PIPE_FORMAT_R16_FLOAT:
-	case PIPE_FORMAT_A16_FLOAT:
-	case PIPE_FORMAT_L16_FLOAT:
-	case PIPE_FORMAT_I16_FLOAT:
-		return COLORX_16_FLOAT;
 
 	/* 32-bit buffers. */
-	case PIPE_FORMAT_A8B8G8R8_SRGB:
-	case PIPE_FORMAT_A8B8G8R8_UNORM:
-	case PIPE_FORMAT_A8R8G8B8_UNORM:
-	case PIPE_FORMAT_B8G8R8A8_SRGB:
 	case PIPE_FORMAT_B8G8R8A8_UNORM:
 	case PIPE_FORMAT_B8G8R8X8_UNORM:
-	case PIPE_FORMAT_R8G8B8A8_SNORM:
 	case PIPE_FORMAT_R8G8B8A8_UNORM:
 	case PIPE_FORMAT_R8G8B8X8_UNORM:
-	case PIPE_FORMAT_R8SG8SB8UX8U_NORM:
-	case PIPE_FORMAT_X8B8G8R8_UNORM:
-	case PIPE_FORMAT_X8R8G8B8_UNORM:
-	case PIPE_FORMAT_R8G8B8_UNORM:
-	case PIPE_FORMAT_R8G8B8A8_SINT:
-	case PIPE_FORMAT_R8G8B8A8_UINT:
-	case PIPE_FORMAT_Z24_UNORM_S8_UINT:
-	case PIPE_FORMAT_Z24X8_UNORM:
 		return COLORX_8_8_8_8;
-	case PIPE_FORMAT_R32_FLOAT:
-	case PIPE_FORMAT_A32_FLOAT:
-	case PIPE_FORMAT_L32_FLOAT:
-	case PIPE_FORMAT_I32_FLOAT:
-	case PIPE_FORMAT_Z32_FLOAT:
-		return COLORX_32_FLOAT;
+	/* Note: snorm untested */
+	case PIPE_FORMAT_R8G8B8A8_SNORM:
+	case PIPE_FORMAT_R8G8B8X8_SNORM:
+		return COLORX_S8_8_8_8;
+
+	/* float buffers */
+	case PIPE_FORMAT_R16_FLOAT:
+		return COLORX_16_FLOAT;
 	case PIPE_FORMAT_R16G16_FLOAT:
-	case PIPE_FORMAT_L16A16_FLOAT:
 		return COLORX_16_16_FLOAT;
-
-	/* 64-bit buffers. */
 	case PIPE_FORMAT_R16G16B16A16_FLOAT:
 		return COLORX_16_16_16_16_FLOAT;
+	case PIPE_FORMAT_R32_FLOAT:
+		return COLORX_32_FLOAT;
 	case PIPE_FORMAT_R32G32_FLOAT:
-	case PIPE_FORMAT_L32A32_FLOAT:
 		return COLORX_32_32_FLOAT;
-
-	/* 128-bit buffers. */
 	case PIPE_FORMAT_R32G32B32A32_FLOAT:
 		return COLORX_32_32_32_32_FLOAT;